Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
THE SEVEN DEADLY SINS
OF PSYCHOLOGY
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
THE SEVEN DEADLY SINS
OF PSYCHOLOGY
A Manifesto for Reforming the Culture
of Scientific Practice
C H R I S C H A M B E R S
Princeton University Press
Princeton & Oxford
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
CONTENTS
Preface ix
Chapter 1. The Sin of Bias 1
A Brief History of the “Yes Man” 4
Neophilia: When the Positive and New Trumps the Negative but True 8
Replicating Concepts Instead of Experiments 13
Reinventing History 16
The Battle against Bias 20
Chapter 2. The Sin of Hidden Flexibility 22
p-Hacking 24
Peculiar Patterns of p 29
Ghost Hunting 34
Unconscious Analytic “Tuning” 35
Biased Debugging 39
Are Research Psychologists Just Poorly Paid Lawyers? 40
Solutions to Hidden Flexibility 41
Chapter 3. The Sin of Unreliability 46
Sources of Unreliability in Psychology 48
Reason 1: Disregard for Direct Replication 48
Reason 2: Lack of Power 55
Reason 3: Failure to Disclose Methods 61
Reason 4: Statistical Fallacies 63
Reason 5: Failure to Retract 65
Solutions to Unreliability 67
Chapter 4. The Sin of Data Hoarding 75
The Untold Benefits of Data Sharing 77
Failure to Share 78
Secret Sharing 80
How Failing to Share Hides Misconduct 81
Making Data Sharing the Norm 84
Grassroots, Carrots, and Sticks 88
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
viii | Contents
Unlocking the Black Box 91
Preventing Bad Habits 94
Chapter 5. The Sin of Corruptibility 96
The Anatomy of Fraud 99
The Thin Gray Line 105
When Junior Scientists Go Astray 112
Kate’s Story 117
The Dirty Dozen: How to Get Away with Fraud 122
Chapter 6. The Sin of Internment 126
The Basics of Open Access Publishing 128
Why Do Psychologists Support Barrier-Based Publishing? 129
Hybrid OA as Both a Solution and a Problem 132
Calling in the Guerrillas 136
Counterarguments 138
An Open Road 147
Chapter 7. The Sin of Bean Counting 149
Roads to Nowhere 151
Impact Factors and Modern-Day Astrology 151
Wagging the Dog 160
The Murky Mess of Academic Authorship 163
Roads to Somewhere 168
Chapter 8. Redemption 171
Solving the Sins of Bias and Hidden Flexibility 174
Registered Reports: A Vaccine against Bias 174
Preregistration without Peer Review 196
Solving the Sin of Unreliability 198
Solving the Sin of Data Hoarding 202
Solving the Sin of Corruptibility 205
Solving the Sin of Internment 208
Solving the Sin of Bean Counting 210
Concrete Steps for Reform 213
Coda 215
Notes 219
Index 263
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
PREFACE
This book is borne out of what I can only describe as a deep personal frustration
with the working culture of psychological science. I have always
thought of our professional culture as a castle—a sanctuary of endeavor built
long ago by our forebears. Like any home it needs constant care and attention,
but instead of repairing it as we go we have allowed it to fall into a state
of disrepair. The windows are dirty and opaque. The roof is leaking and
won’t keep out the rain for much longer. Monsters live in the dungeon.
Despite its many flaws, the castle has served me well. It sheltered me
during my formative years as a junior researcher and advanced me to a
position where I can now talk openly about the need for renovation. And
I stress renovation because I am not suggesting we demolish our stronghold
and start over. The foundations of psychology are solid, and the field has
a proud legacy of discovery. Our founders—Helmholtz, Wundt, James—
built it to last.
After spending fifteen years in psychology and its cousin, cognitive neuroscience,
I have nevertheless reached an unsettling conclusion. If we continue
as we are then psychology will diminish as a reputable science and
could very well disappear. If we ignore the warning signs now, then in a
hundred years or less, psychology may be regarded as one in a long line of
quaint scholarly indulgences, much as we now regard alchemy or phrenology.
Our descendants will smile tolerantly at this pocket of academic antiquity,
nod sagely to one another about the protoscience that was psychology,
and conclude that we were subject to the “limitations of the time.” Of
course, few sciences are likely to withstand the judgment of history, but it
is by our research practices rather than our discoveries that psychology will
be judged most harshly. And that judgment will be this: like so many other
“soft” sciences, we found ourselves trapped within a culture where the appearance
of science was seen as an appropriate replacement for the practice
of science.
In this book I’m going to show how this distortion penetrates many aspects
of our professional lives as scientists. The journey will be grim in
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
x | Preface
places. Using the seven deadly sins as a metaphor, I will explain how unchecked
bias fools us into seeing what we want to see; how we have turned
our backs on fundamental principles of the scientific method; how we treat
the data we acquire as personal property rather than a public resource; how
we permit academic fraud to cause untold damage to the most vulnerable
members of our community; how we waste public resources on outdated
forms of publishing; and how, in assessing the value of science and scientists,
we have surrendered expert judgment to superficial bean counting. I
will hope to convince you that in the quest for genuine understanding, we
must be unflinching in recognizing these failings and relentless in fixing
them.
Within each chapter, and in a separate final chapter, I will recommend
various reforms that highlight two core aspects of science: transparency
and reproducibility. To survive in the twenty-first century and beyond we
must transform our secretive and fragile culture into a truly open and rigorous
science—one that celebrates openness as much as it appreciates innovation,
that prizes robustness as much as novelty. We must recognize that the
old way of doing things is no longer fit for purpose and find a new path.
At its broadest level this book is intended for anyone who is interested
in the practice and culture of science. Even those with no specific interest
in psychology have reasons to care about the problems we face. Malpractice
in any field wastes precious public funding by pursuing lines of enquiry that
may turn out to be misleading or bogus. For example, by suppressing certain
types of results from the published record, we risk introducing ineffective
clinical treatments for mental health conditions such as depression and
schizophrenia. In the UK, where the socioeconomic impact of research is
measured as part of a regular national exercise called the Research Excellence
Framework (REF), psychology has also been shown to influence a
wide range of real-world applications. The 2014 REF reported over 450
“impact case studies” where psychological research has shaped public policy
or practice, including (to name just a few) the design and uptake of electric
cars, strategies for minimizing exam anxiety, the development of improved
police interviewing techniques that account for the limits of human memory,
setting of urban speed limits based on discoveries in vision science,
human factors that are important for effective space exploration, government
strategies for dealing with climate change that take into account public
perception of risk, and plain packaging of tobacco products.1 From its
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Preface | xi
most basic roots to its most applied branches, psychology is a rich part of
public life and a key to understanding many global problems; therefore the
deadly sins discussed here are a problem for society as a whole.
Some of the content, particularly sections on statistical methods, will be
most relevant to the recently embarked researcher—the undergraduate student,
PhD student, or early-career scientist—but there are also important
messages throughout the book for more senior academics who manage their
own laboratories or institutions, and many issues are also relevant to journalists
and science writers. To aid the accessibility of source material for
different audiences I have referred as much as possible to open access literature.
For articles that are not open access, a Google Scholar search of the
article title will often reveal a freely available electronic copy. I have also
drawn on more contemporary forms of communication, including freely
available blog entries and social media.
I owe a great debt to many friends, academic colleagues, journal editors,
science writers, journalists, press officers, and policy experts, for years of
inspiration, critical discussions, arguments, and in some cases interviews
that fed into this work, including: Rachel Adams, Chris Allen, Micah Allen,
Adam Aron, Vaughan Bell, Sven Bestmann, Ananyo Bhattacharya, Dorothy
Bishop, Fred Boy, Todd Braver, Björn Brembs, Jon Brock, Jon Butterworth,
Kate Button, Iain Chalmers, David Colquhoun, Molly Crockett, Stephen
Curry, Helen Czerski, Zoltan Dienes, the late Jon Driver, Malte Elson, Alex
Etz, John Evans, Eva Feredoes, Matt Field, Agneta Fischer, Birte Forstmann,
Fiona Fox, Andrew Gelman, Tom Hardwicke, Chris Hartgerink, Tom Hartley,
Mark Haselgrove, Steven Hill, Alex Holcombe, Aidan Horner, Macartan
Humphreys, Hans Ijzerman, Helen Jamieson, Alok Jha, Gabi Jiga-Boy, Ben
Johnson, Rogier Kievit, James Kilner, Daniël Lakens, Natalia Lawrence,
Keith Laws, Katie Mack, Leah Maizey, Jason Mattingley, Rob McIntosh,
Susan Michie, Candice Morey, Richard Morey, Simon Moss, Ross Mounce,
Nils Mulhert, Kevin Murphy, Suresh Muthukumaraswamy, Bas Neggers,
Neuroskeptic, Kia Nobre, Dave Nussbaum, Hans Op de Beeck, Ivan Oransky,
Damian Pattinson, Andrew Przybylski, James Randerson, Geraint Rees,
Ged Ridgway, Robert Rosenthal, Pia Rotshtein, Jeff Rouder, Elena Rusconi,
Adam Rutherford, Chris Said, Ayse Saygin, Anne Scheel, Sam Schwarzkopf,
Sophie Scott, Dan Simons, Jon Simons, Uri Simonsohn, Sanjay Srivastava,
Mark Stokes, Petroc Sumner, Mike Taylor, Jon Tennant, Eric Turner, Carien
van Reekum, Simine Vazire, Essi Viding, Solveiga Vivian-Griffiths, Matt
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
xii | Preface
Wall, Tony Weidberg, Robert West, Jelte Wicherts, Ed Wilding, Andrew
Wilson, Tal Yarkoni, Ed Yong, and Rolf Zwaan. Sincere thanks go to Sergio
Della Sala and Toby Charkin for their collaboration and fortitude in championing
Registered Reports at Cortex, Brian Nosek, David Mellor, and Sara
Bowman for providing Registered Reports with such a welcoming home at
the Center for Open Science, and to the Royal Society, particularly publisher
Phil Hurst and publishing director Stuart Taylor, for embracing Registered
Reports long before any other multidisciplinary journal. I also owe
a debt of gratitude to Marcus Munafò for joining me in promoting Registered
Reports at every turn, and to the 83 scientists who signed our Guardian
open letter calling for installation of the format within all life science
journals. Finally, I extend a special thanks to Dorothy Bishop, the late (and
much missed) Alex Danchev, Dee Danchev, Zoltan Dienes, Pete Etchells,
Hal Pashler, Frederick Verbruggen, and E. J. Wagenmakers for extensive
draft reading and discussion, to Anastasiya Tarasenko for creating the chapter
illustrations, and to my editors Sarah Caro and Eric Schwartz for their
patience and sage advice throughout this journey.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
THE SEVEN DEADLY SINS
OF PSYCHOLOGY
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
C H A P T E R 1
The Sin of Bias
The human understanding when it has once adopted
an opinion . . . draws all things else to support
and agree with it.
—Francis Bacon, 1620
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
History may look back on 2011 as the year that changed psychology
forever. It all began when the Journal of Personality and Social Psychology
published an article called “Feeling the Future: Experimental Evidence
for Anomalous Retroactive Influences on Cognition and Affect.”1 The
paper, written by Daryl Bem of Cornell University, reported a series of experiments
on psi or “precognition,” a supernatural phenomenon that supposedly
enables people to see events in the future. Bem, himself a reputable
psychologist, took an innovative approach to studying psi. Instead of
using discredited parapsychological methods such as card tasks or dice
tests, he selected a series of gold-standard psychological techniques and
modified them in clever ways.
One such method was a reversed priming task. In a typical priming task,
people decide whether a picture shown on a computer screen is linked to a
positive or negative emotion. So, for example, the participant might decide
whether a picture of kittens is pleasant or unpleasant. If a word that “primes”
the same emotion is presented immediately before the picture (such as the
word “joy” followed by the picture of kittens), then people find it easier to
judge the emotion of the picture, and they respond faster. But if the prime
and target trigger opposite emotions then the task becomes more difficult
because the emotions conflict (e.g., the word “murder” followed by kittens).
To test for the existence of precognition, Bem reversed the order of this
experiment and found that primes delivered after people had responded
seemed to influence their reaction times. He also reported similar “retroactive”
effects on memory. In one of his experiments, people were overall
better at recalling specific words from a list that were also included in a
practice task, with the catch that the so-called practice was undertaken
after the recall task rather than before. On this basis, Bem argued that the
participants were able to benefit in the past from practice they had completed
in the future.
As you might expect, Bem’s results generated a flood of confusion and
controversy. How could an event in the future possibly influence someone’s
reaction time or memory in the past? If precognition truly did exist, in even
a tiny minority of the population, how is it that casinos or stock markets
turn profits? And how could such a bizarre conclusion find a home in a
reputable scientific journal?
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bias | 3
Scrutiny at first turned to Bem’s experimental procedures. Perhaps there
was some flaw in the methods that could explain his results, such as failing
to randomize the order of events, or some other subtle experimental error.
But these aspects of the experiment seemed to pass muster, leaving the
research community facing a dilemma. If true, precognition would be the
most sensational discovery in modern science. We would have to accept the
existence of time travel and reshape our entire understanding of cause and
effect. But if false, Bem’s results would instead point to deep flaws in standard
research practices—after all, if accepted practices could generate such
nonsensical findings, how can any published findings in psychology be
trusted? And so psychologists faced an unenviable choice between, on the
one hand, accepting an impossible scientific conclusion and, on the other
hand, swallowing an unpalatable professional reality.
The scientific community was instinctively skeptical of Bem’s conclusions.
Responding to a preprint of the article that appeared in late 2010,
the psychologist Joachim Krueger said: “My personal view is that this is
ridiculous and can’t be true.”2 After all, extraordinary claims require
extraordinary evidence, and despite being published in a prestigious journal,
the statistical strength of Bem’s evidence was considered far from
extraordinary.
Bem himself realized that his results defied explanation and stressed
the need for independent researchers to replicate his findings. Yet doing
so proved more challenging than you might imagine. One replication attempt
by Chris French and Stuart Ritchie showed no evidence whatsoever
of precognition but was rejected by the same journal that published Bem’s
paper. In this case the journal didn’t even bother to peer review French
and Ritchie’s paper before rejecting it, explaining that it “does not publish
replication studies, whether successful or unsuccessful.”3 This decision
may sound bizarre, but, as we will see, contempt for replication is common
in psychology compared with more established sciences. The most prominent
psychology journals selectively publish findings that they consider to
be original, novel, neat, and above all positive. This publication bias, also
known as the “file-drawer effect,” means that studies that fail to show statistically
significant effects, or that reproduce the work of others, have such
low priority that they are effectively censored from the scientific record.
They either end up in the file drawer or are never conducted in the first
place.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
4 | Chapter 1
Publication bias is one form of what is arguably the most powerful fallacy
in human reasoning: confirmation bias. When we fall prey to confirmation
bias, we seek out and favor evidence that agrees with our existing beliefs,
while at the same time ignoring or devaluing evidence that doesn’t. Confirmation
bias corrupts psychological science in several ways. In its simplest
form, it favors the publication of positive results—that is, hypothesis tests
that reveal statistically significant differences or associations between conditions
(e.g., A is greater than B; A is related to B, vs. A is the same as B; A
is unrelated to B). More insidiously, it contrives a measure of scientific reproducibility
in which it is possible to replicate but never falsify previous
findings, and it encourages altering the hypotheses of experiments after the
fact to “predict” unexpected outcomes. One of the most troubling aspects
of psychology is that the academic community has refused to unanimously
condemn such behavior. On the contrary, many psychologists acquiesce to
these practices and even embrace them as survival skills in a culture where
researchers must publish or perish.
Within months of appearing in a top academic journal, Bem’s claims
about precognition were having a powerful, albeit unintended, effect on the
psychological community. Established methods and accepted publishing
practices fell under renewed scrutiny for producing results that appear convincing
but are almost certainly false. As psychologist Eric-Jan Wagenmakers
and colleagues noted in a statistical demolition of Bem’s paper: “Our
assessment suggests that something is deeply wrong with the way experimental
psychologists design their studies and report their statistical results.”4
With these words, the storm had broken.
A Brief History of the “Yes Man”
To understand the different ways that bias influences psychological science,
we need to take a step back and consider the historical origins and basic
research on confirmation bias. Philosophers and scholars have long recognized
the “yes man” of human reasoning. As early as the fifth century BC,
the historian Thucydides noted words to the effect that “[w]hen a man finds
a conclusion agreeable, he accepts it without argument, but when he finds
it disagreeable, he will bring against it all the forces of logic and reason.”
Similar sentiments were echoed by Dante, Bacon, and Tolstoy. By the mid-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bias | 5
twentieth century, the question had evolved from one of philosophy to one
of science, as psychologists devised ways to measure confirmation bias in
controlled laboratory experiments.
Since the mid-1950s, a convergence of studies has suggested that when
people are faced with a set of observations (data) and a possible explanation
(hypothesis), they favor tests of the hypothesis that seek to confirm it rather
than falsify it. Formally, what this means is that people are biased toward
estimating the probability of data if a particular hypothesis is true,
p(data|hypothesis) rather than the opposite probability of it being false,
p(data|~hypothesis). In other words, people prefer to ask questions to which
the answer is “yes,” ignoring the maxim of philosopher Georg Henrik von
Wright that “no confirming instance of a law is a verifying instance, but . . .
any disconfirming instance is a falsifying instance.”5
Psychologist Peter Wason was one of the first researchers to provide
laboratory evidence of confirmation bias. In one of several innovative experiments
conducted in the 1960s and 1970s, he gave participants a sequence
of numbers, such as 2-4-6, and asked them to figure out the rule
that produced it (in this case: three numbers in increasing order of magnitude).6
Having formed a hypothesis, participants were then allowed to write
down their own sequence, after which they were told whether their sequence
was consistent or inconsistent with the actual rule. Wason found
that participants showed a strong bias to test various hypotheses by confirming
them, even when the outcome of doing so failed to eliminate plausible
alternatives (such as three even numbers). Wason’s participants used this
strategy despite being told in advance that “your aim is not simply to find
numbers which conform to the rule, but to discover the rule itself.”
Since then, many studies have explored the basis of confirmation bias in
a range of laboratory-controlled situations. Perhaps the most famous of
these is the ingenious Selection Task, which was also developed by Wason
in 1968.7 The Selection Task works like this. Suppose I were to show you
four cards on a table, labeled D, B, 3, and 7 (see figure 1.1). I tell you that if
the card shows a letter on one side then it will have a number on the other
side, and I provide you with a more specific rule (hypothesis) that may be
true or false: “If there is a D on one side of any card, then there is a 3 on its other
side.” Finally, I ask you to tell me which cards you would need to turn over
in order to determine whether this rule is true or false. Leaving an informative
card unturned or turning over an uninformative card (i.e., one that
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
6 | Chapter 1
doesn’t test the rule) would be considered an incorrect response. Before
reading further, take a moment and ask yourself, which cards would you
choose and which would you avoid?
If you chose D and avoided B then you’re in good company. Both responses
are correct and are made by the majority of participants. Selecting
D seeks to test the rule by confirming it, whereas avoiding B is correct because
the flip side would be uninformative regardless of the outcome.
Did you choose 3? Wason found that most participants did, even though
3 should be avoided. This is because if the flip side isn’t a D, we learn nothing—the
rule states that cards with D on one side are paired a 3 on the
other, not that D is the only letter to be paired with a 3 (drawing such a
conclusion would be a logical fallacy known as “affirming the consequent”).
And even if the flip side is a D then the outcome would be consistent with
the rule but wouldn’t confirm it, for exactly the same reason.
Finally, did you choose 7 or avoid it? Interestingly, Wason found that few
participants selected 7, even though doing so is correct—in fact, it is just as
correct as selecting D. If the flip side to 7 were discovered to be a D then
the rule would be categorically disproven—a logical test of what’s known
as the “contrapositive.” And herein lies the key result: the fact that most
participants correctly select D but fail to select 7 provides evidence that
people seek to test rules or hypotheses by confirming them rather than by
falsifying them.
Wason’s findings provided the first laboratory-controlled evidence of confirmation
bias, but centuries of informal observations already pointed
strongly to its existence. In a landmark review, psychologist Raymond Nickerson
noted how confirmation bias dominated in the witchcraft trials of the
middle ages.8 Many of these proceedings were a foregone conclusion, seeking
only to obtain evidence that confirmed the guilt of the accused. For
73BD
Figure 1.1. Peter Wason’s Selection Task for measuring confirmation bias. Four cards are
placed face down on a table. You’re told that if there is letter on one side then there will
always be a number on the other side. Then you are given a specific hypothesis: If there is
a D on one side then there is a 3 on its other side. Which cards would you turn over to test
whether this hypothesis is true or false?
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bias | 7
instance, to test whether a person was a witch, the suspect would often be
plunged into water with stones tied to her feet. If she rose then she would
be proven a witch and burned at the stake. If she drowned then she was
usually considered innocent or a witch of lesser power. Either way, being
suspected of witchcraft was tantamount to a death sentence within a legal
framework that sought only to confirm accusations. Similar biases are apparent
in many aspects of modern life. Popular TV programs such as CSI
fuel the impression that forensic science is bias-free and infallible, but in
reality the field is plagued by confirmation bias.9 Even at the most highly
regarded agencies in the world, forensic examiners can be biased toward
interpreting evidence that confirms existing suspicions. Doing so can lead
to wrongful convictions, even when evidence is based on harder data such
as fingerprints and DNA tests.
Confirmation bias also crops up in the world of science communication.
For many years it was assumed that the key to more effective public communication
of science was to fill the public’s lack of knowledge with facts—
the so-called deficit model.10 More recently, however, this idea has been
discredited because it fails to take into account the prior beliefs of the audience.
The extent to which we assimilate new information about popular
issues such as climate change, vaccines, or genetically modified foods is
susceptible to a confirmation bias in which evidence that is consistent with
our preconceptions is favored, while evidence that flies in the face of them
is ignored or attacked. Because of this bias, simply handing people more
facts doesn’t lead to more rational beliefs. The same problem is reflected in
politics. In his landmark 2012 book, the Geek Manifesto, Mark Henderson
laments the cherry-picking of evidence by politicians in order to reinforce
a predetermined agenda. The resulting “policy-based evidence” is a perfect
example of confirmation bias in practice and represents the antithesis of
how science should be used in the formulation of evidence-based policy.
If confirmation bias is so irrational and counterproductive, then why
does it exist? Many different explanations have been suggested based on
cognitive or motivational factors. Some researchers have argued that it reflects
a fundamental limit of human cognition. According to this view, the
fact that we have incomplete information about the world forces us to rely
on the memories that are most easily retrieved (the so-called availability
heuristic), and this reliance could fuel a bias toward what we think we already
know. On the other hand, others have argued that confirmation bias
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
8 | Chapter 1
is the consequence of an innate “positive-test strategy”—a term coined in
1987 by psychologists Joshua Klayman and Young-Won Ha.11 We already
know that people find it easier to judge whether a positive statement is true
or false (e.g., “there are apples in the basket”) compared to a negative one
(“there are no apples in the basket”). Because judgments of presence are
easier than judgments of absence, it could be that we prefer positive tests
of reality over negative ones. By taking the easy road, this bias toward positive
thoughts could lead us to wrongly accept evidence that agrees positively
with our prior beliefs.
Against this backdrop of explanations for why an irrational bias is so
pervasive, psychologists Hugo Mercier and Dan Sperber have suggested that
confirmation bias is in fact perfectly rational in a society where winning
arguments is more important than establishing truths.12 Throughout our
upbringing, we are taught to defend and justify the beliefs we hold, and less
so to challenge them. By interpreting new information according to our
existing preconceptions we boost our self-confidence and can argue more
convincingly, which in turn increases our chances of being regarded as
powerful and socially persuasive. This observation leads us to an obvious
proposition: If human society is constructed so as to reward the act of winning
rather than being correct, who would be surprised to find such incentives
mirrored in scientific practices?
Neophilia: When the Positive and New
Trumps the Negative but True
The core of any research psychologist’s career—and indeed many scientists
in general—is the rate at which they publish empirical articles in highquality
peer-reviewed journals. Since the peer-review process is competitive
(and sometimes extremely so), publishing in the most prominent journals
equates to a form of “winning” in the academic game of life.
Journal editors and reviewers assess submitted manuscripts on many
grounds. They look for flaws in the experimental logic, the research methodology,
and the analyses. They study the introduction to determine
whether the hypotheses are appropriately grounded in previous research.
They scrutinize the discussion to decide whether the paper’s conclusions
are justified by the evidence. But reviewers do more than merely critique
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bias | 9
the rationale, methodology, and interpretation of a paper. They also study
the results themselves. How important are they? How exciting? How much
have we learned from this study? Is it a breakthrough? One of the central
(and as we will see, lamentable) truths in psychology is that exciting positive
results are a key factor in publishing—and often a requirement. The
message to researchers is simple: if you want to win in academia, publish
as many papers as possible in which you provide positive, novel results.
What does it mean to find “positive” results? Positivity in this context
doesn’t mean that the results are uplifting or good news—it refers to
whether the researchers found a reliable difference in measurements, or a
reliable relationship, between two or more study variables. For example,
suppose you wanted to test the effect of a cognitive training intervention
on the success of dieting in people trying to lose weight. First you conduct
a literature review, and, based on previous studies, you decide that boosting
people’s self-control might help. Armed with a good understanding of existing
work, you design a study that includes two groups. The experimental
group perform a computer task in which they are trained to respond to images
of foods, but crucially, to refrain from responding to images of particular
junk foods. They perform this task every day for six weeks, and you
measure how much weight they lose by the end of the experiment. The
control group does a similar task with the same images but responds to all
of them—and you measure weight loss in that group as well.
The null hypothesis (called “H0”) in this case is that there should be no
difference in weight loss—your training intervention has no effect on
whether people gain or lose weight. The alternative hypothesis (called “H1”)
is that the training intervention should boost people’s ability to refrain from
eating junk foods, and so the amount of weight loss should be greater in the
treatment group compared with the control group. A positive result would
be finding a statistically significant difference in weight loss between the
groups (or in technical terms, “rejecting H0”), and a negative result would
be failing to show any significant difference (or in other words, “failing to
reject H0”). Note how I use the term “failing.” This language is key because,
in our current academic culture, journals indeed regard such outcomes as
scientific failures. Regardless of the fact that the rationale and methods are
identical in each outcome, psychologists find negative results much harder
to publish than positive results. This is because positive results are regarded
by journals as reflecting a greater degree of scientific advance and interest
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
10 | Chapter 1
to readers. As one journal editor said to me, “Some results are just more
interesting and important than others. If I do a randomized trial on a novel
intervention based on a long-shot and find no effect that is not a great leap
forward. However, if the same study shows a huge benefit that is a more
important finding.”
This publication bias toward positive results also arises because of the
nature of conventional statistical analyses in psychology. Using standard
methods developed by Neyman and Pearson, positive results reject H0 in
favor of the alternative hypothesis (H1). This statistical approach—called
null hypothesis significance testing—estimates the probability (p) of an effect
of the same or greater size being obtained if the null hypothesis were
true. Crucially, it doesn’t estimate the probability of the null hypothesis itself
being true: p values estimate the probability of a given effect or more extreme
arising given the hypothesis, rather than the probability of a particular
hypothesis given the effect. This means that while a statistically significant
result (by convention, p<.05) allows the researcher to reject H0, a
statistically nonsignificant result (p>.05) doesn’t allow the researcher to
accept H0. All the researcher can conclude from a statistically nonsignificant
outcome is that H0 might be true, or that the data might be insensitive.
The interpretation of statistically nonsignificant effects is therefore inherently
inconclusive.
Consider the thought process this creates in the minds of researchers. If
we can’t test directly whether there is no difference between experimental
conditions, then it makes little sense to design an experiment in which the
null hypothesis would ever be the focus of interest. Instead, psychologists
are trained to design experiments in which findings of interest would always
be positive. This bias in experimental design, in turn, means that
students in psychology enter their research careers reciting the mantra
“Never predict the null hypothesis.” If researchers can never predict the
null hypothesis, and if positive results are considered more interesting to
journals than negative results, then the inevitable outcome is a bias in
which the peer-reviewed literature is dominated by positive findings that
reject H0 in favor of H1, and in which most of the negative or nonsignificant
results remain unpublished. To ensure that they keep winning in the academic
game, researchers are thus pushed into finding positive results that
agree with their expectations—a mechanism that incentivizes and rewards
confirmation bias.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bias | 11
All this might sound possible in theory, but is it true? Psychologists have
known since the 1950s that journals are predisposed toward publishing
positive results, but, historically, it has been difficult to quantify how much
publication bias there really is in psychology.13 One of the most compelling
analyses was reported in 2010 by psychologist Daniele Fanelli from the
University of Edinburgh.14 Fanelli reasoned, as above, that any domain of
the scientific literature that suffers from publication bias should be dominated
by positive results that support the stated hypothesis (H1). To test this
idea, he collected a random sample of more than 2,000 published journal
articles from across the full spectrum of science, ranging from the space
sciences to physics and chemistry, through to biology, psychology, and psychiatry.
The results were striking. Across all sciences, positive outcomes
were more common than negative ones. Even for space science, which published
the highest percentage of negative findings, 70 percent of the sampled
articles supported the stated hypothesis. Crucially, this bias was highest in
psychology, topping out at 91 percent. It is ironic that psychology—the discipline
that produced the first empirical evidence of confirmation bias—is
at the same time one of the most vulnerable to confirmation bias.
The drive to publish positive results is a key cause of publication bias, but
it still explains only half the problem. The other half is the quest for novelty.
To compete for publication at many journals, articles must either adopt a
novel methodology or produce a novel finding—and preferably both. Most
journals that publish psychological research judge the merit of manuscripts,
in part, according to novelty. Some even refer explicitly to novelty as a policy
for publication. The journal Nature states that to be considered for peer
review, results must be “novel” and “arresting,”15 while the journal Cortex
notes that empirical Research Reports must “report important and novel
material.”16 The journal Brain warns authors that “some [manuscripts] are
rejected without peer review owing to lack of novelty,”17 and Cerebral Cortex
goes one step further, noting that even after peer review, “final acceptance
of papers depends not just on technical merit, but also on subjective ratings
of novelty.”18 Within psychology proper, Psychological Science, a journal that
claims to be the highest-ranked in psychology, prioritizes papers that produce
“breathtaking” findings.19
At this point, you might well ask: what’s wrong with novelty? After all,
in order for something to be marked as discovered, surely it can’t have been
observed already (so it must be a novel result), and isn’t it also reasonable
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
12 | Chapter 1
to assume that researchers seeking to produce novel results might need to
adopt new methods? In other words, by valuing novelty aren’t journals simply
valuing discovery? The problem with this argument is the underlying
assumption that every observation in psychological research can be called
a discovery—that every paper reports a clear and definitive fact. As with all
scientific disciplines, this is far from the truth. Most research findings in
psychology are probabilistic rather than deterministic: conventional statistical
tests talk to us in terms of probabilities rather than proofs. This in turn
means that no single study and no one paper can lay claim to a discovery.
Discovery depends wholly and without exception on the extent to which
the original results can be repeated or replicated by other scientists, and not
just once but over and over again. For example, it would not be enough to
report only once that a particular cognitive therapy was effective at reducing
depression; the result would need to be repeated many times in different
groups of patients, and by different groups of researchers, for it be widely
adopted as a public health intervention. Once a result has been replicated
a satisfactory number of times using the same experimental method, it can
then be considered replicable and, in combination with other replicable evidence,
can contribute meaningfully to the theoretical or applied framework
in which it resides. Over time, this mass accumulation of replicable evidence
within different fields can allow theories to become accepted through consensus
and in some cases can even become laws.
In science, prioritizing novelty hinders rather than helps discovery because
it dismisses the value of direct (or close) replication. As we have seen,
journals are the gatekeepers to an academic career, so if they value findings
that are positive and novel, why would scientists ever attempt to replicate
each other? Under a neophilic incentive structure, direct replication is discarded
as boring, uncreative, and lacking in intellectual prowess.
Yet even in a research system dominated by positive bias and neophilia,
psychologists have retained some realization that reproducibility matters.
So, in place of unattractive direct replication, the community has reached
for an alternative form of validation in which one experiment can be said
to replicate the key concept or theme of another by following a different
(novel) experimental method—a process known as conceptual replication.
On its face, this redefinition of replication appears to satisfy the need to
validate previous findings while also preserving novelty. Unfortunately, all
it really does is introduce an entirely new and pernicious form of confirmation
bias.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bias | 13
Replicating Concepts Instead of Experiments
In early 2012, a professor of psychology at Yale University named John
Bargh launched a stinging public attack on a group of researchers who failed
to replicate one of his previous findings.20 The study in question, published
by Bargh and colleagues in 1996, reported that priming participants unconsciously
to think about concepts related to elderly people (e.g., words
such as “retired,” “wrinkle,” and “old”) caused them to walk more slowly
when leaving the lab at the end of the experiment.21 Based on these findings,
Bargh claimed that people are remarkably susceptible to automatic
effects of being primed by social constructs.
Bargh’s paper was an instant hit and to date has been cited more than
3,800 times. Within social psychology it spawned a whole generation of
research on social priming, which has since been applied in a variety of
different contexts. Because of the impact the paper achieved, it would be
reasonable to expect that the central finding must have been replicated
many times and confirmed as being sound. Appearances, however, can be
deceiving.
Several researchers had reported failures to replicate Bargh’s original
study, but few of these nonreplications have been published, owing to the
fact that journals (and reviewers) disapprove of negative findings and often
refuse to publish direct replications. One such attempted replication in 2008
by Hal Pashler and colleagues from the University of California San Diego
was never published in an academic journal and instead resides at an online
repository called PsychFileDrawer.22 Despite more than doubling the sample
size reported in the original study, Pashler and his team found no evidence
of such priming effects—if anything they found the opposite result.
Does this mean Bargh was wrong? Not necessarily. As psychologist Dan
Simons from the University of Illinois has noted, failing to replicate an effect
does not necessarily mean the original finding was in error.23 Nonreplications
can emerge by chance, can be due to subtle changes in experimental
methods between studies, or can be caused by the poor methodology of
the researchers attempting the replication. Thus, nonreplications are themselves
subject to the same tests of replicability as the studies they seek to
replicate.
Nevertheless, the failed replication by Pashler and colleagues—themselves
an experienced research team—raised a question mark over the status
of Bargh’s original study and hinted at the existence of an invisible file drawer
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
14 | Chapter 1
of unpublished failed replications. In 2012, another of these attempted replications
came to light when Stéphane Doyen and colleagues from the University
of Cambridge and Université Libre de Bruxelles also failed to replicate
the elderly priming effect.24 Their article appeared prominently in the peerreviewed
journal PLOS ONE, one of the few outlets worldwide that explicitly
renounces neophilia and publication bias. The ethos of PLOS ONE is to publish
any methodologically sound scientific research, regardless of subjective
judgments as to its perceived importance or originality. In their study, Doyen
and colleagues not only failed to replicate Bargh’s original finding but also
provided an alternative explanation for the original effect—rather than being
due to a priming manipulation, it was the experimenters themselves who
unwittingly induced the participants to walk more slowly by behaving differently
or even revealing the hypothesis.
The response from Bargh was swift and contemptuous. In a highly publicized
blogpost at psychologytoday.com entitled “Nothing in Their
Heads,”25 he attacked not only Doyen and colleagues as “incompetent or
ill-informed,” but also science writer Ed Yong (who covered the story)26 for
engaging in “superficial online science journalism,” and PLOS ONE as a
journal that “quite obviously does not receive the usual high scientific journal
standards of peer-review scrutiny.” Amid a widespread backlash against
Bargh, his blogpost was swiftly (and silently) deleted but not before igniting
a fierce debate about the reliability of social priming research and the status
of replication in psychology more generally.
Doyen’s article, and the response it generated, didn’t just question the
authenticity of the elderly priming effect; it also exposed a crucial disagreement
about the definition of replication. Some psychologists, including
Bargh himself, claimed that the original 1996 study had been replicated at
length, while others claimed that it had never been replicated. How is this
possible?
The answer, it turned out, was that different researchers were defining
replication differently. Those who argued that the elderly priming effect had
never been replicated were referring to direct replications: studies that repeat
the method of a previous experiment as exactly as possible in order to reproduce
the finding. At the time of writing, Bargh’s central finding has been
directly replicated just twice, and in each case with only partial success. In
the first attempt, published six years after the original study,27 the researchers
showed the same effect but only in a subgroup of participants who scored
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bias | 15
high on self-consciousness. In the second attempt, published another four
years later, a different group of authors showed that priming elderly concepts
slowed walking only in participants who held positive attitudes about elderly
people; those who harbored negative attitudes showed the opposite effect.28
Whether these partial replications are themselves replicable is unknown,
but as we will see in chapter 2, hidden flexibility in the choices researchers
make when analyzing their data (particularly concerning subgroup analyses)
can produce spurious differences where none truly exist.
In contrast, those who argued that the elderly priming effect had been
replicated many times were referring to the notion of “conceptual replication”:
the idea that the principle of unconscious social priming demonstrated
in Bargh’s 1996 study has been extended and applied in many different
contexts. In a later blog post at psychologytoday.com called “Priming Effects
Replicate Just Fine, Thanks,” Bargh referred to some of these conceptual
replications in variety of social behaviors, including attitudes and stereotypes
unrelated to the elderly.29
The logic of “conceptual replication” is that if an experiment shows evidence
for a particular phenomenon, you can replicate it by using a different
method that the experimenter believes measures the same class of phenomenon.
Psychologist Rolf Zwaan argues that conceptual replication has a
legitimate role in psychology (and indeed all sciences) to test the extent to
which particular phenomena depend on specific laboratory conditions, and
to determine whether they can be generalized to new contexts.30 The current
academic culture, however, has gone further than merely valuing conceptual
replication—it has allowed it to usurp direct replication. As much
as we all agree about the importance of converging evidence, should we be
seeking it out at the expense of knowing whether the phenomenon being
generalized exists in the first place?
A reliance on conceptual replication is dangerous for three reasons.31
The first is the problem of subjectivity. A conceptual replication can hold
only if the different methods used in two different studies are measuring
the same phenomenon. For this to be the case, some evidence must exist
that they are. Even if we meet this standard, this raises the question of how
similar the methods must be for a study to qualify as being conceptually
replicated. Who decides and by what criteria?
The second problem is that a reliance on conceptual replications risks
findings becoming unreplicated in the future. To illustrate how this could
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
16 | Chapter 1
happen, suppose we have three researchers, Smith, Jones, and Brown, who
publish three scientific papers in sequence. Smith publishes the first paper,
showing evidence for a particular phenomenon. Jones then uses a different
method to show evidence for a phenomenon that appears similar to the one
that Smith discovered. The psychological community decide that the similarity
crosses some subjective threshold and so conclude that Jones “conceptually
replicates” Smith. Now enter Brown. Brown isn’t convinced that
Smith and Jones are measuring the same phenomenon and suspects they
are in fact describing different phenomena. Brown obtains evidence suggesting
that this is indeed the case. In this way, Smith’s finding that was previously
considered replicated by Jones now assumes the bizarre status of becoming
unreplicated.
Finally, conceptual replication fuels an obvious confirmation bias. When
two studies draw similar conclusions using different methods, the second
study can be said to conceptually replicate the first. But what if the second
study draws a very different conclusion—would it be claimed to conceptually
falsify the first study? Of course not. Believers of the original finding
would immediately (and correctly) point to the multitude of differences in
methodology to explain the different results. Conceptual replications thus
force science down a one-way street in which it is possible to confirm but
never disconfirm previous findings. Through a reliance on conceptual replication,
psychology has found yet another way to become enslaved to confirmation
bias.
Reinventing History
So far we have seen how confirmation bias influences psychological science
in two ways: through the pressure to publish results that are novel and positive,
and by ousting direct replication in favor of bias-prone conceptual
replication. A third, and especially insidious, manifestation of confirmation
bias can be found in the phenomenon of hindsight bias. Hindsight bias is a
form of creeping determinism in which we fool ourselves (and others) into
believing that an observation was expected even though it actually came as
a surprise.
It may seem extraordinary that any scientific discipline should be vulnerable
to a fallacy that attempts to reinvent history. Indeed, under the classic
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bias | 17
hypothetico-deductive (H-D) model of the scientific method, the research
process is supposed to be protected against such bias (see figure 1.2). According
to the H-D method, to which psychology at least nominally adheres,
a scientist begins by formulating a hypothesis that addresses some aspect
of a relevant theory. With the hypothesis decided, the scientist then conducts
an experiment and allows the data to determine whether or not the
Generate and
specify hypotheses
Design study
Low statistical power
Lack of replication
p-hacking
Publication bias
Lack of data sharing
p-hacking
HARKing
Publish
Analyze data and
test hypotheses
Collect data
Interpret
data
Figure 1.2. The hypothetico-deductive model of the scientific method is compromised by
a range of questionable research practices. Lack of replication impedes the elimination
of false discoveries and weakens the evidence base underpinning theory. Low statistical
power (to be discussed in chapter 3) increases the chances of missing true discoveries and
reduces the probability that obtained positive effects are real. Exploiting researcher degrees
of freedom (p-hacking—to be discussed in chapter 2) manifests in two general
forms: collecting data until analyses return statistically significant effects, and selectively
reporting analyses that reveal desirable outcomes. HARKing, or Hypothesizing After Results
are Known, involves generating a hypothesis from the data and then presenting it as
a priori. Publication bias occurs when journals reject manuscripts on the basis that they
report negative or otherwise unattractive findings. Finally, lack of data sharing (to be
discussed in chapter 4) prevents detailed meta-analysis and hinders the detection of data
fabrication.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
18 | Chapter 1
hypothesis was supported. This outcome then feeds into revision (and possible
rejection) of the theory, stimulating an iterative cycle of hypothesis
generation, hypothesis testing, and theoretical advance. A central feature
of the H-D method is that the hypothesis is decided before the scientist collects
and analyzes the data. By separating in time the prediction (hypothesis)
from the estimate of reality (data), this method is designed to protect
scientists from their own hindsight bias.
Unfortunately, much psychological research seems to pay little heed to
this aspect of the scientific method. Since the hypothesis of an experiment
is only rarely published in advance, researchers can covertly alter their predictions
after the data have been analyzed in the interests of narrative flair.
In psychology this practice is referred to as Hypothesizing After Results are
Known (HARKing), a term coined in 1998 by psychologist Norbert Kerr.32
HARKing is a form of academic deception in which the experimental hypothesis
(H1) of a study is altered after analyzing the data in order to pretend
that the authors predicted results that, in reality, were unexpected. By
engaging in HARKing, authors are able to present results that seem neat
and consistent with (at least some) existing research or their own previously
published findings. This flexibility allows the research community to produce
the kind of clean and confirmatory papers that psychology journals
prefer while also maintaining the illusion that the research is hypothesis
driven and thus consistent with the H-D method.
HARKing can take many forms, but one simple approach involves reversing
the predictions after inspecting the data. Suppose that a researcher
formulates the hypothesis that, based on the associations we form across
our lifetime between the color red and various behavioral acts of stopping
(e.g., traffic lights; stop signs; hazard signs), people should become more
cautious in a gambling task when the stimuli used are red rather than white.
After running the experiment, however, the researcher finds the opposite
result: people gambled more when exposed to red stimuli. According to the
H-D method, the correct approach here would be to report that the hypothesis
was unsupported, admitting that additional experiments may be required
to understand how this unexpected result arose and its theoretical
implications. However, the researcher realizes that this conclusion may be
difficult to publish without conducting those additional experiments, and
he or she also knows that nobody reviewing the paper would be aware that
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bias | 19
the original hypothesis was unsupported. So, to create a more compelling
narrative, the researcher returns to the literature and searches for studies
suggesting that being exposed to the color red can lead people to “see red,”
losing control and becoming more impulsive. Armed with a small number
of cherry-picked findings, the researcher ignores the original (better
grounded) rationale and rewrites the hypothesis to predict that people will
actually gamble more when exposed to red stimuli. In the final published
paper, the introduction section is written with this post hoc hypothesis
presented as a priori.
Just how prevalent is this kind of HARKing? Norbert Kerr’s survey of
156 psychologists in 1998 suggested that about 40 percent of respondents
had observed HARKing by other researchers; strikingly, the surveyed psychologists
also suspected that HARKing was about 20 percent more prevalent
than the classic H-D method.33 A more recent survey of 2,155 psychologists
by Leslie John and colleagues estimated the true prevalence rate to be
as high as 90 percent despite a self-admission rate of just 35 percent.34
Remarkably, not all psychologists agree that HARKing is a problem.
Nearly 25 years before suggesting the existence of precognition, Daryl Bem
claimed that if data are “strong enough” then researchers are justified in
“subordinating or even ignoring [their] original hypotheses.”35 In other
words, Bem argued that it is legitimate to subvert the H-D method, and to
do so covertly, in order to preserve the narrative structure of a scientific
paper.
Norbert Kerr and others have objected to this point of view, as well they
might. First and foremost, because HARKing relies on deception, it violates
the fundamental ethical principle that research should be reported honestly
and completely. Deliberate HARKing may therefore lie on the same continuum
of malpractice as research fraud. Secondly, the act of deception in
HARKing leads the reader to believe that an obtained finding was more
expected, and hence more reliable, than it truly is—this, in turn, risks distorting
the scientific record to place undue certainty in particular findings
and theories. Finally, in cases where a post hoc hypothesis is pitted against
an alternative account that the author already knows was unsupported,
HARKing creates the illusion of competitive hypothesis testing. Since a
HARKed hypothesis can, by definition, never be disconfirmed, this contrived
scenario further exacerbates confirmation bias.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
20 | Chapter 1
The Battle against Bias
If confirmation bias is so much a part of human nature then what hope can
we have of defeating it in science? In an academic culture that prizes novel
results that confirm our expectations, is there any real chance of reform?
We have known about the various manifestations of bias in psychology since
the 1950s—and have done little to counteract them—so it is easy to see why
many psychologists are cynical about the prospect of change. However, the
tide is turning. Chapter 8 will address the set of changes we must make—
and are already launching—to protect psychological science against bias
and the other “deadly sins” that have become part of our academic landscape.
Some of these reforms are already bearing fruit.
Our starting point for any program of reform must be the acceptance
that we can never completely eliminate confirmation bias—in Nietzsche’s
words we are human, all too human. Decades of psychological research shows
how bias is woven into the fabric of cognition and, in many situations, operates
unconsciously. So, rather than waging a fruitless war on our own nature,
we would do better to accept imperfection and implement measures
that protect the outcome of science as much as possible from our inherent
flaws as human practitioners.
One such protection against bias is study preregistration. We will return
to the details of preregistration in chapter 8, but for now it is useful to consider
how publicly registering our research intentions before we collect data
can help neutralize bias. Consider the three main manifestations of confirmation
bias in psychology: publication bias, conceptual replication, and
HARKing. In each case, a strong motivation for engaging in these practices
is not to generate high-quality, replicable science, but to produce results that
are publishable and perceived to be of interest to other scientists. Journals
enforce publication bias because they believe that novel, positive results are
more likely to indicate discoveries that their readers will want to see; by
comparison, replications and negative findings are considered boring and
relatively lacking in intellectual merit. To fit with the demands of journals,
psychologists have thus replaced direct replication with conceptual replication,
maintaining the comfortable but futile delusion that our science values
replication while still satisfying the demands of novelty and originality.
Finally, as we have seen, many researchers engage in HARKing because
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bias | 21
they realize that failing to confirm their own hypothesis is regarded as a
form of intellectual failure.
Study preregistration helps overcome these problems by changing the
incentive structure to value “good science” over and above “good results.”
The essence of preregistration is that the study rationale, hypotheses, experimental
methods, and analysis plan are stated publicly in advance of
collecting data. When this process is undertaken through a peer-reviewed
journal, it forces journal editors to make publishing decisions before results
exist. This, in turn, prevents publication bias by ensuring that whether results
are positive or negative, novel or familiar, groundbreaking or incremental,
is irrelevant to whether the science will be published. Similarly,
since authors will have stated their hypotheses in advance, preregistration
prevents HARKing and ensures adherence to the H-D model of the scientific
method. As we will see in chapter 2, preregistration also prevents researchers
from cherry-picking results that they believe generate a desirable
narrative.
In addition to study preregistration, bias can be reduced by reforming
statistical practice. As discussed earlier, one reason negative findings are
regarded as less interesting is our cultural reliance on null hypothesis significance
testing (NHST). NHST can only ever tell us whether the null
hypothesis is rejected, and never whether it is supported. Our reliance on
this one-sided statistical approach inherently places greater weight on positive
findings. However, by shifting to alternative Bayesian statistical methods,
we can test all potential hypotheses (H0, H1 . . . Hn) fairly as legitimate
possible outcomes. We will explore this alternative method in more detail
in chapter 3.
As we take this journey it is crucial that individual scientists from every
level feel empowered to promote reform without damaging their careers.
Confirmation bias is closely allied with “groupthink”—a pernicious social
phenomenon in which a consensus of behavior is mistaken for a convergence
of informed evidence. The herd doesn’t always make the most rational
or intelligent decisions, and groupthink can stifle innovation and critical
reflection. To ensure the future of psychological science, it is incumbent on
us as psychologists to recognize and challenge our own biases.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
C H A P T E R 2
The Sin of Hidden Flexibility
Torture numbers and they will confess to anything.
—Gregg Easterbrook, 1999
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
In 2008, British illusionist Derren Brown presented a TV program called
The System in which he claimed he could predict, with certainty, which
horse would win at the racetrack. The show follows Khadisha, a member
of the public, as Brown provides her with tips on upcoming races. In each
case the tips pay off, and after succeeding five times in a row Khadisha decides
to bet as much money as possible on a sixth and final race. The twist
in the program is that Brown has no system—Khadisha is benefiting from
nothing more than chance. Unknown to her until after placing her final
bet, Brown initially recruited 7,776 members of the public and provided
each of them with a unique combination of potential winners. Participants
with a losing horse were successively eliminated at each of six stages, eventually
leaving just one participant who had won every time—and that person
just happened to be Khadisha. By presenting the story from Khadisha’s
perspective, Brown created the illusion that her winning streak was too
unlikely to be random—and so must be caused by The System—when in
fact it was explained entirely by chance.
Unfortunately for science, the hidden flexibility that Brown used to generate
false belief in The System is the same mechanism that psychologists
exploit to produce results that are attractive and easy to publish. Faced with
the career pressure to publish positive findings in the most prestigious and
selective journals, it is now standard practice for researchers to analyze
complex data in many different ways and report only the most interesting
and statistically significant outcomes. Doing so deceives the audience into
believing that such outcomes are credible, rather than existing within an
ocean of unreported negative or inconclusive findings. Any conclusions
drawn from such tests will, at best, overestimate of the size of any real effect.
At worst they could be entirely false.
By torturing numbers until they produce publishable outcomes, psychology
commits our second mortal sin: that of exploiting hidden analytic flexibility.
Formally, hidden flexibility is one manifestation of the “fallacy of
incomplete evidence,” which arises when we frame an argument without
taking into account the full set of information available. Although hidden
flexibility is itself a form of research bias, its deep and ubiquitous nature in
psychology earns it a dedicated place in our hall of shame.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
24 | Chapter 2
p-Hacking
As we saw earlier, the dominant approach for statistical analysis in psychological
science is a set of techniques called null hypothesis significance
testing (NHST). NHST estimates the probability of an obtained positive
effect, or one greater, being observed in a set of data if the null hypothesis
(H0) is true and no effect truly exists. Importantly, the p value doesn’t tell
us the probability of H0 itself being true, and it doesn’t indicate the size or
reliability of the obtained effect—instead what it tells us is how surprised
we should be to obtain the current effect, or one more extreme, if H0 were
to be true.1 The smaller the p value, the greater our surprise would be and
the more confidently we can reject H0.
Since the 1920s, the convention in psychology has been to require a p
value of less than .05 in order to categorically reject H0. This significance
threshold is known as α—the probability of falsely declaring a positive effect
when, in fact, there isn’t one. Under NHST, a false positive or Type I
error occurs when we incorrectly reject a true H0. The α threshold thus
indicates the maximum allowable probability of a Type I error in order to
reject H0 and conclude that a statistically significant effect is present.
Why is α set to .05, you might ask? The .05 convention is arbitrary,
as noted by Ronald Fisher—one of the architects of NHST—nearly a century
ago:
If one in twenty does not seem high enough odds, we may, if we prefer
it, draw the line at one in fifty (the 2 per cent point), or one in a
hundred (the 1 per cent point). Personally, the writer prefers to set a
low standard of significance at the 5 per cent point, and ignore entirely
all results which fail to reach this level.2
Setting the α threshold to .05 theoretically allows up to 1 in 20 false
rejections of H0 across a set of independent significance tests. Some have
argued that this threshold is too liberal and leads to a scientific literature
built on weak findings that are unlikely to replicate.3 Furthermore, even if
we believe that it is acceptable for 5 percent of statistically significant results
to be false positives, the truth is that exploiting analytic flexibility increases
α even more, increasing the actual rate of false positives.
This flexibility arises because researchers make analytic decisions after
inspecting their data and are faced with many analysis options that can be
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Hidden Flexibility | 25
considered defensible yet produce slightly different p values. For instance,
given a distribution of reaction time values, authors have the option of excluding
statistical outliers (such as very slow responses) within each participant.
They also have the option of excluding entire participants on the
same basis. If they decide to adopt either or both of these approaches, there
are then many available methods they could use, each of which could produce
slightly different results. As well as being flexible, a key feature of such
decisions is that they are hidden and never published. The rules of engagement
do not require authors to specify which analytic decisions were a priori
(confirmatory) and which were post hoc (exploratory)—in fact, such transparency
is likely to penalize authors competing for publication in the most
prestigious journals. This combination of culture and incentives inevitably
leads to all analyses being portrayed as confirmatory and hypothesis driven
even where many were exploratory. In this way, authors can generate a
product that is attractive to journals while also maintaining the illusion
(and possibly delusion) that they have adhered to the hypothetico-deductive
model of the scientific method.
The decision space in which these exploratory analyses reside is referred
to as “researcher degrees of freedom.” Beyond the exclusion of outliers, it
can include decisions such as which conditions to enter into a wider analysis
of multiple factors, which covariates or regressors to take into account,
whether or not to collect additional participants, and even how to define
the dependent measure itself. In even the simplest experimental design,
these pathways quickly branch out to form a complex decision tree that a
researcher can navigate either deliberately or unconsciously in order to
generate statistically significant effects. By selecting the most desirable outcomes,
it is possible to reject H0 in almost any set of data—and by combining
selective reporting with HARKing (as described in chapter 1) it is possible
to do so in favor of almost any alternative hypothesis.
Exploiting researcher degrees of freedom to generate statistical significance
is known as “p-hacking” and was brought to prominence in 2011 by
Joe Simmons, Leif Nelson, and Uri Simonsohn from the University of Pennsylvania
and University of California Berkeley.4 Through a combination of
real experiments and simulations, Simmons and colleagues showed how
selective reporting of exploratory analyses can generate meaningless p values.
In one such demonstration, the authors simulated a simple experiment
involving one independent variable (the intervention), two dependent vari-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
26 | Chapter 2
ables (behaviors being measured), and a single covariate (gender of the
participant). The simulation was configured so that there was no effect of
the manipulation, that is, H0 was predetermined to be true. They then
simulated the outcome of the experiment 15,000 times and asked how often
at least one statistically significant effect was observed (p<.05). Given that
H0 was true in this scenario, a nominal rate of 5 percent false positives can
be assumed at α = .05. The central question posed by Simmons and colleagues
was what would happen if they embedded hidden flexibility within
the analysis decisions. In particular, they tested the effect of analyzing either
of the two dependent variables (reporting if a positive effect was obtained
on either one), including gender as a covariate or not, increasing the
number of participants after analyzing the results, and dropping one or
more of the conditions. Allowing maximal combinatorial flexibility between
these four options increased the false positive rate from a nominal 5 percent
to an alarming 60.7 percent.
As striking as this is, 60.7 percent is probably still an underestimate of
the true rate of false positives in many psychology experiments. Simmons
and colleagues didn’t even include other common forms of hidden flexibility,
such as variable criteria for excluding outliers or conducting exploratory
analyses within subgroups (e.g., males only or females only). Their
simulated experimental design was also relatively simple and produced
only a limited range of researcher degrees of freedom. In contrast, many
designs in psychology are more complicated and will include many more
options. Using a standard research design with four independent variables
and one dependent variable, psychologist Dorothy Bishop has showed that
at least one statistically significant main effect or interaction can be expected
by chance in more than 50 percent of analyses—an order of magnitude
higher than the conventional α threshold.5 Crucially, this rate of
false positives occurs even without exploiting the researcher degrees of
freedom illustrated by Simmons and colleagues. Thus, where p-hacking
occurs in more complex designs it is likely to render the obtained p values
completely worthless.
One key source of hidden flexibility in the simulations by Simmons and
colleagues was the option to add participants after inspecting the results.
There are all kinds of reasons why researchers peek at data before data collection
is complete, but one central motivation is efficiency: in an environment
with limited resources it can often seem sensible to stop data collec-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Hidden Flexibility | 27
tion as soon as all-important statistical significance is either obtained or
seems out of reach. This temptation to peek and chase p<.05 is of course
motivated by the fact that psychology journals typically require the main
conclusions of a paper to be underpinned by statistically significant results.
If a critical statistical test returns p = .07, the researcher knows that reviewers
and editors will regard the result as weak and unconvincing, and that
the paper has little chance of being published in a competitive journal.
Many researchers will therefore add participants in an attempt to nudge
the p value over the line, without reporting in the published paper that they
did so.
This kind of behavior may seem rational within a publishing system
where what is best for science conflicts with the incentives that drive individual
scientists. After all, if scientists are engaging in these practices then
it is surely because they believe they have no other choice in the race for
jobs, grant funding, and professional esteem. Unfortunately, however, chasing
statistical significance by peeking and adding data completely undermines
the philosophy of NHST. A central but often overlooked requirement
of NHST is that researchers prespecify a stopping rule, which is the final
sample size at which data collection must cease. Peeking at data prior to
this end point in order to decide whether to continue or to stop increases
the chances of a false positive. This is because NHST estimates the probability
of the observed data (or more extreme) under the null hypothesis, and
since the aggregate data pattern can vary randomly as individual data points
are added, repeated hypothesis testing increases the odds that the data will,
by chance alone, fall below the nominated α level (see figure 2.1). This increase
in false positives is similar to that obtained when testing multiple
hypotheses simultaneously.
Just how common is p-hacking? The lack of transparency in the research
community makes a definitive answer impossible to glean, but important
clues can be found by returning to the 2012 study by Leslie John and colleagues
from chapter 1. Based on a survey of more than 2,000 American
psychologists, they estimated that 100 percent have, on at least one occasion,
selectively excluded data after looking at the impact of doing so, and
that 100 percent have collected more data for an experiment after seeing
whether results were statistically significant. They also estimate that 75
percent of psychologists have failed to report all conditions in an experiment,
and that more than 50 percent have stopped data collection after
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
28 | Chapter 2
Falsepositiverate:pr(p<0.05)
0.30
0.25
0.20
0.10
0.15
0.05
0.00
pvalue
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
1.0
0.0
Number of participants between successive analyses
1510 00
Sample size
302010 400
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Hidden Flexibility | 29
achieving a “desired result.” These results indicate that, far from being a
rare practice, p-hacking in psychology may be the norm.
Peculiar Patterns of p
The survey by John and colleagues suggests that p-hacking is common in
psychology, but can we detect more objective evidence of its existence? One
possible clue lies in the way p values are reported. If p-hacking is as common
as claimed, then it should distort the distribution of p values in published
work. To illustrate why, consider the following scenarios. In one, a
team of researchers collect data by adding one participant at a time and
successively analyze the results after each participant until statistical significance
is obtained. Given this strategy, what p value would you expect
to see at the end of the experiment? In another scenario, the team obtain
a p value of .10 but don’t have the option to collect additional data. Instead,
they try ten different methods for excluding statistical outliers. Most of
these produce p values higher than .05, but one reduces the p value to .049.
They therefore select that option in the declared analysis and don’t report
the other “failed” attempts. Finally, consider a situation where p = .08 and
the researchers have several degrees of freedom available, in terms of charFigure
2.1. The peril of adopting a flexible stopping rule in null hypothesis significance
testing. In the upper panel, an experiment is simulated in which the null hypothesis (H0)
is true and a statistical test is conducted after each new participant up to a maximum
sample size of 50. A researcher who strategically p-hacks would stop as soon as p drops
below .05 (dotted line). In this simulation, p crosses the significance threshold after collecting
data for 19 participants (red symbols), despite the fact that there is no real effect to
be discovered. In the lower panel we see how the frequency of interim analyses influences
the false positive rate, defined here as the probability of finding a p value < .05 before
reaching the maximum sample size. Each symbol in this plot is the average false
positive rate across 10,000 simulated experiments, based on a variable stopping rule when
H0 is true. If the researcher initially collects data for five participants and then reanalyses
after every subsequent set of 20 participants, the false positive rate is 0.12 (rightmost symbol),
which is roughly twice the α threshold of .05 (dotted line). Moving leftward on the
plot, as successive analyses become more frequent (i.e., peeking after fewer and fewer
participants), the false positive rate increases. If the researcher checks after every single
participant then it climbs to as high as 0.26, yielding a 1 in 4 chance that a positive result
(p<.05) would be falsely declared where none truly exists.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
30 | Chapter 2
acterizing the dependent variable—specifically, they are able report the
results either in terms of response times or performance accuracy, or an
integrated measure of both. After analyzing all these different measures
they find that the integrated measure “works best,” revealing a p value of
.037, whereas the individual measures alone reveal only nonsignificant effects
(p>.05).
Although each of these scenarios is different, they all share one thing:
in each case the researcher is attempting to push the p value just over the
line. If this is your goal then it would make sense to stop p-hacking as soon
as the p value drops below .05—after all, why spend additional resources
only to risk that an effect that is currently publishable might “disappear”
with the addition of more participants or by looking at the data in a different
way? By focusing on merely crossing the significance threshold, the
outcome should be to create a cluster of p values just below .05.
A number of individual cases of such behavior have been alleged. In a
Science paper published in 2012, researchers presented evidence that religious
beliefs could be reduced by instructing people to complete a series of
tasks that require rational, analytic thinking. Despite sample sizes across
four experiments ranging from 57 to 179, each experiment returned p values
within a range of p = .03 to p = .04. Critics have argued either that the
authors knew precisely how many participants would be required in each
experiment to attain statistical significance or that they p-hacked, consciously
or unconsciously, in order to always narrowly reach it.6 There is,
however, an entirely innocent explanation. Through no fault of the authors,
their paper could be one of many unbiased studies considered by Science,
with the journal selectively publishing the one that “struck gold” in finding
a sequence of four statistically significant effects. Where it is impossible to
distinguish biased practices by researchers from publication bias by journals,
the authors naturally deserve the benefit of the doubt.
Because of the difficulty in differentiating publication bias from researcher
bias, individual cases of p-hacking are difficult, if not impossible,
to prove. However, if p-hacking is the norm then such cases may accrue
across the literature to produce an overall preponderance of p values just
below the significance threshold. In 2012, psychologists E. J. Masicampo
and Daniel Lalande asked this question for the first time by examining the
distribution of 3,627 p values sampled from three of the most prestigious
psychology journals. Overall they found that smaller p values were more
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Hidden Flexibility | 31
likely to be published than larger ones, but they also discovered that the
number of p values just below .05 was about five times higher than
expected.7
Masicampo and Lalande’s findings have since been replicated by Nathan
Leggett and colleagues from the University of Adelaide. Not only did they
find the same spike in p values just below .05, but they also showed that the
spike increased between 1965 and 2005.8 The reason for growing numbers
of “just-significant” results is not known for certain (and has itself been
robustly challenged),9 but if it is a genuine phenomenon then one possible
explanation is the huge advancement in computing technology and statistical
software. Undertaking NHST in 1965 was cumbersome and laborious
(and often done by hand), which acted as a natural disincentive toward
p-hacking. In contrast, modern software packages such as SPSS and R can
reanalyze data many ways in just seconds.
It has been suggested that the studies of the Masicampo team and the
Leggett team reveal evidence of p-hacking on a massive scale across thousands
of studies, but is it possible to show such effects within more specific
fields? A tool developed by Simonsohn, Leif, and Simmons called “p-curve”
analysis promises to do just this.10 The logic of p-curve is that the distribution
of statistically significant p values within a set of studies reveals their
evidential value (see figure 2.2). For unbiased (non-p-hacked) results where
H0 is false, we should see more p values clustered toward the lower end of
the spectrum (e.g., p<.01) than immediately below the significance threshold
(e.g., p values between .04 to .05). This, in turn, should produce a distribution
of p values between p = 0 and p = .05 that is positively skewed.
In contrast, when researchers engage in p-hacking we should see a clustering
of p values that is greatest just below .05, with fewer instances at lower
values—thus the distribution of p values should be negatively skewed. Although
p-curve has attracted controversy, it is a promising addition to the
existing array of tools to detect hidden analytic flexibility.11
The problem of p-hacking is not unique to psychology. Compared with
typical behavioral experiments, functional brain imaging includes far more
researcher degrees of freedom. As the blogger Neuroskeptic has pointed
out, the decision space of even the simplest fMRI study can include hundreds
of analysis options, providing ample room for p-hacking.12 At the time
of writing, no studies of the distribution of p values have yet been undertaken
for fMRI or electroencephalography (EEG), but indirect evidence
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
32 | Chapter 2
Percentageofpvalues
50
40
30
20
10
0
70
60
p value
0.01 0.02 0.050.040.03
H0 = true, p-hacking
Percentageofpvalues
50
40
30
20
10
0
70
60
p value
0.01 0.02 0.050.040.03
H0 = false, no p-hacking
Percentageofpvalues
50
40
30
20
10
0
60
p value
0.01 0.02 0.050.040.03
H0 = true, no p-hacking
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Hidden Flexibility | 33
suggests that p-hacking may be just as common in these fields as in psychological
science. Josh Carp of the University of Michigan has reported that
out of 241 randomly selected fMRI studies, 207 employed unique analysis
pipelines; this implies that fMRI researchers have numerous defensible options
at their disposal and are making those analysis decisions after inspecting
the data. As expected by a culture of p-hacking, earlier work has shown
that the test-retest reliability of fMRI is moderate to low, with an estimated
rate of false positives within the range of 10–40 percent.13 We will return
to problems with unreliability in chapter 3, but for now it is sufficient to
note that p-hacking presents a serious risk to the validity of both psychology
and cognitive neuroscience.
Institutionalized p-hacking damages the integrity of science and may be
on the rise. If virtually all psychologists engage in p-hacking (even unconsciously)
at least some of the time, and if p-hacking increases the rate of
false positives to 50 percent or higher, then much of the psychological literature
will be false to some degree. This situation is both harmful and,
crucially, preventable. It prompts us to consider how the scientific record
would change if p-hacking wasn’t the norm and challenges us to reflect on
how such practices can be tolerated in any community of scientists. Unfortunately,
p-hacking appears to have crept up on the psychological community
and become perceived as a necessary evil—the price we must pay to
publish in the most competitive journals. Frustration with the practice is,
however, growing. As Uri Simonsohn said in 2012 during a debate with
fellow psychologist Norbert Schwarz:
I don’t know of anybody who runs a study, conducts one test, and
publishes it no matter what the p-value is. . . . We are all p hackers,
those of us who realize it want change.14
Figure 2.2. The logic of the p-curve tool developed by Uri Simonsohn and colleagues.
Each plot shows a hypothetical distribution of p values between 0 and .05. For example,
the x-value of .05 corresponds to all p values between .04 and .05, while .01 corresponded
to all p values between 0 and .01. In the upper panel, the null hypothesis (H0) is true, and
there is no p-hacking; therefore p values in this plot are uniformly distributed. In the
middle panel, H0 is false, leading to a greater number of smaller p values than larger ones.
The positive (rightward) skew in this plot doesn’t rule out the presence of p-hacking but
does suggest that the sample of p values is likely to contain evidential value. In the lower
panel, H0 is true and more p values are observed closer to .05. The negative (leftward)
skew in this plot suggests the presence of p-hacking.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
34 | Chapter 2
Ghost Hunting
Could a greater emphasis on replication help solve the problems created by
p-hacking?15 In particular, if we take a case where a p-hacked finding is later
replicated, does the fact that the replication succeeded mean we don’t need
to worry whether the original finding exploited researcher degrees of
freedom?
If we assume that true discoveries will replicate more often than false
discoveries then a coordinated program of replication certainly has the
potential to weed out p-hacked findings, provided that the procedures and
analyses of the original study and the replication are identical. However,
the argument that replication neutralizes p-hacking has two major shortcomings.
The first is that while direct (close) replication is vital, even a
widespread and systematic replication initiative wouldn’t solve the problem
that p-hacked studies already waste resources by creating blind alleys. Second,
as we saw earlier and will see again later, such a direct replication
program simply doesn’t exist in psychology. Instead, the psychological community
has come to rely on the more loosely defined “conceptual replication”
to validate previous findings, which satisfies the need for journals to
publish novel and original results. Within a system that depends on conceptual
replication, researcher degrees of freedom can be just as easily exploited
to “replicate” a false discovery as to create one from scratch. A phacked
conceptual replication of a p-hacked study tells us very little about
reality apart from our ability to deceive ourselves.
The case of John Bargh and the elderly priming effect provides an interesting
case where researcher degrees of freedom may have led to so-called
phantom replication. Recall that at least two attempts to exactly replicate
the original elderly priming effect have failed.16 By way of rebuttal, Bargh
has argued that two other studies did successfully replicate the effect.17
However, if we look closely at those studies, we find that in neither case
was there an overall effect of elderly priming—in one study the effect was
statistically significant only once participants were divided into subgroups
of low or high self-consciousness; and in the other study the effect was only
significant when dividing participants into those with positive or negative
attitudes toward elderly people. Furthermore, each of these replications
used different methods for handling statistical outliers, and each analysis
included covariates that were not part of the original elderly priming experi-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Hidden Flexibility | 35
ments. These differences hint at a potential phantom replication. Driven by
the confirmation bias to replicate the original (high-profile) elderly priming
effect, the researchers in these subsequent studies may have consciously or
unconsciously exploited researcher degrees of freedom to produce a successful
replication and, in turn, a more easily marketable publication.
Whether these conceptual replications were contaminated by p-hacking
cannot be known for certain, but we have good reason to be suspicious.
Despite one-time prevalence estimates approaching 100 percent, researchers
usually deny that they p-hack.18 In Leslie John’s survey in 2012, only
about 60 percent of psychologists admitted to “Collecting more data after
seeing whether results were significant,” whereas the prevalence estimate
derived from this admission estimate was 100 percent. Similarly, while ~30
percent admitted to “Failing to report all conditions” and ~40 percent admitted
to “Excluding data after the impact of doing so,” the estimated prevalence
rates in each case were ~70 percent and ~100 percent, respectively.
These figures needn’t imply dishonesty. Researchers may sincerely deny
p-hacking yet still do it unconsciously by failing to remember and document
all the analysis decisions made after inspecting data. Some psychologists
may even do it consciously but believe that such practices are acceptable
in the interests of data exploration and narrative exposition. Yet, regardless
of whether researchers are p-hacking consciously or unconsciously, the
solution is the same. The only way to verify that studies are not p-hacked
is to show that the authors planned their methods and analysis before
they analyzed the data—and the only way to prove that is through study
preregistration.
Unconscious Analytic “Tuning”
What do we mean exactly when we say p-hacking can happen unconsciously?
Statisticians Andrew Gelman and Eric Loken have suggested that
subtle forms of p-hacking and HARKing can join forces to produce false
discoveries.19 In many cases, they argue, researchers may behave completely
honestly, believing they are following best practice while still exploiting
researcher degrees of freedom.
To illustrate how, consider a scenario where a team of researchers design
an experiment to test the a priori hypothesis that listening to classical music
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
36 | Chapter 2
improves attention span. After consulting the literature they decide that a
visual search task provides an ideal way of measuring attention. The researchers
choose a version of this task in which participants view a screen
of objects and must search for a specific target object among distractors,
such as the letter “O” among many “Q”s. On each trial of the task, the participant
judges as quickly as possible whether the “O” is present or absent
by pressing a button—half the time the “O” is present and half the time it
is absent. To vary the need for attention, the researchers also manipulate
the number of distractors (Qs) between three conditions: 4 distractors
(low difficulty, i.e., the “O” pops out when it is present), 8 distractors (medium
difficulty) and 16 distractors (high difficulty). The key dependent
variables are the reaction times and error rates in judging whether the
letter “O” is present or absent. Most studies report reaction times as the
measure for this task and find that reaction times increase with the number
of distractors. Many studies also report error rates. The researchers
decide to adopt a repeated measures design in which each participant
performs this task twice, once while listening to classical music and once
while listening to nonclassical music (their control condition). They decide
to test 20 participants.
So far the researchers feel they have done everything right: they have a
prespecified hypothesis, a task selected with a clear rationale, and a sample
size that is on parity with previous studies on visual search. Once the study
is complete, the first signs of their analysis are encouraging: they successfully
replicate two effects that are typically observed in visual search tasks,
namely that participants are significantly slower and more error prone
under conditions with more distractors (16 is more difficult than 8 which,
in turn, is more difficult than 4), and that they are significantly slower to
judge when a target is absent compared to when it is present. So far so
good. On this basis, the authors judge that the task successfully measured
attention.
But then it gets trickier. The researchers find no statistically significant
main effect of classical music on either reaction times or error rates, which
does not allow them to reject the null hypothesis. However they do find a
significant interaction between the type of music (classical, nonclassical)
and the number of distractors (4, 8, 16) for error rates (p = .01) but not for
reaction times (p = .7). What this interaction means is that, for error rates,
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Hidden Flexibility | 37
the effect of classical music differed significantly between the different distractor
conditions. Post hoc comparisons show that error rates were significantly
reduced when participants were exposed to classical music compared
with control music, in displays with 16 distractors (p = .01) but not in displays
with 8 or 4 distractors (both p>.2). In a separate analysis they also
find that reaction times on target absent trials (i.e., no “O” present) are
significantly faster when exposed to classical music compared with control
music (p = .03). The same advantage isn’t significant for trials where the
“O” was present (p = .55) or when target-present and target-absent trials
are averaged together (as shown by the lack of a significant main effect).
The researchers think carefully about their results. After doing some
additional reading they learn that error rates can sometimes provide more
sensitive measures of attention than reaction times, which would explain
why classical music influenced only error rates. They also know that the
“target absent” condition is more difficult for participants and therefore
perhaps a more sensitive measure of attentional capacity—that would also
explain why classical music boosted performance on trials without targets.
Finally, they are pleased to see that the benefit of classical music on error
rates with 16 distractors goes in the direction predicted by their hypothesis.
Therefore, despite the fact that the main effect of classical music is not statistically
significant on either of the measures (reaction times or error rates),
the researchers conclude that the results support the hypothesis: classical
music improves visual attention, particularly under conditions of heightened
task difficulty. In the introduction of their article they phrase their hypothesis
as, “We predicted that classical music would improve visual search
performance. Since misidentification of visual stimuli can provide a more
sensitive measure of attention than reaction time (e.g., Smith 2000), such
effects may be expected to occur most clearly in error rates, especially under
conditions of heightened attentional load or task difficulty.” In the discussion
section of the paper, the authors note that their hypothesis was supported
and argue that their results conceptually replicate a previous study,
which showed that classical music can improve the ability to detect typographic
errors in printed text.
What, if anything, did the researchers do wrong in this scenario? Many
psychologists would argue that their behavior is impeccable. After all, they
didn’t engage in questionable practices such as adding participants until
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
38 | Chapter 2
statistical significance was obtained, selectively removing outliers, or testing
the effect of various covariates on statistical significance. Moreover,
they had an a priori hypothesis, which they tested, and they employed a
manipulation check to confirm that their visual search task measured attention.
However, as Gelman and Loken point out, the situation isn’t so
simple—researcher degrees of freedom have still crept insidiously into their
conclusions.
The first problem is the lack of precision in the researchers’ a priori hypothesis,
which doesn’t specify the dependent variable that should show the
effect of classical music (either reaction time or error rates, or both) and
doesn’t state under what conditions that hypothesis would or would not be
supported. By proposing such a vague hypothesis, the researchers have allowed
any one of many different outcomes to support their expectations,
ignoring the fact that doing so inflates the Type I error rate (α) and invites
confirmation bias. Put differently, although they have an a priori scientific
hypothesis, it is consistent with multiple statistical hypotheses.
The second problem is that the researchers ignore the lack of a statistically
significant main effect of the intervention on either of the measures;
instead they find a significant interaction between the main manipulation
(classical music vs. control music) and the number of distractors (4, 8, 16)—
but for error rates only. Since the researchers had no specific hypothesis
that the effect of classical music would increase at greater distractor set
sizes for error rates only, this result was unexpected. Yet by framing this
analysis as though it was a priori, the researchers tested more null hypotheses
than were implied in their original (vague) hypothesis, which in turn
increases the chances of a false positive.
Third, the researchers engage in HARKing. Even though their general
hypothesis was decided before conducting the study, it was then refined
based on the data and is presented in the introduction in its refined state.
This behavior is a subtle form of HARKing that conflates hypothesis testing
with post hoc explanation of unexpected results. Since the researchers did
have an a priori hypothesis (albeit vague) they would no doubt deny that
they engaged in HARKing. Yet even if blurred in their own recollections,
the fact is that they adjusted and refined their hypothesis to appear consistent
with unexpected results.
Finally, despite the fact that their findings are less definitive than advertised,
the researchers treat them as a conceptual replication of previous
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Hidden Flexibility | 39
work—a corroboration of the general idea that exposure to classical music
improves attention. Interestingly, it is in this final stage of the process where
the lack of precision in their original hypothesis is most clearly apparent.
This practice also highlights the inherent weakness of conceptual replication,
which risks constructing bodies of knowledge on weak units of
evidence.
Is this scenario dishonest? No. Is it fraudulent? No. But does it reflect
questionable research practices? Yes. Unconscious as it may be, the fact is
that the researchers in this scenario allowed imprecision and confirmation
bias to distort the scientific record. Even among honest scientists, researcher
degrees of freedom pose a serious threat to discovery.
Biased Debugging
Sometimes hidden flexibility can be so enmeshed with confirmation bias
that it becomes virtually invisible. In 2013, Mark Stokes from the University
of Oxford highlighted a situation where an analysis strategy that seems
completely sensible can lead to publication of false discoveries.20 Suppose
a researcher completes two experiments. Each experiment involves a different
method to provide convergent tests for an overarching theory. In each
case the data analysis is complicated and requires the researcher to write
an automated script. After checking through each of the two scripts for
obvious mistakes, the researcher runs the analyses. In one of the experiments
the data support the hypothesis indicated by the theory. In the second
experiment, however, the results are inconsistent. Puzzled, the researcher
studies the analysis script for the second experiment and discovers a subtle
but serious error. Upon correcting the error, the results of the second experiment
become consistent with the first experiment. The author is delighted
and concludes that the results of both experiments provide convergent
support for the theory in question.
What’s wrong with this scenario? Shouldn’t we applaud the researcher
for being careful? Yes and no. On the one hand, the researcher has caught
a genuine error and prevented publication of a false conclusion. But note
that the researcher didn’t bother to double-check the analysis of the first
experiment because the results in that case turned out as expected and—
perhaps more importantly—as desired. Only the second experiment at-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
40 | Chapter 2
tracted scrutiny because it ran counter to expectations, and running counter
to expectations was considered sufficient grounds to believe it was
erroneous. Stokes argues that this kind of results-led debugging—also
termed “selective scrutiny”—threatens to magnify false discoveries substantially,
especially if it occurs across an entire field of research.21 And since
researchers never report which parts of their code were debugged or not
(and rarely publish the code itself), biased debugging represents a particularly
insidious form of hidden analytic flexibility.
Are Research Psychologists Just Poorly Paid Lawyers?
The specter of bias and hidden analytic flexibility inevitably prompts us to
ask: what is the job of a scientist? Is it to accumulate evidence as dispassionately
as possible and decide on the weight of that evidence what conclusion
to draw? Or is it our responsibility to advocate a particular viewpoint,
seeking out evidence to support that view? One is the job of a scientist; the
other is the job of a lawyer. As psychologist John Johnson from Pennsylvania
State University said in a 2013 blog post at Psychology Today:
Scientists are not supposed to begin with the goal of convincing others
that a particular idea is true and then assemble as much evidence
as possible in favor of that idea. Scientists are supposed to be more
like detectives looking for clues to get to the bottom of what is actually
going on. They are supposed to be willing to follow the trail of
clues, wherever that may lead them. They are supposed to be interested
in searching for the critical data that will help decide what is
actually true, not just for data that supports a preconceived idea. Science
is supposed to be more like detective work than lawyering.22
Unfortunately, as we have seen so far, psychology falls short of meeting
this standard. Whether conscious or unconscious, the psychological community
tortures data until the numbers say what we want them to say—indeed
what many psychologists, deep down, would admit we need them to
say in the competition for high-impact publications, jobs, and grant funding.
This situation exposes a widening gulf between the needs of the scientists
and the needs of science. Until these needs are aligned in favor of science
and the public who fund it, the needs of scientists will always win—to our
own detriment and to that of future generations.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Hidden Flexibility | 41
Solutions to Hidden Flexibility
Hidden flexibility is a problem for any science where the act of discovery
involves the accumulation of evidence. This process is less certain in some
sciences than in others. If your evidence is clearly one way or the other,
such as the discovery of a new fossil or galaxy, then inferences from statistics
may be unnecessary. In such cases no analytic flexibility is required,
whether hidden or disclosed—so none takes place. It is perhaps this association
between statistical analysis and noisy evidence that prompted physicist
Ernest Rutherford to allegedly once remark: “If your experiment needs statistics,
you ought to have done a better experiment.”
In many life sciences, including psychology, discovery isn’t a black-andwhite
issue; it is matter of determining, from one experiment to the next,
the theoretical contribution made by various shades of gray. When psychologists
set arbitrary criteria (p<.05) on the precise shade of gray required to
achieve publication—and hence career success—they also incentivize a host
of conscious and unconscious strategies to cross that threshold. In the battle
between science and storytelling, there is simply no competition: storytelling
wins every time.
How can we get out of this mess? Chapter 8 will outline a manifesto for
reform, many aspects of which are already being adopted. The remainder
of this chapter will summarize some of the methods we can use to counter
hidden flexibility.
Preregistration. The most thorough solution to p-hacking and other
forms of hidden flexibility (including HARKing) is to prespecify our hypotheses
and primary analysis strategies before examining data. Preregistration
ensures that readers can distinguish the strategies that were independent
of the data from those that were (or might have been) data led.
This is not to suggest that data-led strategies are necessarily incorrect or
misleading. Some of the most remarkable advances in science have
emerged from exploration, and there is nothing inherently wrong with
analytic flexibility. The problems arise when that flexibility is hidden from
the reader, and possibly from the researcher’s own awareness. By unmasking
this process, preregistration protects the outcome of science from our
own biases as human practitioners.
Study preregistration has now been standard practice in clinical medicine
for decades, driven by concerns over the effects of hidden flexibility
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
42 | Chapter 2
and publication bias on public health. For basic science, these risks may be
less immediate but they are no less serious. Hidden flexibility distorts the
scientific record, and, since basic research influences and feeds into more
applied areas (including clinical science), corruption of basic literature necessarily
threatens any forward applications of discovery.
Recent years have witnessed a concerted push to highlight the benefits
of preregistration in psychology. An increasing number of journals are now
offering preregistered article formats in which part of the peer review process
happens before researchers conduct experiments. This type of review
ensures adherence to the hypothetico-deductive model of the scientific
method, and it also prevents publication bias. Resources such as the Open
Science Framework also provide the means for researchers to preregister
their study protocols.
p-curve. The p-curve tool developed by Simonsohn and colleagues is useful
for estimating the prevalence of p-hacking in published literature. It
achieves this by assuming that collections of studies dominated by
p-hacking will exhibit a concentration of p values that peaks just below
.05. In contrast, an evidence base containing positive results in which
p-hacking is rare or absent will produce a positively skewed distribution
where smaller p values are more common than larger ones. Although this
tool cannot diagnose p-hacking within individual studies, it can tell us
which fields within psychology suffer more from hidden flexibility. Having
identified them, the community can then take appropriate corrective
action such as an intensive program of direct replication.
Disclosure statements. In 2012, Joe Simmons and colleagues proposed a
straightforward way to combat hidden flexibility: simply ask the researchers.23
This approach assumes that most researchers (a) are inherently honest
and will not deliberately lie, and (b) recognize that exploiting researcher
degrees of freedom reduces the credibility of their own research.
Therefore, requiring researchers to state whether or not they engaged in
questionable practices should act as a disincentive to p-hacking, except for
researchers who are either willing to admit to doing it (potentially suffering
loss of academic reputation) or are willing to lie (active fraud).
The disclosure statements suggested by the Simmons team would require
authors to state in the methods section of submitted manuscripts how they
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Hidden Flexibility | 43
made certain decisions about the study design and analysis. These include
whether the sample size was determined before the study began and if any
of experimental conditions or data were excluded. Their “21 word solution”
(as they call it) is:
We report how we determined our sample size, all data exclusions (if
any), all manipulations, and all measures in the study.
Although elegant in their simplicity, disclosure statements have several
limitations. They can’t catch forms of p-hacking that exploit defensible ambiguities
in analysis decisions, such as testing the effect of adding a covariate
to the design or focusing the analysis on a particular subgroup (e.g., males
only) following exploration. Furthermore, the greater transparency of methods,
while laudable in its own right, doesn’t stop the practice of HARKing;
note that researchers are not asked whether their a priori hypotheses were
altered after inspecting the data. Finally, disclosure statements cannot catch
unconscious exploitation of researcher degrees of freedom, such as forgetting
the full range of analyses undertaken or more subtle forms of HARKing
(as proposed by Gelman and Loken). Notwithstanding these limitations,
disclosure statements are a worthy addition to the range of tools for counteracting
hidden analytic flexibility.
Data sharing. The general lack of data transparency is a major concern in
psychology, and will be discussed in detail in chapter 4. For now, it is sufficient
to note that data sharing provides a natural antidote to some forms
of p-hacking—particularly those where statistical significance is based on
post hoc analytic decisions such as different methods for excluding outliers
or focusing on specific subgroups. Publishing the raw data allows independent
scientists with no vested interest to test how robust the outcome
is to alternative analysis pathways. If such examinations were to reveal
that the authors’ published approach was the only one out of a much larger
subset to produce p<.05, the community would be justifiably skeptical of
the study’s conclusions. Even though relatively few scientists scrutinize
each other’s raw data, the mere possibility that this could happen should
act as a natural deterrent to deliberate p-hacking.
Solutions to allow “optional stopping.” Leslie John’s survey showed how
a major source of p-hacking is violation of stopping rules; that is, continu-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
44 | Chapter 2
ously adding participants to an experiment until the p value drops below
the significance threshold. When H0 is true, p values between 0 and 1 are
all equally likely to occur, therefore with the addition of enough participants
a p value below .05 will eventually be found by chance. Psychologists
often neglect stopping rules because, in most studies, there is no strong
motivation for selecting a particular sample size in the first place.
Fixed stopping rules present a problem for psychology because the size
of the effect being investigated is often small and poorly defined. Fortunately,
there are two solutions that psychologists can use to avoid violating
stopping rules. The first, highlighted by Michael Strube from Washington
University (and more recently by Daniël Lakens), allows researchers to use
a variable stopping rule with NHST by lowering the α level in accordance
with how regularly the researcher peeks at the results.24 This correction is
similar to more conventional corrections for multiple comparisons. The
second approach is to adopt Bayesian hypothesis testing in place of NHST.25
Unlike NHST, which estimates the probability of observed (or more extreme)
data under the null hypothesis, Bayesian tests estimate the relative
probabilities of the data under competing hypotheses. In chapter 3 we will
see how this approach has numerous advantages over more conventional
NHST methods, including the ability to directly estimate the probability of
H0 relative to H1. Moreover, Bayesian tests allow researchers to sequentially
add participants to an experiment until the weight of evidence supports
either H0 or H1. According to Bayesian statistical philosophy “there is nothing
wrong with gathering more data, examining these data, and then deciding
whether or not to stop collecting new data—no special corrections are
needed.”26 This flexibility is afforded by the likelihood principle, which
holds that “[w]ithin the framework of a statistical model, all of the information
which the data provide concerning the relative merits of two hypotheses
is contained in the likelihood ratio of those hypotheses.”27 In other
words, the more data you have, the more accurately you can conclude support
for one hypothesis over another.
Standardization of research practices. One reason p-hacking is so common
is that arbitrary decisions are easy to justify. Even within a simple
experiment, there are dozens of different analytic pathways that a researcher
can take, all of which may have a precedent in the published literature
and all of which may be considered defensible. This ambiguity
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Hidden Flexibility | 45
enables researchers to pick and choose the most desirable outcome from a
smorgasbord of statistical tests, reporting the one that “worked” as though
it was the only analysis that was attempted. One solution to this problem
is to apply more constraints on the range of acceptable approaches, applying
institutional standards to such practices as outlier exclusion or the use
of covariates.
Scientists are generally resistant to such changes, particularly when a
clear best practice fails to stand out from the crowd. Such standards can
also be difficult to apply in emerging fields, such as functional brain imaging,
owing to the rapid developments in methodology and analysis strategies.
However, where standardization isn’t possible and any one of several
arbitrary decisions is possible, there is a strong argument that researchers
should report all of them and then summarize the robustness of the outcomes
across all contingencies.
Moving beyond the moral argument. Is p-hacking a form of fraud?
Whether questionable research practices such as p-hacking and HARKing
are fraudulent—or even on the same continuum as fraud—is controversial.
Psychologist Dave Nussbaum from the University of Chicago has argued
that clearly fraudulent behavior, such as data fabrication, is categorically
different from questionable practices such as p-hacking because, with
p-hacking, the intent of the researcher cannot be known.28 Nussbaum is
right. As we have seen in this chapter, many cases of p-hacking and HARKing
are likely to be unconscious acts of self-deception. Yet in cases where
researchers deliberately p-hack in order to achieve statistical significance,
Nussbaum agrees that such behavior is on the same continuum of extreme
fraud.
We will return to the issue of fraud, but for now we can ask: does it matter
if p-hacking is fraudulent? Whether the sin of hidden flexibility is deliberate
or an act of self-deception is a distraction from the goal of reforming
science. Regardless of what lies in the minds of researchers, the effects of
p-hacking are clear, populating the literature with post hoc hypotheses,
false discoveries, and blind alleys. The solutions are also the same.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
C H A P T E R 3
The Sin of Unreliability
And it’s this type of integrity, this kind of care not to fool
yourself, that is missing to a large extent in much of the
research in cargo cult science.
—Richard Feynman, 1974
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Particles break light-speed limit,” announced Nature News.1 “Faster than
light particles threaten Einstein,” declared Reuters.2 “Was Einstein
wrong?” asked Time.3 So read the headlines in September 2011 when a
team of physicists published evidence suggesting that subatomic particles
called neutrinos could travel faster than the speed of light. If true, this discovery
would revolutionize modern physics. Teams of scientists immediately
began the task of repeating the experiment. By June 2012, three independent
groups had failed to replicate the original result: the neutrinos in
their experiments traveled at approximately the speed of light, just as predicted
by special relativity. One month later, the original team reported
that their findings were caused by a loose fiber-optic cable.
The case of faster-than-light neutrinos may sound like an example of science
going awry, but, in fact, it is just the opposite. After one team of scientists
made what looked like an extraordinary discovery, the scientific
community responded by attempting to reproduce the result. When those
attempts failed, the original team scrutinized their experiment more closely,
discovering a technical error that explained the anomaly. Science can never
escape the risk of human error, but it can and must ensure that it selfcorrects.
Imagine for a moment what kind of physics we would have if
faster-than-light neutrinos had simply been believed, their status as a discovery
left unreplicated and unchallenged.
Replication is the immune system of science, identifying false discoveries
by testing whether other scientists can repeat them. Without replication,
we have no way of knowing which discoveries are genuine and which are
caused by technical error, researcher bias, fraud, or the play of chance. And
if we don’t know which results are reliable, how can we generate meaningful
theories?
Unfortunately, as we saw earlier, the process of replication—so intrinsic
to the scientific method—is largely ignored or distorted in psychology. Recall
from chapter 1 that the claims of psychic precognition by Daryl Bem
soared into print at the one of the most prestigious psychology journals in
the world. Yet the crucial nonreplication by Stuart Richie and Chris French
took much longer to appear and was initially rejected, on principle alone,
by the same journal that published Bem’s findings. Instead of valuing the
reproducibility of results, psychology has embraced a tabloid culture where
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
48 | Chapter 3
novelty and interest-value are paramount and the truth is left begging. Psychology
thus succumbs to our third major transgression: the sin of
unreliability.
Sources of Unreliability in Psychology
Science generates an understanding of the natural world by using empirical
evidence to reduce uncertainty. The hypothetico-deductive model of the
scientific method, introduced in chapter 1, achieves this by systematically
testing hypotheses borne from theory. Once the results of high-quality experiments
are verified through direct (close) replication, the evidence can
refine the theory in question, generating further hypotheses and honing
our ability to understand and predict reality (see figure 3.1).
Unfortunately, psychology fails to adhere to this philosophy, and nowhere
is this culture expressed more blatantly than by the indifference—
and, in many cases, hostility—toward direct replication. Yet the lack of
replication isn’t the only reason psychological science faces a crisis of reliability.
As we will see, other concerns include low statistical power, failure
to disclose full methodological details, statistical fallacies, and the refusal
to retract irreproducible findings from the literature. Together these problems
not only threaten the truth-value of psychological evidence; they
threaten the status of psychology as a science.
Reason 1: Disregard for Direct Replication
Direct replication is intrinsic to all sciences. During a direct replication, a
researcher seeks to the test the repeatability of a previous finding by duplicating
the methodology as exactly as possible.4 The importance of direct
replication in science is reflected by the simple fact that empirical journal
articles in all sciences include method sections. A method is supposed to
provide researchers with all the information they would need to replicate
the experimental “recipe.”
Despite the clear importance of replication, we saw in chapter 1 how the
academic culture in psychology places little emphasis in repeating the experimental
methods of other psychologists. Such work is seen to lack in-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Unreliability | 49
novation within a system that instead seeks to validate previous findings by
conducting novel experiments that test a related (but different) idea using
a different method: the approach referred to as “conceptual replication.”
Although used widely in psychology, the term conceptual replication does
not feature in the scientific method of other disciplines. In fact, the term
itself is misleading because conceptual replications don’t actually replicate
previous experiments; they instead assume (rather than test) the truth of
the findings revealed in those experiments, infer the underlying cause, and
then seek converging evidence for that inferred cause using an entirely different
experimental procedure. Viewed within the framework of the H-D
scientific method, this process can be thought of as extrapolating from a
body of findings to refine theory and generate new hypotheses (see figure
3.1). As important as this step is, it depends first and foremost on the reliability
of the underlying evidence base. To rely solely on extrapolation at
the expense of direct replication is to build a house on sand.
Direct
replication
Replicable?
Experiment
Hypothesis
generation
Yes
No
Direct
replication
Replicable?
Experiment
Hypothesis
generation
Yes
No
Direct
replication
Replicable?
Experiment
Hypothesis
generation
Yes
NoDirect
replication
Replicable?
Experiment
Hypothesis
generation
Yes
No
Increasing knowledge
Scientific
law
Theoretical
refinement
Theory
Figure 3.1. Science usually advances in steady increments rather than leaps. According to
the deductive scientific method, a hypothesis is generated from current theory, and an
experiment is designed to test the hypothesis against one or more competitors. If the outcome
of the experiment can be directly replicated to the community’s satisfaction, then
the theory is refined, and a new hypothesis is formulated. The endpoint of this process of
knowledge accumulation is the generation of scientific laws. Note that the refining of
theoretical precision in no way implies that the original theoretical model remains intact
or must be in any way sustained: the increase of knowledge can, and does, lead to theories
being discarded altogether. Regardless, the theoretical framework always becomes more
precise with the addition of new evidence.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
50 | Chapter 3
If direct replications are considered trivial and uninteresting then they
should be seldom seen in the published literature. In his famous 1974 lecture
“Cargo Cult Science,” Richard Feynman recounts his experience of suggesting
to a psychology student that she perform a direct replication of a previous
experiment before attempting a novel one:
She was very delighted with this new idea, and went to her professor.
And his reply was, no, you cannot do that, because the experiment
has already been done and you would be wasting time. This was in
about 1947 or so, and it seems to have been the general policy then to
not try to repeat psychological experiments, but only to change the
conditions and see what happened.5
Remarkably, since 1947 there has been little analysis of how frequently
psychologists replicate each other’s work even partially, let alone directly.
In 2012, Matthew Makel, Jonathan Plucker, and Boyd Hegarty conducted
the first systematic investigation of replication rates in psychology.6 They
searched the top 100 psychology journals from 1900 to 2012 and found that
only 1.57 percent of 321,411 articles mentioned a word beginning with “replicat*.”
This may sound low enough but was already optimistically high.
Within a randomly selected subsample of 500 of the 1.57 percent, only 342
actually reported some form of replication—and, of these, just 62 articles
reported a direct replication of a previous experiment. On top of that, only
47 percent of replications within the subsample were produced by independent
researchers. The implications of the Makel study are sobering: for
every 1,000 papers published in psychology, only two will seek to directly
replicate a previous experiment, and just one of those will be published by
a different team to the original study.
What happens to a scientific discipline when it abandons direct replication,
as psychology has done? The immediate consequence, estimated by
medical researcher John Ioannidis, is that up to 98 percent of published
findings may be either unconfirmed genuine discoveries or unchallenged
fallacies.7 Failure to attempt replications thus sabotages the ability for psychological
science to self-correct: where physicists succeeded (quickly) at
refuting and explaining the observation of faster-than-light neutrinos, psychologists
would fail by not trying in the first place. This contempt for verification,
in turn, has a profound effect on theory development, condemning
the field to the publication of theoretical frameworks that can be neither
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Unreliability | 51
confirmed nor falsified. In the words of psychologists Chris Ferguson and
Moritz Heene, the outcome is a discipline that plays host to a “graveyard of
undead theories,” each of them as immovable as they are impotent.8
Why is psychology so implacably opposed to direct replication? The short
answer is that we don’t know for certain—the aversion is so entrenched that
the causes and effects have become obscured. The longer answer, as discussed
in chapter 1, is that together with other life sciences, psychology has
evolved an incentive structure that rewards empiricists who can produce
novel, positive, eye-catching results that confirm the hypothesis and offer
pithy interpretations. Pressure not to replicate is applied from all directions.
At the supply end, funding agencies are loath to award money for merely
repeating previous research—in the UK, even grant applications that describe
original research (let alone replications) are often rejected for lack
of novelty and innovation. Meanwhile, at the demand end, it is a struggle
to publish direct replications in respected journals, whether successful or
not. When such attempts succeed they are generally seen as boring and
contributing little of value (“we already knew this.”; “what does this add?”),
despite the fact that replications frame the certainty we can justify in prior
discoveries. And when replications fail, defenders of the original work are
liable to try and block publication or respond aggressively when such work
is published.
The negative attitude toward replication in psychology is epitomized by
an incident that became known rather infamously as “Repligate.” In May
2014, the journal Social Psychology bucked the academic trend and reported
an ambitious initiative to reproduce a series of influential psychological
discoveries claimed since the 1950s.9 Many of the findings could not be
replicated, and in most cases these nonreplications were met with cordial
interactions between researchers. For instance, confronted with the news
that one of his prior findings in social priming could not be reproduced, Dr.
Eugene Caruso from the University of Chicago said, “This was certainly
disappointing at a personal level. But when I take a broader perspective, it’s
apparent that we can always learn something from a carefully designed and
executed study.”
Not all researchers were so gracious. Dr. Simone Schnall from the University
of Cambridge argued that her work on social priming was treated
unfairly and “defamed.” In a remarkable public statement, Schnall claimed
that she was bullied by the researchers who sought (unsuccessfully) to rep-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
52 | Chapter 3
licate her findings and that the journal editors who agreed to publish the
failed replications of her work behaved unethically.10 She wrote, “I feel like
a criminal suspect who has no right to a defence and there is no way to win:
The accusations that come with a “failed” replication can do great damage
to my reputation, but if I challenge the findings I come across as a ‘sore
loser.’”
Schnall’s strong reaction to the failed replication of her own work provoked
a mixed reaction from the psychological community. While many
psychologists were bewildered by her response, a number of prominent US
psychologists voiced support for her position. Dan Gilbert from Harvard
University likened Schnall’s battle to the plight of Rosa Parks,11 and he referred
to some psychologists who conducted or supported replications as
“bullies,” “replication police,” “second stringers,” McCarthyists, and “god’s
chosen soldiers in a great jihad.”12 Others accused the so-called replicators
of being “Nazis,” “fascists,” and “mafia.” Rather than viewing replication as
an intrinsic part of best scientific practice, Gilbert and his supporters
framed it as a threat to the reputation of the (presumably brilliant) researchers
who publish irreproducible findings, stifling their creativity and
innovation.13
For some psychologists, the reputational damage in such cases is grave—
so grave that they believe we should limit the freedom of researchers to
pursue replications. In the wake of Repligate, Nobel laureate Daniel Kahneman
called for a new rule in which replication attempts should be “prohibited”
unless the researchers conducting the replication consult beforehand
with the authors of the original work.14 Kahneman said, “Authors, whose
work and reputation are at stake, should have the right to participate as
advisers in the replication of their research.” Why? Because the method
sections published by psychology journals are too vague to provide a recipe
that can be repeated by others. Kahneman argued that successfully reproducing
original effects could depend on seemingly irrelevant factors—hidden
secrets that only the original authors would know. “For example, experimental
instructions are commonly paraphrased in the methods section,
although their wording and even the font in which they are printed are
known to be significant.”
For many psychologists, Kahneman’s cure is worse than the disease.
Andrew Wilson from Leeds Metropolitan University immediately rejected
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Unreliability | 53
the suggestion of new regulations, writing: “If you can’t stand the replication
heat, get out of the empirical kitchen because publishing your work
means you think it’s ready for prime time, and if other people can’t make it
work based on your published methods then that’s your problem and not
theirs.”15 Are replications in other sciences ever regarded as an act of aggression,
as Schnall and others suggest? Jon Butterworth, head of the Department
of Physics and Astronomy at University College London, finds this
view of replication completely alien. “Thinking someone’s result is interesting
and important enough to be checked is more like flattery,” he told me.
For Butterworth there is no question that the published methods of a scientific
paper should be sufficient for trained specialists in the field to repeat
experiments, without the need to learn unpublished secrets from the original
authors. “Certainly no physicist I know would dare claim their result
depended on hidden ‘craft.’”16
Helen Czerski, broadcaster, physicist and oceanographer at University
College London, offers a similar perspective. In her field, contacting the
authors of a paper to find out how to replicate their results would be seen
as odd. “You might have a chat with them along the way about the difficulties
encountered and the issues associated with that research,” she said to
me, “but you certainly wouldn’t ask them what secret method they used to
come up with the results.” Czerski questions whether Kahneman’s proposed
rules may breach research ethics. “My gut response is that asking that question
is close to scientific misconduct, if you were asking solely for the purpose
of increasing your chances of replication, rather than to learn more
about the experimental issues associated with that test. The attitude in the
fields I’ve worked in is definitely that you should be able to conduct your
own test of a hypothesis, and that you shouldn’t need any guidance from
the original authors.”17
One thing seems clear—the culture of replication in the physical sciences
is a world apart from the one that prevails in psychology. Katie Mack, astrophysicist
at the University of Melbourne, told me that in her field there
are many situations where reproducing a result is considered essential for
moving the area forward. “A major result produced by only one group, or
with only one instrument, is rarely taken as definitive.” Mack points out
that even findings that have been replicated many times over are valued,
such as the Hubble constant, which describes the rate of expansion of the
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
54 | Chapter 3
universe. “Many groups have measured it with many different methods (or
in some cases the same method), and each new result is considered noteworthy
and definitely publishable.”18
New rules and constraints on replications—especially when issued by
heavyweights such as Kahneman—are likely to discourage academic journals
from publishing them, and in this respect journals hardly need further
discouragement. With rare exceptions, none of the most prominent
journals in psychology or neuroscience regularly publish direct replications,
and even the megajournal PLOS ONE—one of the few outlets to
explicitly disavow the traditional focus on novelty—warns authors that
submissions that “replicate or are derivative of existing work will likely be
rejected if authors do not provide adequate justification.”19 Populating the
literature with false or unconfirmed discoveries is thus deemed acceptable
while verifying the replicability of those discoveries demands special
explanation.
In contrast to physicists, many senior psychologists seem content with a
lack of direct replication. In 2014, Jason Mitchell from Harvard University
argued that “hand-wringing over failed replications in social psychology is
largely pointless, because unsuccessful experiments have no meaningful
scientific value.”20 Mitchell’s central thesis was that a failed replication is
most likely the result of human error on the part of the researcher attempting
the replication rather than the unreliability of the phenomenon originally
reported. But in a lapse of logic, Mitchell ignored the possibility that
the very same human error can give rise to false discoveries.21
Psychologists Wolfgang Stroebe and Fritz Strack have also published a
provocative defense of the status quo.22 Like Kahneman, they argue that in
at least some areas of psychology, direct replications are not possible because
the original results depend on a host of hidden methodological variables
that are just as invisible to the original researchers as to those attempting
the replication. Therefore, they argue, when a direct replication fails it
is most likely due to the dependence of the effect on hidden moderators
rather than the unreliability of the original finding. Instead of defending
direct replication, Stroebe and Strack take the view that conceptual replications
are more informative because they test the “underlying theoretical
construct.” They conclude that direct replications are uninteresting regardless
of the outcome because they bring us no closer to knowing whether the
original study was “a good test of the theory.”
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Unreliability | 55
Stroebe and Strack’s article triggered a trenchant rebuke from psychologist
Dan Simons. Simons argues that this characterization of psychology
renders it blatantly unscientific because if the failure of a direct replication
can always be explained by hidden variables (or moderators) then the original
result, by definition, can never be falsified. In such a world there can
be no false discoveries. The task of ensuring reproducibility thus becomes
impossible because, as Simons puts it:
the number of possible moderators is infinite: perhaps the effect depends
on the phases of the moon, perhaps it only works at a particular
longitude and latitude, perhaps it requires subjects who ate a lot
of corn as children. . . . We cannot accumulate evidence for the reliability
of any effect. Instead, all findings, both positive and negative,
can be attributed to moderators unless proven otherwise. And we
can never prove otherwise.23
In addition, Strack and Stroebe’s attack on direct replication can be leveled
equally at their preferred alternative of conceptual replication. Whether
it succeeds or fails, a conceptual replication could also arise because of the
action of hidden moderators or random causes; indeed the fact that a conceptual
replication adopts a different methodology can only increase its
vulnerability to such confounds. Hidden moderators or (as yet) unexplained
causal factors are no doubt a reality in much psychological research, as in
other sciences. The rational solution to such complexity is a program of
comprehensive direct replication followed by cautious, incremental advances.
But discounting direct replication because of the hypothetical actions
of unobserved (and unobservable) moderators is tantamount to an
argument for magic.
Reason 2: Lack of Power
In chapter 2 we saw how hidden analytic flexibility corrupts science by elevating
the rate of false positives. As shown by Joe Simmons and colleagues,
exploiting researcher degrees of freedom can have a profound effect on the
α level—the probability of erroneously rejecting a true null hypothesis—increasing
it from a nominal .05 to a startling .60 or higher. These questionable
practices are certainly a major source of unreliable findings in psychol-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
56 | Chapter 3
ogy. However, an additional and frequently overlooked source is the
prospect of experiments missing true discoveries. Under the logic of NHST,
this is referred to as β: the probability of failing to reject a false null hypothesis.
To draw a courtroom analogy, α can be seen as the probability of convicting
an innocent defendant, while β is the probability of acquitting a
guilty one. The probability of correctly rejecting a false null hypothesis—
that is, of correctly convicting a guilty perpetrator—is thus calculated as
1−β. This value is referred to as statistical power and tells us the probability
that a statistical test will detect an effect of a given size, where that effect
truly exists.
Since the 1960s, researchers have known that psychology suffers from
low statistical power. Psychologist Jacob Cohen was among the first to draw
the issue of power and β into prominence. Cohen surveyed all articles that
appeared in 1960 across three issues of the Journal of Abnormal and Social
Psychology. He found that, on average, they had only a 48 percent chance
of detecting effects of medium size and only an 18 percent chance of detecting
even smaller effects.24 Even for large effect sizes, Cohen found that the
studies in his cohort bore a 17 percent chance of missing them.
Cohen’s analysis of statistical power was groundbreaking and led to a
chain of investigations into power throughout the 1960s and 1970s. Even
so, these studies had little influence on research practices. When psychologists
Peter Sedlmeier and Gerd Gigerenzer returned to the question in 1989
they found that the average statistical power in psychology had scarcely
changed between 1960 and 1984.25 In 2001, Scott Bezeau and Roger Graves
took yet another look, repeating the analysis on 66 studies published in the
field of clinical neuropsychology between 1998 and 1999. Again, they confirmed
an overall power to detect medium effect sizes of about 0.50—virtually
identical to previous investigations.26 For four decades, psychological
research has remained steadfastly underpowered, despite widespread
awareness of the problem. Most recently, in 2013, Kate Button and colleagues
from the University of Bristol extended these findings to neuroscience,
uncovering an even lower median power of 0.21.27
Why do psychologists neglect power? Psychologist Klaus Fiedler and colleagues
have argued that the prevalence of underpowered studies follows
from a cultural fixation on α at the expense of β. Doing so, they argue, leads
to high rates of false negatives—that is, missed true discoveries—that can
cause lasting damage to theory generation. While false positives at least
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Unreliability | 57
have the potential to be disconfirmed by additional research (albeit minimally,
owing to low rates of replication), the unrelenting pressure for researchers
to produce publishable positive results means that failed hypotheses
are likely to be swiftly discarded and forgotten. Fiedler and colleagues
warn that when a correct hypothesis is wrongly rejected and abandoned
because of a false negative then “even the strictest tests of the remaining
hypotheses can only create an illusion of validity.”28
Not surprisingly, a common finding from the studies of Cohen onward
is that very few published studies in psychology determine sample sizes
through a priori (prospective) power analysis. Instead, sample sizes tend to
be decided by a rule of thumb in which the sample size is roughly matched
to previous experiments that succeeded in obtaining statistically significant
effects. The problem with this approach is that it ignores the statistical
properties of replication: when a study detects a true positive with a p value
just below .05 (e.g., p = .049) then an exact replication with the same sample
size (assuming the same effect size) has only about a 50 percent chance of
successfully repeating the discovery.29 Increasing the power to 0.8 or higher
generally requires researchers to at least double the original sample size.
Given the prevalence of “just significant” results in psychology (see chapter
2), when researchers determine sample sizes based on “what worked last
time” then the overall statistical power of a field will converge on 0.5 or
less. So it is no coincidence that Cohen and others have consistently estimated
the power of psychological research to depend on the flip of a coin.
Valuing our instincts in sample size selection over and above formal a priori
power analysis places a glass ceiling on the statistical sensitivity of any scientific
endeavor.
Another reason for the neglect of statistical power in psychology is the
almost complete absence of direct replication. Because no two studies employ
exactly the same design, this allows researchers to shrug off concerns
about power on the grounds that nothing exactly the same has ever been
done before, so there is nothing comparable in the literature on which to
base a power analysis. This argument is, of course, fallacious for a number
of reasons. First, even conceptual replications often apply only small tweaks
to previous experimental methods; therefore an estimate of the expected
effect size (and hence power) can be gleaned, especially when looking over
a body of previous similar work. Second, even in rare cases where power
cannot be estimated from previous work, researchers can at least motivate
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
58 | Chapter 3
the sample size to detect an effect of a particular size (e.g., small, medium,
or large effects, as defined by Cohen).30
It may seem obvious that low power reduces the reliability of psychological
research by increasing the rate of false negatives (β), but what about the
risk of false positives (α)? Because underpowered studies lack sensitivity to
detect true effects, any effects they do reveal must be large in order to
achieve statistical significance. Therefore, you might think that a positive
result (p<.05) from a low-powered study would actually be more convincing
than the equivalent finding from a high-powered study—after all, isn’t the
bar for detecting a true discovery in a low-powered study necessarily higher?
While it is the case that the effect size required to achieve statistical
significance is greater for low-powered experiments, this doesn’t mean that
the probability of those significant results being true is necessarily higher.
Here we must be careful not to confuse the probability of the data under
the null hypothesis (p) with the probability that an obtained positive result
is a true positive. The probability of a true positive cannot be inferred directly
from the p value of an experiment; it must instead be estimated
through the positive predictive value (PPV). If you imagine traversing a desert
in search of water, the PPV can be thought of as the chance that a shimmer
on the horizon is an oasis rather than a mirage. Mathematically it is
calculated as the number of true positives (oases) divided by the total number
of all positive observations (shimmers), both true (oases) and false
(mirages):
Positive predictive value PPV =
Total number of positive observations
Number of true positives
To see how the PPV relates to power, we need to substitute these quantities
for probabilities. In the numerator, the number of true positives can be
replaced by the statistical power of the experiment (1−β) multiplied by the
prior probability (R) that the effect being studied is truly positive:
PPV =
Total number of positive observations
(1 – β) × R
Then, since the total number of obtained positive results is the sum of
all true positives and false positives, we can replace the denominator with
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Unreliability | 59
the probability of obtaining a true positive, (1−β) × R, added to the probability
of obtaining a false positive (α):
Positive predictive value PPV =
(1 – β) × R + α
(1 – β) × R
As we can see in figure 3.2, the PPV is therefore directly related to statistical
power: as power increases, so does the chance that a statistically
significant effect is a true positive. For example, if we assume a relatively
low statistical power of 0.15, combined with the prior probability that the
null hypothesis is false of 0.20 (R), then the probability of a positive result
being a true positive is only 0.375 (PPV). Increasing the statistical power
from 0.15 to 0.95 increases the PPV from 0.375 to 0.79. Therefore, high
Positivepredictivevalue
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.0
Statistical power
0.80.60.4 1.00.20.0
Figure 3.2. The statistical power of an experiment tells us its positive predictive value
(PPV): the probability that a statistically significant effect is a true positive. This curve
shows the relationship between power and PPV for a hypothetical experiment where the
prior probability of the null hypothesis being false is 0.2 (i.e., a reasonably unlikely hypothesis)
and α =.05. When power is just 0.15 (lower-left dotted line), then the PPV is also
just .375. But when power is increased to .95 (upper-right dotted line), the PPV rises to
0.79. This curve plateaus at 0.8; to achieve a higher PPV, the researcher would either need
to lower the α or test a more plausible H1.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
60 | Chapter 3
power not only helps us limit the rate of false negatives, but it also caps the
rate of false positives—just as a powerful experiment increases the chances
of finding an oasis, it also reduces the chances of leading us toward a
mirage.
As well as failing to recognize the importance of statistical power, psychologists
routinely underestimate how complexity in experimental designs
reduces power. Psychologist Scott Maxwell from the University of Notre
Dame has argued that failure to appreciate the price of complexity could be
a major cause of the historically low rates of power in psychology. Even in
simple factorial designs, power losses quickly accumulate. Suppose, for instance,
you wanted to know the effect of a new cognitive training intervention
on weight loss in men and women. To address this question you set up
a 2 × 2 factorial design. One factor is the type of training: whether participants
receive the new intervention or a control condition (i.e., a placebo).
The other factor is the gender of the participants (male or female). Each
male and female participant is randomly assigned to one of the treatment
groups. Let’s further suppose you have key research questions about each
of the factors. First, you want to know whether your new intervention
works better than the control intervention. This will be the main effect of
training type, collapsed across gender. Second, you are interested in
whether men or women are generally more successful at losing weight. This
will be the main effect of gender, collapsed across training type. And finally,
you want to know whether training intervention is more effective in one
gender than the other. This will be the interaction of gender × training
type. Based on similar studies in this area, you settle on a sample size of
120 participants, with 30 participants in each of the four groups. What
would your power be to detect a medium-sized effect for each of the main
effects and the interaction? Maxwell calculated it to be just 0.47, thus on
more than 50 percent of such experiments at least one of your tests would
return a false negative.31
In 2002, Rachel Smith and colleagues from the University of Michigan
reported a series of computer simulations to test the chances of obtaining
β errors in experimental designs that have even more factors and explore
more complex interactions. What they found was disturbing—at effect sizes
and sample sizes commonly observed in psychological research, the rate of
false negative conclusions was as high as 84 percent.32 Their message was
clear: researchers ignore power at their peril.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Unreliability | 61
Reason 3: Failure to Disclose Methods
The main purpose of including method sections in published research articles
is to provide readers with enough information about the design and
analysis to be able to replicate the experiment. A critical reader of any
method section should be asking not only whether the reported procedure
is sound but also whether it provides sufficient details to be repeatable.
Unfortunately, an additional source of unreliability in psychology lies in the
systematic failure of studies to disclose sufficient methodological detail to
allow exact replication.
Evidence of a systematic lack of disclosure was first uncovered in the
survey of 2,000 psychologists by Leslie John and colleagues reported in
chapter 2. Based on self-reports, John and colleagues estimated that more
than 70 percent of psychologists have failed to report all experimental conditions
in a paper, and that virtually all psychologists have failed to fully
disclose the experimental measures in a study. In 2013, a team of psychologists
led by Etienne LeBel from the University of Western Ontario sought to
determine the prevalence of such practices. LeBel and colleagues contacted
a random 50 percent of authors who had published in four of the most prestigious
psychology journals throughout 2012 and asked whether they had
fully disclosed the exclusion of observations, experimental conditions, experimental
measures, and the method used to determine final sample size.33
Of 347 authors contacted, 161 replied—and of these, 11 percent reported
that they had failed to report the exclusion of observations or experimental
conditions. But even more startling was that 45 percent of respondents admitted
failing to disclose all the experimental measures they acquired. The
most popular reason for concealing these details was that the excluded measure
was “unrelated to the research question.”34 LeBel and colleagues conclude
that the time has come for disclosure statements to become a mandatory
part of the manuscript submission process at all psychology journals.
Disclosure statements could be as simple as the “21 word” solution proposed
in the same year by Simmons and colleagues (see chapter 2), which simply
requires authors to say: “We report how we determined our sample size, all
data exclusions (if any), all manipulations, and all measures in the study.”35
As we saw earlier, concealing details about the range of attempted and
statistically nonsignificant analyses can dramatically elevate the rate of false
positives. At the same time, failing to disclose details of experimental pro-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
62 | Chapter 3
cedures obstructs other scientists from replicating the work and, in cases
where such replications are attempted, sparks additional contention. A case
in point returns us to the elderly priming controversy of 2012 between
Bargh and Doyen. Recall that Doyen and colleagues failed to replicate
Bargh’s influential finding that priming people with concepts related to
elderly people can lead them to walk more slowly. Doyen claimed that the
original result was due to the experimenter in Bargh’s original experiments
being unblinded to the experimental conditions and inadvertently “priming”
the subjects. When Doyen and colleagues repeated the experiment in
such a way as to ensure effective blinding, the effect disappeared.
In response, Bargh claimed that the experimenter in his original study
was blinded to the experimental condition. Unfortunately, it is impossible
to tell either way because the method section in Bargh’s original study was
too vague. Because of this ambiguity, the debate between Bargh and Doyen
focused on what the original study did rather than the scientific validity of
the elderly priming effect itself—and this dispute spread among the wider
psychological community. Such arguments lead nowhere and generate ill
feeling: where methodological imprecision is allowed to sully the scientific
record, any failed replication can be attacked for not following some unpublished—but
apparently critical—detail of the original methodology. And
where the original researchers seek to clarify their methodology after the
fact, such attempts can be perceived, rightly or wrongly, as face-saving and
dishonest.
If lack of methodological disclosure is such a problem, why don’t journals
simply require more detail? There are no good answers to this question but
plenty of bad ones. Space limitations in printed journals require authors to
cut details from manuscripts that may be seen by editors as extraneous or
unnecessary. Strangely, these policies have persisted even in online (unprinted)
journals where space is no concern. Even worse, because direct
replication is so rare, much of the detail necessary for replication is regarded
as unnecessary by authors, reviewers, and editors. The absence of this detail,
in turn, feeds a vicious cycle that makes direct replication of such work
difficult or impossible. Methods sections have thus faded from their original
purpose of providing the recipe for replication, instead acting as a way for
reviewers and editors to check that the authors followed procedural norms
and avoided obvious methodological flaws. While this is a necessary role of
a method section, it is far from sufficient.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Unreliability | 63
Reason 4: Statistical Fallacies
In chapter 2 we saw how undisclosed analytic flexibility can lead to a range
of questionable practices in psychology that undermine reliability, such as
p-hacking and HARKing. Beneath these errors in the application of statistics,
however, lie even deeper and more fundamental misunderstandings
of null hypothesis significance testing (NHST).
The most frequently misunderstood aspect of NHST is the p value itself.36
To illustrate just how confusing p values can be, consider a scenario
where you conduct an experiment testing the effectiveness of a new kind
of intervention on the number of people who successfully quit smoking. You
give one group of participants the new intervention while the other group
is given a control (baseline) intervention. You find that significantly more
participants quit following the new intervention compared with the control
intervention. The statistical test returns p = .04, which at α = .05 allows
you to reject the null hypothesis.
Now consider which of the following statements is correct:
A. The obtained p value of .04 indicates a 4 percent probability that
the new intervention was not effective (the null hypothesis).
B. The obtained p value of .04 indicates a 4 percent probability that
the results are due to random chance rather than the new
intervention.
C. The fact that there is a statistically significant difference means
that your intervention works and is clinically significant.
D. The obtained p value indicates that the observed data would
occur only 4 percent of the time if the null hypothesis were true.
E. Previous research found that a similar intervention failed to produce
a significant improvement compared the control intervention,
returning a significance level of p = .17. This indicates that
your new intervention is more promising that this previous
intervention.
F. After your study is published, another research group publishes
a paper reporting that a different intervention also leads to a
statistically significant improvement compared to the same control
intervention (p = .001). Because their p value is smaller
than your p value (.04), they conclude that their treatment is
more effective than your treatment.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
64 | Chapter 3
Which of these statements is correct? In fact, all of them are wrong in
different ways. Statement A is the most prevalent misconception of NHST:
that a p value indicates the probability of the null hypothesis (or any hypothesis)
being true. Another common variant of this fallacy is that p = .04
reflects a 96 percent chance of that the effect is “real.” These statements
confuse the p value with the posterior probability of the null hypothesis.
Recall that NHST estimates the probability of the observed data given the
hypothesis, p(data|hypothesis), rather than the probability of the hypothesis
given the data, p(hypothesis|data). To determine the posterior probability
of any hypothesis being true we need to use Bayes’ theorem, which we will
turn to later in this chapter. Statement B is a variant of statement A and
equally wrong. For the same reason that a p value can’t tell us the probability
that there is no effect, it can’t tell us the probability that the effect is due to
chance.37
What about statement C? Can we conclude that the new treatment was
clinically significant because it was statistically more effective than the
control intervention? No, because the p value tells us nothing about the size
of the effect. As statistical power increases, so does the chance of detecting
increasingly small effects. Therefore, depending on the statistical power of
your experiment, p = .04 could indicate an effect so small as to be trivial
in terms of clinical significance. Statistical significance and “real world”
significance are independent concepts and should not be confused, even
though they frequently are.
Statement D moves us closer to the true definition of a p value but is still
crucially wrong. The p value doesn’t tell us the probability of observing your
data if the null hypothesis is true—it tells us the probability of observing
your data or more extreme data if the null hypothesis is true.
In statement E, we’re asked whether your intervention, which yielded a
statistically significant effect (p = .04), can be said to be more effective than
an alternative intervention that produced a statistically nonsignificant effect
(p = .17). Although this interpretation is tempting—and extremely common—it
is incorrect. To conclude that two effects are different requires a
statistical test of their difference: in other words, you would need to show
that your intervention produced a significantly larger improvement in smoking
cessation compared with the alternative intervention. In most cases this
requires a test of the statistical interaction, and only if that interaction was
statistically significant could you conclude a quantitative dissociation be-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Unreliability | 65
tween effects. In 2011, psychologists Sander Nieuwenhuis, Birte Forstmann,
and E. J. Wagenmakers explored the prevalence of this fallacy in 157 articles
published in four of the most prestigious journals in neuroscience. Remarkably,
they found that 50 percent of papers assumed a meaningful difference
between a statistically significant effect and a statistically nonsignificant
effect, without testing for the critical interaction that would justify such a
conclusion.38
Finally, let’s consider statement F. This scenario is similar to statement
E except that both p values are statistically significant. Again, the statement
is false without performing a test to compare the magnitude of the
improvement between your intervention and the newer intervention. Remember
that a p value on its own tells us nothing about the size of an
observed effect—it merely tells us how surprised we should be to observe
that effect, or larger, if the null hypothesis were true. It is entirely possible
that a large effect that yields p = .04 is more important for theory or applications
than a smaller effect from a much larger sample that produces
p = .001. This tendency for scientists to draw conclusions about effect
sizes based on p values reflects the same statistical fallacy as statement
A: that we tend to interpret p values as telling us the likelihood of our
hypothesis being true, with smaller p values making us feel more confident
that we are “right.” This thinking is not allowed under NHST because
a p value can never tell us the probability of truth. Many psychologists
(and scientists in related fields) have fallen into the trap of confusing p
values with a measure that does tell us of the relative likelihood of one
hypothesis over another: the Bayes factor. We will return to Bayes factors
later in this chapter.
Reason 5: Failure to Retract
Scientific publishing is largely a matter of trust, which means that when
major errors are identified it is important that untrustworthy papers are
retracted promptly from the literature. Articles can be retracted for many
reasons, ranging from honest mistakes such as technical errors or failure
to reproduce findings, to research misconduct and fraud.39 If replication is
the immune system of science, then retraction can be thought of as the last
line of defense—a surgical excision.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
66 | Chapter 3
Putting aside fraud, in many sciences the failure to replicate a previous
result because of technical error or unknown reasons is sufficient grounds
for retracting the original paper.40 In physics, chemistry, and some areas of
biology, results are often so clearly positive or negative that failure to replicate
indicates a critical mistake with either the original study or the replication
attempt. Psychology, however, rarely retracts articles because of
replication failures alone. In 2013, researchers Minhua Zhang and Michael
Grieneisen found that the social sciences, including psychology, are several
times less likely than other sciences to retract articles because of “distrust
of data or interpretations.”41 In a previous study they showed that overall
retraction rates in psychology are just 27 percent of the average calculated
across more than 200 other research areas.42
The resistance to retraction in psychology is so hardened that it can lead
to farcical interactions between researchers. One recent case highlighted
by Dan Simons relates again to the work of Yale psychologist John Bargh.43
In 2012, Bargh and colleague Idit Shalev published a study claiming that
lonelier people prefer warmer baths and showers, thereby compensating for
a lack of “social warmth” through physical warmth.44 In 2014, psychologist
Brent Donnellan and colleagues reported a failure to replicate this finding—and
not just in a single experiment but across nine experiments and
more than 3,000 participants, over 30 times the sample size of the original
study.45 Despite this failure to replicate, as well as the presence of unexplained
anomalies in the original data, Bargh and Shalev refused to retract
their original paper. In many other sciences, a false discovery of this magnitude
would automatically trigger excision of the original work from the
scientific record. In psychology, unreliability is business as usual.
What does the rate of retractions in a particular scientific field say about
the quality of science in that field? On the one hand you might think (optimistically)
that fewer retractions is a good sign, indicating that fewer studies
are in need of retraction. More realistically, fewer retractions can be thought
to indicate lower confidence scientists place in their own methods and a
lower bar for publication in the first place. If we take a science such as experimental
physics, where studies tend to have high statistical power, methods
are well defined and de facto preregistered, then the failure to reproduce
a previous result is considered a major cause for concern. But in a weaker
science where lax statistical standards and questionable research practices
are the norm, attempts to reproduce prior work will often fail, and it should
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Unreliability | 67
therefore come as no surprise that retraction is rare. By lowering the bar for
publication, we necessarily raise it for retraction. A strict policy requiring
retraction of unreliable results could cause the thread of psychological research
to unravel, consigning vast swathes of the literature to the scrap heap.
Solutions to Unreliability
What is the point of science if we can’t trust what it tells us? Lack of reliability
is unquestionably one of the gravest challenges facing psychological
science. However there are reasons to be hopeful. In the remainder of this
chapter we will consider some solutions to the five main problems outlined
above, including lack of replication, low statistical power, lack of disclosure
about methods, statistical misunderstandings, and failure to retract flawed
studies.
Replication and power. If the problem is a lack of replication and statistical
power, then the obvious solution is to produce more high-powered research
and instill a culture where, like other sciences, replication is viewed
as a foundation of discovery. Our challenge is how to achieve this reform
within an academic culture that places little to no value on direct replication,
and where jobs and grant funding are awarded based on the quantity
of high-impact publications rather than the reliability of the underlying
science. Placing a premium on work that can be independently replicated
is one major part of the solution. Psychologist Will Gervais from the University
of Kentucky has shown how a policy of replication automatically
rewards researchers who conduct high-powered studies.46 Under the current
system, a researcher who publishes many underpowered studies (“Dr.
Mayfly”; N = 40 participants per experiment) will have a substantial career
advantage over a researcher who publishes a smaller number of highpowered
studies (“Dr. Elephant”; N = 300 participants per experiment).
Using simple statistical modeling, Gervais found that after a six-year period,
Dr. Mayfly will have published nearly twice the number of papers as
Dr. Elephant. However this advantage is reversed when publication requires
every experiment to be directly replicated just once. Both researchers
now publish fewer papers, but because Dr. Mayfly wastes so much time
and resources chasing down false leads, Dr. Elephant publishes more than
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
68 | Chapter 3
double the number of papers as Dr. Mayfly. By aligning the incentive to
publish more papers with the incentive to publish replicable science, the
careful scientist is rewarded over the cowboy.
So much for theory, but how can we incentivize replications in practice?
E. J. Wagenmakers and Birte Forstmann have proposed an innovative solution
in which journal editors issue public calls for replication attempts of
studies that hold particular interest or weight in the literature. Teams of
scientists would then compete for the bid, with the winning team guaranteed
a publication in the journal that issues the call.47 The journal Perspectives
on Psychological Science recently launched a similar initiative called
Registered Replication Reports in which groups of scientists work collaboratively
to directly replicate findings of particular importance, with publication
of preregistered protocols guaranteed regardless of the outcome.
Sanjay Srivastava has gone even further and called for a “Pottery Barn rule”
(you break it, you buy it) in which the journal that publishes any original
finding is required to publish direct replications of the study, regardless of
the outcome.48 Srivastava’s idea is challenging to implement but would incentivize
journals to publish only the work they see as credible. It would
also capitalize on evidence that replication studies have the most impact
when they are published in the same journal as the original paper.49
Journal policies must also play a key role in encouraging, rewarding, and
normalizing replication. The journal BMC Psychology, launched in 2013, is
one of a handful journals to explicitly take up the gauntlet, noting in its
guide to authors that it will “publish work deemed by peer reviewers to be
a coherent and sound addition to scientific knowledge and to put less emphasis
on interest levels, provided that the research constitutes a useful
contribution to the field.” In a provocative article for BMC Psychology, editor
Keith Laws has noted that this policy explicitly welcomes negative findings
and replication studies.50
Bayesian hypothesis testing. One of the strangest aspects of conventional
statistical tests is that they don’t actually tell us what we need to know.
When we run an experiment we want to know the probability that our
hypothesis is correct, or the probability that the obtained results were due
to a genuine effect rather than the play of chance. But the doctrine of
NHST doesn’t allow such interpretations—the p value only tells us how
surprised we should be to obtain results at least as extreme if the null hy-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Unreliability | 69
pothesis were true. This counterintuitive reasoning of p(data|hypothesis)
is why many scientists still harbor basic misunderstandings about the definition
of a p value.
A more intuitive approach to statistical testing is to consider the opposite
logic to NHST and estimate p(hypothesis|data)—that is, the probability of
the hypothesis in question being true given the data in hand. But to achieve
this we need to look beyond NHST—a relatively new invention—and return
to an eighteenth-century mathematical law called Bayes’ theorem.
Bayes’ theorem is a simple rule that allows us to calculate the conditional
probability of a particular belief being true (hypothesis) given a particular
set of evidence (data) and our prior belief. Specifically, it estimates the posterior
probability of a proposition by integrating the observed data with the
strength of our prior expectations. Bayes’ theorem is expressed formally as:
P(H|D) =
P(D)
P(D|H) P(H)
where P(H|D) is the posterior probability of the hypothesis given the data,
P(H) is the prior probability of the hypothesis, P(D|H) is the probability of
the data given the hypothesis, and P(D) is the probability of the data
itself.
One of the simplest applications of Bayes’ theorem is in medical diagnosis.
Suppose you are a doctor who suspects your patient may be suffering
from Alzheimer’s disease. To examine this possibility you give your patient
a 15-minute behavioral test for mild cognitive impairment. The test reveals
a positive result, potentially indicative of the disease. So what is the probability
that your patient actually has Alzheimer’s disease? We can represent
this as P(H|D): the posterior probability of the hypothesis that the patient
has Alzheimer’s disease given that test was positive. Using Bayes’ theorem,
we can estimate this probability if we have three other pieces of information:
the sensitivity of the test, which is the probability that the test would
yield a positive result in a patient who actually has Alzheimer’s disease,
P(D|H), the prior probability of the patient having the disease in the first
place, P(H), and the overall probability of a positive test result regardless of
whether or not the patient has the disease, P(D). Now, let’s further suppose
that the test has a sensitivity of 80 percent, which is to say that that it has
an 80 percent chance of detecting Alzheimer’s disease where the disease is
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
70 | Chapter 3
present; thus P(D|H) = 0.80. Let’s also assume that the chance of the patient
having Alzheimer’s disease within the population is 1 percent (P(H) = 0.01).
To complete the picture, all we need to know now is the chance of the test
returning a positive result regardless of whether or not the patient has the
disease. To calculate this we can expand P(D) according to the “law of total
probability,” which proves that the overall probability of a particular occurrence
(in this case testing positive for Alzheimer’s disease) is the sum of its
constituent conditional probabilities:
P(D) = P(D|H) × P(H) +P(D|~H) × P(~H)
In this equation, P(D|H) and P(H) are defined as above, P(D|~H) is the probability
of a positive result if the patient doesn’t have Alzheimer’s disease (i.e.,
the false positive rate of the test), and P(~H) is the overall probability that
the patient doesn’t have Alzheimer’s disease (calculated as 1−P(H) = 0.99).
Let’s assume that that test in this case has a false positive rate of 5 percent,
so that P(D|~H) = 0.05. The overall probability of a positive test result, P(D),
can then be calculated as:
PD = 0.80 × 0.01 + 0.05 × 0.99 = 0.0575
We can now substitute this value of P(D) into Bayes’ theorem to estimate
the probability that the patient truly has Alzheimer’s disease:
P(H|D) =
0.0575
0.80 × 0.01
= .139
Are you surprised by this outcome? Despite the fact that the test has 80
percent sensitivity and just a 5 percent false positive rate, Bayes’ theorem
tells us that there is still only a 13.9 percent chance that a patient with a
positive test result has Alzheimer’s disease.51 This outcome violates our
common intuitions; indeed, when posed with similar scenarios, even medical
practitioners routinely overestimate the accuracy of such diagnostic
tests.52
Bayes’ theorem is a valuable tool in medical screening because it counters
our erroneous intuitions about probabilities, but calculating P(H|D) on its
own isn’t particularly useful in psychological science. In most experimental
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Unreliability | 71
settings we need to know more than the absolute probability of a given hypothesis
given the data; instead, given the data that we have, we need to
know how likely the experimental hypothesis (H1) is relative to the probability
of a comparator (baseline) hypothesis. This baseline can be anything
we choose, but in general it takes the form of the null hypothesis (H0) that
the effect of interest is null or nonexistent in the population being studied.
By calculating the ratio between P(D|H) for each of our hypotheses, H1 and
H0, Bayes’ theorem allows us to decide which hypothesis is better supported
by the evidence. The resulting value is known as a Bayes factor, B, and tells
us the weight of evidence in favor of H1 over H0.
To calculate B we need to know the relative posterior probabilities of
each hypothesis, P(H|D), and their relative prior probabilities P(H). To calculate
the relative posterior probability of H1 vs. H0, we simply divide one
Bayes’ theorem by the other:
( )
( )P(H1|D)
P(D)
P(D|H0) P(H0)
P(D)
P(D|H1) P(H1)
P(H0|D)
=
This rather cumbersome equation can be restructured as:
P(H1|D)
P(D|H0)
P(D|H1)
P(H0|D)
=
P(H0)
P(H1)
×
P(D)
P(D)
×
Note that the values of P(D) here cancel out to 1, leaving us with the
simpler form:
P(H1|D)
P(D|H0)
P(D|H1)
P(H0|D)
=
P(H0)
P(H1)
×
This equation, in turn, can be restructured to calculate the Bayes factor, B:
P(D|H1)
P(H0|D)
P(H1|D)
P(D|H0)
=B =
P(H0)
P(H1)
÷
In other words, the Bayes factor (B) can be thought of as the ratio of the
posterior probabilities, P(H|D), divided by the ratio of the prior probabilities
P(H). This provides a simple estimate of how likely one hypothesis is given
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
72 | Chapter 3
the evidence relative to the other, and therefore how much a rational observer
should update his or her prior beliefs based on that data. A value of
B = 1 is perfectly ambiguous and should lead to no change in prior beliefs,
while B>1 indicates evidence in favor of H1, and B<1 indicates evidence in
favor of H0. Following the pioneering twentieth-century work of statistician
Harold Jeffreys, a value of B between 1 and 3 is generally considered to be
inconclusive in favoring H1 over H0, while B>3 is judged to be “substantial
or moderate evidence,” B>10 is “strong evidence,” B>30 is “very strong evidence,”
and B>100 is “decisive evidence.” These ratios can be reversed to
provide the same standards of evidence in favor of H0 over H1, with B<0.33,
B<0.1, B<0.033, and B<0.01 providing substantial, strong, very strong, and
decisive evidence, respectively.
The rationale of Bayesian hypothesis testing is intuitive, but how exactly
does it stand to improve the reliability of psychological science? Recall from
chapter 2 that a key source of p-hacking in NHST lies in the violation of
stopping rules—if we simply add participants to a study until p<.05 then
we increase the odds that the p value will dip below .05 by chance and thus
lead to a Type I error. Bayesian analysis, however, is protected against this
possibility: according to the likelihood principle the more data we consider,
the more accurate our estimation of reality becomes. Thus, Bayesian researchers
can continuously add data to their experiments until the evidence
clearly supports H1 or H0. This approach not only eliminates the violation
of stopping rules as a source of misleading information, it is also allows the
most efficient allocation of limited scientific resources.
A second major advantage of Bayesian hypothesis testing is that, in contrast
to NHST, it allows direct estimation of the probability of H0 relative
to H1. Recall that a nonsignificant p value allows the researcher only to fail
to reject H0, and never to accept it or estimate its chances of being true.
The biased nature of this inference is one major cause of publication bias
because any negative finding (p>.05) using NHST is by definition inconclusive,
which in turn allows authors and journals to feel justified in reporting
only positive results. Bayesian inference, however, affords no special status
to H0 and permits inferences about the relative probability of any prespecified
hypothesis given the evidence: where NHST is limited to concluding
an absence of evidence, a Bayes factor provides evidence of absence. Third,
Bayesian tests require authors to transparently specify and justify their prior
hypothesis, allowing the scientific community to see just how well grounded
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Unreliability | 73
it is, and challenge it as necessary. Finally, Bayes factors can be combined
and updated in light of new evidence.53 Bayesian testing thus encourages
scientists to engage in programmatic research that is geared toward replications
and accumulated truths rather than the “snapshot science” of NHST
where a single study showing p<.05 can be impossible to refute.
It must be emphasized that Bayesian hypothesis testing isn’t a cure-all
for reliability. A Bayesian analysis can still be “B-hacked” by manipulating
other researcher degrees of freedom such as the selective rejection of outliers
or selection of dependent variables. No system of statistical inference
is immune to cherry-picking, bias, or fraud. But combined with other initiatives,
including study preregistration, Bayesian analysis is a major part of
the solution.54
Adversarial collaborations. Too often in psychology, as in other sciences,
ego overcomes reason, reducing the reliability and credibility of the results
we produce. In the competitive world of reputation management, it is
often more important to be seen winning an argument than to discover
the truth. One proposed solution to this problem is the widespread implementation
of so-called adversarial collaborations. An adversarial collaboration
is one in which two or more researchers (or teams of researchers)
with opposing beliefs jointly formulate a research design that will determine
a “winner.” A typical formulation involves a team of researchers
(proponents) who believe a certain hypothesis to be true, and a counter
team (skeptics) who believe that no such effect exists. Having reached a
consensus on the optimal design, both sides then implement the experiment
and the results are combined at the end to adjudicate between the
competing hypotheses. The advantage of this approach is that by agreeing
to the rules of engagement in advance (and thus preregistering their hypotheses
and analysis plans), both sides are forced to accept the outcome:
there is no scope to claim, after the fact, that an undesirable result was the
outcome of a suboptimal or flawed methodology used by the opponent.55
Improving reporting standards. One obvious way to improve the reliability
of psychological science is to ensure that published methods and
analyses are more easily reproducible by independent researchers. Recently
a number of journals have moved toward more stringent reporting
standards. In 2013, the journal Nature Neuroscience introduced a new
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
74 | Chapter 3
methods checklist to increase research transparency, while at the same
time eliminating words limits on method sections.56 In 2014, Psychological
Science introduced a requirement for authors to complete a simple disclosure
statement upon the submission of manuscripts:
1) Confirm that (a) the total number of excluded observations and
(b) the reasons for making these exclusions have been reported
in the Method section(s). [ ] If no observations were excluded,
check here [ ].
2) Confirm that all independent variables or manipulations,
whether successful or failed, have been reported in the Method
section(s). [ ] If there were no independent variables or manipulations,
as in the case of correlational research, check here [ ].
3) Confirm that all dependent variables or measures that were analyzed
for this article’s target research question have been reported
in the Methods section(s). [ ]
4) Confirm that (a) how sample size was determined and (b) your
data-collection stopping rule have been reported in the Method
section(s) [ ] and provide the page number(s) on which this information
appears in your manuscript.57
Meanwhile, the Center for Open Science, led by psychologist Brian
Nosek from the University of Virginia, has championed an initiative to offer
badges for adherence to transparent research practices. These include sharing
of primary research data, experimental materials, and study preregistration.58
Broad uptake and adherence to these new standards remains to be
seen, but they are positive and vital moves that emphasize the pressing need
to improve reliability and credibility of psychology.
Nudging career incentives. How can we motivate researchers to care
more about reproducibility? One approach would be to develop a quantitative
metric of reproducibility that counts in career advancement—a system
where your score is improved by how often your work is independently
replicated. In 2013, Adam Marcus and Ivan Oransky proposed such
an index for journals.59 We will return to this possibility in chapter 8 and
consider a concrete mechanism for how it could work in psychological
science.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
C H A P T E R 4
The Sin of Data Hoarding
Code and data or it didn’t happen.
—Anon.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
If passing aliens were to glance at terrestrial science they might conclude
that data sharing is a standard global practice. Scanning the policy of the
National Institutes of Health, one of the major US funding agencies, they
would see that it “requires resource producers to release primary data
along with an initial interpretation . . . to the appropriate public databases
as soon as the data is verified. All data will be deposited to public databases
. . . and these pre-publication data will be available for all to use.”1
Turning to the journal Science, one of the most prestigious peer-reviewed
outlets on the planet, they would see that data sets “must be deposited in
an approved database, and an accession number or a specific access address
must be included in the published paper.”2 They would nod approvingly
and move on.
But our visitors should have dug deeper. Had they studied the policies of
the NIH and Science magazine more closely, they would have discovered
that such rules apply only to specific kinds of scientific information, such
as DNA and protein sequences, molecular structures, and climate data.
Psychology isn’t subject to such requirements; on the contrary it is standard
practice for psychology researchers to withhold data unless motivated to
share out of self-interest or (rarely) when required to share by a higher authority.
Unlike fields such as genomics or climatology, many psychologists
effectively claim personal ownership over data—data that in the majority
of cases is provided by volunteers through the use of public funding and
resources.
Many psychologists consider sharing data only where doing so brings
professional gains, such as working partnerships that lead to joint authorship
of papers. As MIT neuroscientist Earl Miller has said: “I’m for sharing
raw data. But as collaborations. Data is not a public water fountain.”3 Many
psychologists and neuroscientists share Miller’s views. They fear that unfettered
access to data would be to surrender intellectual property to rivals,
allow undeserving competitors to benefit from our own hard work and
invite unwelcome attention from critics. The widespread prevalence of questionable
research practices further reduces the incentive to share. After all,
in the world of major league science, who would want their mistakes or
dubious decision making exposed to the scrutiny of a rival researcher or
(worse) a professional statistician? By treating data as personal property and
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Data Hoarding | 77
adopting a defensive agenda, psychology is guilty of our fourth major transgression:
the sin of data hoarding.
The Untold Benefits of Data Sharing
Why do some areas of science enjoy a culture of transparency while others,
like psychology, are so closed? Are psychologists more selfish than other
scientists? Are they more sensitive to competition or criticism? Could it
simply be that they are less aware of the gains that data sharing can bring
both to wider science and individual scientists?
It is no exaggeration to say that data sharing carries enormous benefits
for science. The most immediate plus is that it allows independent scientists
to repeat the analyses reported by the authors, verifying that they were
performed and reported appropriately. An unbiased perspective can be tremendously
helpful in correcting honest mistakes and ensuring the robustness
of the scientific record. Sharing study materials (e.g., computer code
and stimuli) as well as data can help other scientists repeat experiments,
thus aiding in the vital task of direct replication.
Data sharing also makes it easier to detect questionable research practices—were
I to conduct 100 statistical tests on a data set and report only
the one analysis that was statistically significant (an egregious form of
p-hacking) then this would become obvious to anyone who tries the 99
analyses on the same data that “didn’t work.” Beyond the gray area of questionable
practices, data sharing also enables the detection of fraud. In 2013,
psychologist Uri Simonsohn published a simple and elegant statistical
method for detecting fraudulent manipulation of data.4 As we will see, Simonsohn
used this tool on the raw data associated with several published
papers and in so doing unmasked multiple cases of data fabrication. His
efforts in turn led to university investigations and resignations by established
academics.
In the longer term, sharing data and materials at the point of publication
prevents the permanent loss of information—over time, computers crash;
and scientists change institutions, change contact details, and eventually
die, and unshared data inevitably dies with them.5 This anecdote by psychologists
Jelte Wicherts and Marjan Bakker from the University of Amsterdam
paints an all-too-familiar picture in psychology: “One of us once
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
78 | Chapter 4
requested data from a close colleague, who responded: ‘Sure I will send you
those data, but it’s like seven computers ago, and so please allow me some
time to hunt them down.’”6
Finally, publishing data allows independent scientists, now and in the
future, to perform analyses that the original authors never imagined or had
any interest in pursuing. Complex data sets can hold hidden gems, and aggregated
data across many studies can allow researchers to perform vital
meta-analysis.7
To all these benefits a critic might respond, “Sure, anyone can see how
data sharing benefits science, but what about me as an individual researcher?
What do I stand to gain?” Looking beyond the fact that all publicly
funded scientists have an ethical duty to be as open and transparent as possible,
one of the most selfish advantages is in citation rates. An analysis of
over 10,000 gene expression microarray studies found that those that shared
data attracted up to 30 percent more citations than those that did not, after
controlling for a range of extraneous factors.8 Within the field of cancer
research this relationship is even stronger, with open data associated with
a 69 percent rise in citations.9 The potential career benefits of data sharing
are therefore substantial.
Failure to Share
With so many good reasons to embrace transparency, how many psychologists
actually lift the lid on their own data? Public data deposition in psychology
is unquestionably rare—very few researchers make their data available
to anyone beyond their immediate research team. Nevertheless, all
major psychology journals require authors to retain data for several years
and share it with readers on request. This raises the question of how many
psychologists meet these requirements and share data when asked.
In 2006, Jelte Wicherts and colleagues decided to put this policy to the
test by requesting data from 249 studies that appeared in four major journals
published by the American Psychological Association (APA). Their
experience was unsettling:
Unfortunately, 6 months later, after writing more than 400 e-mails—
and sending some corresponding authors detailed descriptions of our
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Data Hoarding | 79
study aims, approvals of our ethical committee, signed assurances
not to share data with others, and even our full resumes—we ended
up with a meager 38 positive reactions and the actual data sets from
64 studies (25.7% of the total number of 249 data sets). This means
that 73% of the authors did not share their data.10
In the 73 percent of cases where data wasn’t obtained, the original researchers
either failed to respond to e-mails (18 percent), refused to share
the data (35 percent), or promised to send it but never did (20 percent). This
was despite the fact that all the studies were published within the prior 18
months, and despite clear APA guidelines requiring the authors to retain
data and provide it to interested readers.11 In a separate interview, Wicherts
told me that among the specific reasons for refusing were statements such
as, “I am up for tenure and I have no time to do this”; “This is an idiotic
amount of work. I am sorry but I can’t help you”; “This is an ongoing project”;
and “I am afraid your request is not possible.”12
As a solution, Wicherts and colleagues called for public data deposition
to become a mandatory condition for publication. A decade later, the APA
is still yet to act on their recommendation, and only a few major psychology
journals have issued a blanket rule requiring public data deposition.
One reason why psychologists might refuse to share data is for fear of
their results being disputed owing to questionable or sloppy research practices.
If this is the case then statistical errors, particularly those favoring
statistical significance, should be more common in studies where researchers
fail to share. In 2011, Wicherts and colleagues tackled this question by
requesting the raw data for 49 studies published by two APA journals: the
Journal of Personality and Social Psychology and the Journal of Experimental
Psychology: Learning, Memory and Cognition.13 Consistent with their earlier
investigation, they found that fewer than half of the authors (43 percent)
were willing and able to share at least some data. They also found that misreporting
of p values was more than twice as frequent in studies where the
authors failed to provide the data. Most of these errors were clear but relatively
minor—more worrying were the 10 cases where the correct calculation
of p values, based on details in the published article, transformed a
result from being statistically significant to statistically nonsignificant. For
all 10 of these errors, across 7 studies, the authors were unable or unwilling
to produce the data supporting their conclusions.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
80 | Chapter 4
In the face of such evidence you might think that the APA would, at a
minimum, reprimand authors who fail to share data and even retract the
offending papers. Not so. Despite violating the APA’s code of conduct, none
of the cases documented by Wicherts and colleagues has met with official
sanction or criticism of any kind. Wicherts reflects, “The APA did not pursue
any specific authors for not sharing data with us. I am also not aware
of any case in which the failure to share data has led to a formal investigation
or sanction in psychology. In fact, if a given author refuses to share
data, I fear that there is little you can do. In the last 8 years, the APA has
done nothing to improve data availability in any of its 50+ journals.”14
Thus, while the APA code pays lip service to data sharing, it apparently
does nothing to enforce it. Few psychologists, and least of all the APA, seem
to care whether psychologists share their data or not.
Secret Sharing
Even when psychologists are compelled to share data, the ability of other
researchers to work with it can be hampered by usage restrictions. One such
recent case returns us to the 2012 study by John Bargh and Idit Shalev discussed
in chapter 3.15 Recall that in their high-profile paper, they claimed
that lonelier people prefer warmer baths and showers. Their explanation
for this curious correlation was that lonely people receive less social warmth
and so try to make up the difference by increasing their exposure to physical
warmth.16 In 2014, Brent Donnellan and colleagues attempted to replicate
Bargh and Shalev’s results in a much larger sample and found no such
relationship. To try to understand the basis of Bargh and Shalev’s claims
they requested the raw data of the original study, which was published in
the APA journal Emotion, and so was subject to the following clause in the
APA code of conduct:
After research results are published, psychologists do not withhold
the data on which their conclusions are based from other competent
professionals who seek to verify the substantive claims through re-
analysis.17
According to the APA policy, Bargh should have shared the data with his
fellow professionals. Donnellan explains how Bargh shared the data with
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Data Hoarding | 81
Donnellan’s colleague, Joe Cesario, but only with extraordinary conditions
placed on how it could be used:
Bargh let Joe Cesario look at his data but he forbids us from talking
about what Joe observed. So a gag order is in place. I think this is
bull****. There is no reason why there should be a veil of secrecy
around raw data. How can we have an open and transparent science
if researchers are not allowed to make observations about the underlying
data used to make published claims?18
Psychologist Dan Simons, who had previously spoken up against Bargh
following the replication debacle on elderly priming (see chapter 1), also
expressed concern about these restrictions:
As a field, we should be concerned whenever a scientist imposes a
gag order on the public discussion of their data from published studies,
strikes out against those who, for whatever reason, can’t replicate
results, or deletes public critiques of his arguments. Secrecy without
explanation or justification is antithetical to the open discourse central
to the scientific process. Openness is essential to seeking out the
truth, and secrecy about published results leaves outsiders to wonder
about the reason for all the fog.19
It seems unlikely that Bargh’s restrictions on the use of his data would
be legally enforceable. Nevertheless, Donnellan and colleagues respected
the gag order out of professional courtesy. That researchers should feel
entitled to such secrecy is invidious but unsurprising within a culture
where data is regarded as the personal property of the researcher who collected
it.
How Failing to Share Hides Misconduct
The most unscrupulous “researchers” of all, fraudsters, have every reason
to be wary of sharing data. In 2011 and 2012, Uri Simonsohn undertook a
remarkable investigation to expose two major cases of data fabrication using
statistical methods alone—cases that led to him being popularly dubbed
the “data detective.”20 The first case was triggered by an article on social
embodiment in the Journal of Experimental Social Psychology, led by re-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
82 | Chapter 4
searcher Lawrence Sanna. Sanna, from the University of Michigan, claimed
that participants placed in higher physical locations, such as at the top of
escalators, behaved more generously toward strangers than participants at
lower elevations—this was argued to be consistent with a theorized relationship
between morality and height (as reflected in common phrases such
as “the moral high ground”). Over three studies Sanna claimed that people
in higher locations spent more time helping others, were more compassionate,
and behaved more charitably.
Simonsohn’s suspicions were roused when he noticed an odd pattern in
the results. While the reported averages in each study differed substantially
between the different elevation conditions, the standard deviations were
almost the same. For example, in one of Sanna’s experiments, the average
amount of hot sauce given by research participants to strangers in a taste
test (a common proxy for how compassionate participants are to strangers)
was reported to be 39.74 grams in the “high” condition, 85.74 grams in the
“low” condition, and 65.73 grams in the “control” condition. In contrast,
the standard deviations for each condition were virtually identical at 25.09,
24.58, and 25.65, respectively. So while the averages more than doubled
between conditions the standard deviations of those averages varied by just
one-fiftieth of the same amount. This led Simonsohn to suspect that the
data were too clean to be the result of random sampling.
Of course, just because data look too good to be true doesn’t mean they
are the product of fraud. Any number of more benign explanations must
first be ruled out, such as questionable research practices in which the authors
report only a subset of experiments that showed the clearest effects
or fail to report otherwise legitimate data exclusions. The authors in such
cases might even be entirely blameless. Given the strong bias of journals to
publish clean and striking results (see chapter 1), their article could simply
be the one out of thousands that happened to “strike it lucky”—the so-called
winner’s curse.
To test these possibilities, Simonsohn studied other articles by the same
authors as well as articles that used the same experimental paradigms by
different authors. The pattern he found was worrying. Several articles authored
by Sanna showed the same similarity between standard deviations,
and this similarity was much greater for these articles than for a sample of
similar papers in the same field by other authors. Simonsohn then requested
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Data Hoarding | 83
the raw data from Sanna (which, perhaps surprisingly, he provided), and
using simulations based on the actual distributions of the data he estimated
the probability of such patterns occurring due to chance to be 1 in several
billion. On this basis, Simonsohn alerted the relevant university authorities.
Sanna’s articles were subsequently retracted at his own request, and he
resigned.
Using the same methodology, Simonsohn also exposed another case by
Dirk Smeesters from Erasmus University in the Netherlands. Not only did
Smeesters report standard deviations that were too similar to be obtained
from random samples, he also left another telltale sign of fraudulent intervention—a
human bias stemming from the law of small numbers. When
people are asked to generate a short sequence of random numbers, say between
1 and 10, they typically overestimate the spread of the data that would
arise by chance. So, for instance, in a sequence of a dozen numbers between
1 and 10, a typical sequence produced by a person might consist of 1, 3, 4,
4, 5, 6, 7, 7, 7, 8, 9, 10. Truly random sequences, however, can often involve
seemingly unlikely repetitions, such as 1, 2, 3, 4, 4, 4, 4, 4, 6, 7, 8, 8. When
a human fabricates data they will typically obey the law of small numbers
and overestimate the spread, which can be measured by studying the frequency
of the most commonly occurring number (the mode). For the humanly
generated sequence above the mode is 7 (with a frequency of 3), while
for the randomly generated sequence the mode is 4 (with a frequency of 5).
Using simulations based on raw data provided by Smeesters, Simonsohn
found that the frequency of the mode was consistently too low to be due to
random sampling—it occurred just 21 times in 100,000 simulated experiments,
suggesting that it was generated artificially. Following a university
investigation, several of Smeesters’s articles were retracted, and he resigned
even before the investigation had concluded.
The most striking feature of these cases is that while the aggregate data
that is typically reported in journal articles provided the initial suspicion of
unreliable results, it was the systematic analysis of multiple raw data sets that
provided the smoking gun of data fabrication. In the words of Haldane, “Man
is an orderly animal. He finds it very hard to imitate the disorder of nature.”21
Only by scrutinizing raw data can the scale of such imitation be uncovered,
and only by introducing a culture where data sharing is the norm can such
cases be efficiently detected and purged from the scientific record.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
84 | Chapter 4
Making Data Sharing the Norm
Compulsory data sharing may be ideal, but the road to shifting norms can
be rocky, as the PLOS family of journals discovered. Since March 2014 all
PLOS journals now “require authors to make all data underlying the findings
described in their manuscript fully available without restriction, with
rare exception. . . . Refusal to share data and related metadata and methods
in accordance with this policy will be grounds for rejection.”22 Among the
acceptable methods for data sharing are: deposition in a public archive,
making the data available via a third party, or attached as supplementary
files that are stored and made available by the journal. Recognizing the
impotence of journal policies that require authors to share data on request,
PLOS took the bold step of mandating the release of data as a condition of
publication.
The reaction from some corners of the scientific community was swift
and scathing. “PLOS is letting the inmates run the asylum and this will
kill them,” declared prominent science blogger DrugMonkey.23 Among a
long list of objections, DrugMonkey pointed to the increased personnel time
required to prepare data for public release, a fear that mandatory sharing
could lead to “data orthodoxy” in which rigid and unrealistic standards for
data curation are imposed, and the worry that publicly available datasets
could fuel nitpicking by science denialists. Other critics have also criticized
the infeasibility of uploading very large datasets (such as terabytes of video
files), and the fact that data sharing may not in any case prevent or uncover
the most sophisticated acts of fraud.24 Some researchers further worried
that repeated statistical testing within public data archives could increase
the rate of false discoveries, that sharing could breach the confidentiality
of human participants, and that it could encourage acts of “scooping” in
which rivals scavenge archived data and publish on it before the authors
who produced it have gained the benefits they deserve. It has even been
suggested that mandated sharing could disadvantage third world countries
by forcing them to release their most precious commodity to the slick machinations
of first world labs.25 Under sustained attack, the PLOS data-sharing
policy soon earned an ignominious hashtag on Twitter, #PLOSFail.
The intense criticism of PLOS in turn spurred some robust defenses of
open science. Writing on his blog, geneticist Matthew MacManes of the
University of Hampshire accused DrugMonkey and other bloggers of adopt-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Data Hoarding | 85
ing “myopic” and “laughable” positions. “If you are working in academia,
in the US, you are very likely to be funded by taxpayer money,” MacManes
said. “The data you produce is enabled by taxpayers. . . . So why is it such a
hard friggin’ concept that you should be required to share your data freely,
upon publication?”26 MacManes argued that in most cases it is easy to share
data—concerns that doing so is too hard are “corner cases” that are unrepresentative
of most science published in PLOS. He left readers with a challenge:
“Go to any of the PLOS journals and look at the last month of publications.
How many of these contain data so super-special or large that they
could not be posted?”
Viewed individually, none of the objections to the PLOS policy seem to
present a roadblock to sharing. For example, if we consider fears about false
discovery rates due to “data fishing” in online archives, this concern is easily
addressed by highlighting the provenance of the data, recognizing—with
due caution—that analyses of existing archives should be seen as acts of
hypothesis generation that catalyze future hypothesis testing, and never as
hypothesis testing per se. Concerns about confidentiality can also be met,
in most cases, through careful anonymization, while concerns about scooping
can be solved by archiving some data sets with a future embargo date.27
These so-called dark archives can provide original authors with a head start
of months or even years over their competitors.
While many aspects of the PLOS backlash are addressable, others highlight
how existing mechanisms for open data need to be streamlined. Some
of the most frustrating problems are trivially technical. Many of the available
(free) data-archiving services, such as Figshare, Zenodo and Dataverse, place
a limit on the maximum size of individual files that can be uploaded even
though they don’t limit the overall size of an uploaded archive.28 In some
areas of psychology very large files are routinely acquired, such as videos or
neuroimaging data sets, and arbitrary file size limits are an unnecessary
deterrent to sharing. Even with such barriers removed, there is no escaping
the fact that archiving data and code in a form that others can comprehend
requires a much greater investment of time than researchers are accustomed
to spending. Among all the commodities in science, staff time is among the
most expensive, so a commitment to data sharing will need to be recognized
and supported by both funders and academic institutions.
The PLOS policy itself has teething problems. One key question is what
exactly constitutes raw data and how adherence to a sharing policy can
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
86 | Chapter 4
be policed within the sprawling architecture of a megajournal. Unfortunately,
concerns about enforcement appear to have merit. In writing this
book I drew a random sample of 50 studies employing brain-imaging
methods between 1 November 2014 and 1 May 2015, and it was remarkable
how few researchers shared any data at all.29 In spite of the policy
coming into effect months earlier, just 38 percent of papers in the sample
archived their raw fMRI data for restricted or unrestricted access. Authors
who failed to share data consistently cited legal or ethical restrictions, yet
there was also a great deal of inconsistency in the application of such exemptions;
while some authors claimed to be unable to share data with
anyone, others from the same countries were able to share comparable
data without restriction.
Legal and ethical restrictions are one thing—and in many cases may be
justified—but the most worrying trend in this sample of PLOS articles was
how often authors appeared to simply sidestep the policy. In 25 percent of
papers (13 out of 50), the authors claimed that all study data was present in
the manuscript or the supporting information when it is obvious that neither
was the case. In place of data, the paper and supporting information
instead contained summaries of aggregate data, such as figures or graphics.
Somehow—though it is difficult to conceive how—such studies survived
the review process and were considered by academic editors to be in compliance
with the PLOS data-sharing policy.
In June 2015, I raised these concerns with PLOS executive editor Damian
Pattinson. Following an internal investigation PLOS told me that, as of
December 2015, most of the 13 cases had now shared at least some data,
and that the journal was working on updated guidelines to help avoid future
cases:
The number of instances of authors making their full underlying data
available has risen dramatically since PLOS introduced its data mandate.
Staff carry out a number of checks to submissions in relation to
our data requirements and we support our Academic Editors if they
identify any concerns, since they are best placed to advise on whether
the dataset is complete and adequately pertains to the findings reported
in the manuscript. However we acknowledge that we do not yet have
full compliance on PLOS ONE.
If, upon publication of an article, it transpires that the data are not
fully available, we follow up with the authors to request deposition. We
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Data Hoarding | 87
have done this in the cases raised here and so far data for 10 out of the
13 studies have been deposited, and this information has been made
available to readers via a comment on the articles. We are pursuing our
follow up for the remaining three articles. We will continue to pursue
cases that do not adhere to our policies, and are developing more subjectspecific
guidelines for authors so that it is clear what data are required
and how they should be deposited.30
Compared with PLOS, specialist psychology journals have been slower
to take up the mantle of open data—of the dozens of available journals, very
few require data sharing as a condition of publication. Nevertheless there
is evidence of slow but steady evolution. Recent years have seen data-sharing
policies emerge at the journals Cognition, Experimental Psychology, and the
newly established APA journal Archives of Scientific Psychology.31 Each of
these requires data deposition at the point of publication, although Cognition
and Experimental Psychology permit exceptions for legal, ethical, or practical
reasons. Archives of Scientific Psychology offers a stronger policy, mandating
data sharing for all articles while also stipulating that “Next users [of the
data] agree to offer authorship to the originators of the data on any subsequent
publications.” Whether providing data alone is a sufficient contribution
to warrant coauthorship is controversial, and just how such a policy of
mandatory coauthorship can be enforced remains to be seen.32 Nevertheless,
encouraging collaborative relationships between sharers and users offers
a clear incentive for authors to adopt open practices.
Short of mandating sharing, there is also evidence that so-called nudge
incentives increase open practices. Since 2014, Psychological Science has
published open data kitemarks known as “badges” alongside articles that
share data freely in a public repository.33 Since the implementation of open
data badges, the percentage of articles sharing data has risen steadily from
less than 5 percent in 2013 to 25 percent in 2014 to 38 percent in 2015. In
June 2015, the journal Science went one step further in publishing the Transparency
and Openness Promotion (TOP) guidelines, which like the badges
initiative is the brainchild of Brian Nosek and colleagues at the Center for
Open Science.34 The TOP guidelines, which I helped coauthor, call for journals
to periodically review their adherence to various levels of openness,
including standards for data sharing. Although the guidelines do not mandate
adherence to transparent practices, they raise public awareness of the
journal’s adherence to transparency. At the time of writing, the guidelines
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
88 | Chapter 4
have attracted over 750 journals and 60 major organizations as signatories,
though notably not the APA.35
Top-down reforms are being driven not only by journals. In the UK, the
major research councils are increasing their commitment to ensure openness
and long-term preservation of data. The Economic and Social Research
Council (ESRC), which funds more psychology research than any other
agency, now requires grant holders to deposit data in permanent archives.36
In contrast to the PLOS policy, the ESRC’s directive met with no objections.
Psychologists, it seems, have learned to not bite the hand that feeds them.
Grassroots, Carrots, and Sticks
Changes in policy are no doubt effective in changing practices, but they can
also be slow to take shape. Gatekeeping organizations, such as journals and
funders, are often managed by complex and conservative bureaucracies that
instinctively (and sometimes mindlessly) constrict toward the status quo.
A more radical alternative to driving reform, proposed by Cardiff University
psychologist Richard Morey, is to harness the power of gatekeepers at a
grassroots level—the level of peer review.
Science depends crucially on the unpaid service of thousands of peer
reviewers: researchers who in most cases donate their time freely to critique
the work of other researchers in the assessment for publication. Collectively,
reviewers wield enormous control over science and the careers of individual
scientists—they decide which papers meet a sufficient standard for publication,
assessing whether the scientific question tackled by a study is well
justified, whether the methods are valid, and whether the conclusions of
the authors are based on the evidence. In 2014, Morey and colleagues (including
me) launched a project called the Peer Reviewers’ Openness (PRO)
initiative, which invites reviewers to add one more item to the list of essential
manuscript ingredients: that it should share data in a public archive or
provide a reason for not sharing.37 From 1 January 2017, signatories to the
PRO initiative pledge not to review papers in which authors both refuse to
share data and also refuse to say why they cannot share. In this way, signatories
can withhold reviews just as any reviewer would refuse a paper that
lacked critical components such as a method or results section.
The PRO initiative attempts to shift norms in a unique and (potentially)
very powerful way. Rather than lobbying journals and funders to change
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Data Hoarding | 89
their policies, it harnesses the academic freedom reserved by all individuals
in the scientific community—the choice of each researcher to decide that
the publication of data, where possible, is an integral part of the scientific
process. As more reviewers sign up to PRO, the incentive for researchers to
embrace open practices will, in turn, increase. Authors who choose not to
share (and refuse to justify their closed practices) will struggle to find reviewers
to assess their manuscripts, delaying their publication workflow
and placing them at a career disadvantage compared with those who do
share or are able to justify not sharing. At the same time, within a peerreview
system that is already heavily burdened, journal editors will face
mounting pressure to implement open data policies in an effort to ensure a
steady supply of reviews. PRO also carries a direct reward for reviewers
themselves. Because data deposition is rare in psychology, reviewers who
sign up will find their review workloads substantially reduced, at least in
the short term.
When PRO was announced in November 2014 (under the now superseded
name “Agenda for Open Research”) it was met with a mixture of
enthusiasm and trepidation from the psychological community. On the
Facebook group of the International Social Cognition Network (ISCON),
the initiative was criticized by some psychologists as divisive and heavyhanded.
In a heated debate, one senior psychologist warned that the agenda
was “a brilliant move, but coercive and politicizing.” He said, “I am sympathetic
to the cause now, but am concerned with such mass ‘work stoppage’
in order to gain concessions from the field. . . . An academic boycott is a
dangerous thing and something we have thankfully not seen in our discipline
. . . till now. Are we really prepared to go down this route?” Another
researcher likened the initiative to the academic boycott of Israel and called
instead for “a host of less coercive methods.” He added, “We can hold symposia
on the blessings of data sharing. We write articles and books about
this topic. The sky is the limit. . . . My ideal of open science is a science that
persuades, not a science that forces people to do things against their will.”
In response, Richard Morey argued that the initiative doesn’t force authors
to share their data, only to say why they refuse to share. “Since you
can meet the requirements of the agenda simply by giving an explicit reason
why your data is not open, and then the reviewer is asked to take your word
for it, I can’t understand why so many people here have a problem with it.
All you have to do is explain why your level of openness is reasonable. Why
is this unreasonable?”
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
90 | Chapter 4
The answer, of course, is that the real reasons authors decide against
sharing are at a minimum embarrassing and at worst professionally imperiling.
Were psychologists to answer honestly, the (tongue-in-cheek) truth
would be:
• “I’m not sharing my data because I can’t be bothered to organize
the files in a way that makes sense to anyone else—in fact, they
probably won’t even make sense to me when I look back at them in
six months’ time.”
• “The last thing I need is a bunch of second-stringers combing
through my raw data to look for ‘mistakes’ in my analysis. And,
believe me, given how little time I spent rechecking my analysis
(once I’d found what I was looking for) there are bound be some
mistakes!”
• “Share data? How would I explain the three dependent measures I
strategically omitted from my report because the findings didn’t fit
with my conclusion? Or the five equally defensible ways of analyzing
the data that produced completely different outcomes?”
• “The importance of my data is dwarfed only by my towering professional
reputation. Were I to publish my data, hordes of competitors
would descend like vultures, publishing on my dime and depriving
me of dozens of rightful Nature and Science papers. I’m a
scientist, not a charity.”
• “You only want me to release my study materials and code so that
you can benefit from all my hard work. To which I say, go and
write your own code!” (shakes fist)
• “I can’t possibly share my code because firstly it isn’t commented,
secondly it is ugly and nonsensical, and fourthly was rushed and
probably missing something.”
• “Sharing my experimental materials would only make it easier for
other scientists to fail to replicate my results. As things stand now,
when someone tries to replicate me using my vaguely worded
method sections, I can at least shrug away their nonreplication as
due to incompetence on their part, or some unstated but (obviously)
crucial difference between their study and mine. The moment I
start sharing materials, failed replications might suggest that my
work isn’t reproducible!”
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Data Hoarding | 91
• “I’m not sharing because it would be a huge amount of work for no
immediate reward. This whole ‘transparency movement’ is already
making it harder for me to spin the stories I need to get my papers
published in the Prestigious Journal of Implausible Results. The last
thing we should be doing is encouraging the crusaders.”
Unlocking the Black Box
Some years ago, as a novice journal editor handling one of my first assignments,
I saw for myself how closed practices hinder the advancement of
science. A manuscript I was editing came back with two quite critical, but
overall positive, reviews. In one instance, the reviewer chose to reveal her
identity to the authors because the authors had claimed that a result shown
in one of the reviewer’s earlier studies, and which conflicted with the authors’
current findings, might not hold up under a reanalysis using a different
method. The reviewer responded by arguing that this wasn’t the case,
given her intimate knowledge of her own data, and so asked the authors to
modify their claim. In response, the authors requested the relevant data
from the reviewer so that they could run the analysis for themselves. Remarkably—or
perhaps unremarkably given what we now know about standard
practices in psychology—the reviewer denied the request on the
grounds that “I usually do not share the raw data of my experiments with
people that I’m not collaborating with.” This placed the authors (and me,
as editor) in a difficult situation. Here was a reviewer demanding a revision
of a manuscript based on privileged information that only the reviewer held
and was unwilling to disclose. Given the reviewer’s refusal to be transparent,
I followed what I felt was the only defensible course of action: I disregarded
the reviewer’s concern. To this day, nobody knows what an independent
reanalysis of that data might have revealed. I have no idea who benefits
from such a stalemate, but it certainly isn’t science.
Years later, such stories remain common in psychology, but a combination
of initiatives is shifting the field steadily in the right direction. As
journals and funders strengthen their data-sharing policies (despite the
teething problems at PLOS), the TOP guidelines are raising the profile of
those policies and generating vital discussions. Meanwhile the PRO initiative
is incentivizing open practices at a community level. But even beyond
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
92 | Chapter 4
carrots and sticks, there are also a growing number of researchers leading
by example and embracing open practices for no reward. This raises the
question of why, in the current system, a psychologist would choose to push
against the grain, and what can we learn from such acts of apparent
altruism?
To find out I asked psychologist Frederick Verbruggen from the University
of Exeter about his experiences and opinions on open data. Along with
Jelte Wicherts, Verbruggen is one of a handful of psychologists blazing the
trail on data sharing—since 2013, all publications arising from studies in
the Verbruggen laboratory are accompanied by deposited data in a freely
accessible public archive. Why has Verbruggen taken the step to embrace
open practices? He told me it was a gradual evolution precipitated by an
incident in which he attempted to reanalyze some existing data from a study
he ran several years earlier. “As a consequence of not having a decent data
management plan back then, I now found myself really struggling to find
the appropriate data sets (and lost a lot of time). This was not the first time
that had happened to me, so I finally realised that I had to change my data
storage practices.”38 Verbruggen and data officer Myriam Mertens then set
about developing a series of data management and sharing guidelines in his
laboratory. “Once we started implementing the guidelines, I realised that
it was incredibly straightforward to share the data for individual studies.
This allows reviewers to check our data or analysis protocols if they want
to. Once the paper is published, other researchers are free to check and use
our data as well.”
Verbruggen admits that the transition from closed to open practices was
hard, but nowhere near insurmountable. “At the beginning I had to change
how I was storing data, and this did take some effort. But once the systems
were in place, it all became very straightforward. If you have data sharing
in mind from the moment you start programming your experiments, you
can take a few small steps that make data sharing at the end very easy. This
includes simple things, such as adding a documentation file that briefly
describes the experimental procedure and what you can find in the data
files.” Now that such practices are established in his research he believes
they carry personal benefits in addition to the gains for other scientists. “I
hope that people will trust my work more because I am completely open
about how we analyse our data now.” His lab releases its analysis scripts in
the (free) statistical software package R, allowing readers to trace every step
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Data Hoarding | 93
of the analysis. Doing so, he argues, “may also increase citations and visibility
of the research.” He also points out the praise that sharing attracts from
colleagues. “So far, several reviewers have commented positively on my data
sharing practices. I don’t think it has influenced their overall rating of our
papers, but it is nice to see that the extra effort is appreciated. I also provide
a link to data at the end of my presentations, and some people have commented
positively on that as well.”
Verbruggen is enthusiastic about sharing. “I would strongly recommend
it to everybody,” he says. At the same time, he also believes in the importance
of authors receiving credit for sharing data. “Others can freely use
them, but I do expect people to cite the source of these data. For example,
we upload everything on our institutional repository, which provides a handle,
and for every study we add a statement that says ‘Unrestricted use
permitted but please acknowledge source and include the dataset handle.’
For me this is a matter of scientific integrity.”
What does Verbruggen make of the storm of criticisms thrown at datasharing
policies, such as those launched by PLOS? Although strongly supportive
of data sharing at the point of publication, Verbruggen is emphatic
about the importance of opt-out mechanisms where sharing may be unethical
or infeasible. “I’m the first to admit that data sharing is rather easy for
purely behavioural studies, whereas it may be less straightforward for neuroscience
studies with large data sets. But even for those studies, it should
be possible to share at least some aspects of the data.” He also recognizes
the concerns of researchers who fear being scooped or prefer to share data
in exchange for authorship. “Some studies take a very long time to run and
lead to large, multifaceted data sets. And indeed, in some sub-disciplines, it
is common and perfectly acceptable for researchers to extract multiple papers
from such data sets. So I can understand that the researchers want to
have first dibs, or at least a co-authorship, in such situations.” Is the solution
in such cases not to share? No—instead Verbruggen advocates the use of
embargoed data sets (or so-called dark archives). “You can impose an embargo,
so that only after a specified period the data become publicly available.
Another option is that the owner of the dark archive data has to give permission
for each request for access. The advantage is that data can still be preserved
for the long term, while owners retain control over data re-use.”
One point is clear. Even among its strongest advocates, data sharing in
psychology is far from a black-and-white issue. Dorothy Bishop has warned
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
94 | Chapter 4
that the public availability of large multivariate datasets could encourage a
form of p-hacking in which researchers trawl through archived data to
search for patterns in support of particular conclusions—and in some cases,
such analyses could even be weaponized in order to discredit the original
research (or researchers) in pursuit of a political objective. Bishop also highlights
the dangers of publication bias: “Suppose we have a researcher interested
in the possible impact of diet on children’s cognitive development.
There are several large publicly available datasets that could be used to
address this question. If a positive association is found, then the result is
reported, but if it is not, the researcher is likely to move on to look at something
else—unless they embarked on the analysis with the intent of disproving
the link.”39 As a solution, Bishop suggests that secondary analyses with
clear hypotheses should be transparently preregistered, with unregistered
analyses considered exploratory.
Verbruggen stresses that he would never seek to impose a policy of mandatory
or universal data sharing blind to the consequences. Instead he appeals
to the better angels of our nature, noting simply that he “would just
strongly encourage people to share their data whenever reasonably possible,
which should be possible for the vast majority of cognitive psychology and
basic science studies.” In psychology it appears that the devil lies in negotiating
a consensus in what different researchers consider to be reasonable
and possible sharing of their chief currency. When asked directly, most researchers
agree that transparency and data sharing are good (even best)
professional practice. However, with few incentives to share—and several
incentives not to—reform may be a long road. For data sharing to succeed
in psychology, the community needs to see personal gains. As one Twitter
user, Greg Hale, put it: “I’ll trade access to my 2 TB [terabytes] for the
world’s exabytes any day!”40
Preventing Bad Habits
Verbruggen successfully reformed his lab practices, but what about preventing
suboptimal practices before they can take hold? Psychologists Candice
and Richard Morey have proposed an innovative “data partners scheme” in
which early-career scientists in different labs pair up in order to swap data,
develop a five-year data curation plan, and check that each other’s data can
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Data Hoarding | 95
be understood before submitting to an archive. The overall aim of the
scheme is to steer young scientists toward open practices from the earliest
point, instilling awareness of the importance of openness and the necessary
skill set to achieve it:
For many scientists, open science is a difficult choice because it is
encumbered by a number of unnecessary pragmatic concerns flowing
from habits formed over many years. Openness is not truly a free
choice, driven by the merits of open science. This need not be the
case for the next generation of researchers. Senior researchers have
an important role to play in helping their advisees form good habits
and develop a theory of scientific mind. The data partner scheme,
the five-year plan, and the submission check can help establish good
lab practices, with the benefit that students will be prepared for a
more open science.41
Other sciences show us what we can achieve when we reach the end of
that road. In crystallography, the definitions of what constitute reasonable
and possible data sharing are now unanimous, and the field has embraced
transparency. Stephen Curry, professor of structural biology at Imperial
College London, believes that data deposition has strengthened the quality
of science in his field. “Publication now requires deposition of the coordinates
of the published structure in the Protein Data Bank and I think I’m
right in saying that the X-ray diffraction data on which the structure is based
should also be deposited. This helps to encourage healthier attitudes to data
analysis and should enable improved future analysis as better software tools
become available. The crystallographic community is generally quite proud
of its openness in this regard.”42 With much effort and a little bit of luck,
psychology may one day be able to say the same.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
C H A P T E R 5
The Sin of Corruptibility
I was doing fine, but then I became impatient,
overambitious, reckless.
—Diederik Stapel, 2012
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
It’s late. Everyone has gone home, and you’re tired. After a grueling week
you have finally collected all the data for your latest study. It’s a crucial
experiment, one that your manuscript reviewers demanded in order to
corroborate the four original experiments in your paper. If the results look
good, your work will most likely be accepted for publication in Nature. This
would be a “career maker,” a phrase your supervisor says with a knowing
glint in the eye. But if the results are negative or unclear, the reviewers’
doubts about the value of your study will be confirmed, and it will almost
certainly be rejected. Months of work are at stake. With a Nature paper
under your belt that early-career fellowship is suddenly within reach. With
it would come independence and job security, not to mention prestige and
a dose of jealous admiration from your peers.
You run the analysis. The test returns p = .08. Damn, so close, but for
Nature an inch is as good as a mile. What to do? You run through the potential
solutions. Collect more data to push the p value below the allimportant
.05? This is standard practice among your colleagues, but collecting
more data is time consuming (you are already approaching the
deadline the journal editor gave you), and, in any case, there is no guarantee
that the p value would go down rather than up. Also, you’re sure you read
once that collecting data until p<.05 is frowned on by statisticians. You
remember asking your supervisor about this once, but the reply was just a
snort and a roll of the eyes. No one really cares what the stats geeks say.
Another possibility beckons. Perhaps you can analyze the data differently.
Yes, change the rule for excluding outliers from 3 standard deviations
to 2. Tweaking the outlier exclusion is legitimate isn’t it? It would be harder
to justify why you used a different exclusion rule in this experiment from
the previous one, but stringency looks good to reviewers, and there’s plenty
of precedent in the literature for 2. You rerun the analysis. p = .067. Close
but still no cigar. In annoyance you try 2.5. Then 2.25. Then 1.75. Then 4.
Nothing works. The p value is a rock, unyielding.
Another brainwave. Try a different kind of analysis altogether. Yes, that
would still be justifiable at a stretch. A few clicks. p = .057. Tantalizingly
closer, but still not close enough to even be able to round down to the notorious
white-lie of “p = .05.” Something big must be holding you back. You
study the individual data. Ten participants. Five show the effect you need
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
98 | Chapter 5
to see. Three don’t show much either way. Two show opposite effects, one
of which is substantially against the hypothesis. Yes, that’s the problem, that
one subject with the huge opposite effect. Something must have gone wrong.
Perhaps they performed the task incorrectly? Maybe they figured out the
hypothesis or didn’t follow the instructions? In any case, they’re a Bad Subject.
You wonder what would happen if you just dropped that subject from
the sample—you’re pretty sure you could find a reason if needed. Now p =
.052. Achingly close. Ok, so what if you flipped the results for that subject—
turned condition A into condition B and vice versa. Just this one subject,
just to see what it would have looked like if the results were as expected.
Now p = .001. Magic. Not just borderline significant, but indisputably significant.
A scientist-who-knows-what-they’re-doing level of significance.
Reality returns with a thud. You changed the data. You touched the untouchable.
You click undo. Back to a drab p value that won’t budge, and even
after applying every questionable (but legal) practice you know, it will never
look as impressive as that p = .001. No Nature paper, then.
But it’s late and nobody is around. Nobody else has seen this data. Nobody
else has as much riding on this experiment as you do—whether you
sink or swim is entirely in your own hands. Who would ever know if you
just flipped the raw data for this one subject in this one condition? Sure, it’s
not The Truth, but your four original experiments were as honest as anyone’s,
and they all suggest that this experiment really should have worked.
In any case, the concept of truth is subjective, right? If you alter this data
now you’ll be able to push the rest of your experiments into a high-profile
journal. Then you’ll have a shot at securing that fellowship, and you can
make a major contribution to science. You would surely be a fool not to take
this opportunity. You’ve never altered data before, and you would promise
yourself never to do it again. You’ve worked so hard and deserve this win.
Nobody would ever know. You’re already starting to forgive yourself. Nobody
will ever know.
How many scientists have faced such a dilemma and yielded to the temptation
to fabricate results? How many professors achieved their status because
of fraudulent behavior earlier in their careers? How many scientists
escalate their behavior from the gray area of questionable practices to partial
or entire fabrication of data sets? Unfortunately, recent cases suggest
that fraud is not as rare in psychology as we would hope. Falsifying data
offers a low-risk, high-reward career strategy for scientists who, for whatever
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 99
reason, lose their moral compass and sense of purpose. Like many other
sciences, psychology is based on an honesty system that is poorly equipped
to prevent or detect fraud. Worst of all, when exposed, fraudulent researchers
and their institutions are often minimally accountable while whistleblowers
face vilification. For turning a blind eye toward fraud and its consequences,
psychology perpetrates our fifth major transgression: the sin of
corruptibility.
The Anatomy of Fraud
Diederik Stapel liked to win. From a young age, so he writes, he was competitive
and enjoyed intellectual challenges such as philosophy, politics,
literature, and drama.1 After failing in the theater, he pursued psychological
science and found that it provided an arena that attracted prestige and admiration.
Stapel claims to have begun his career at the University of Amsterdam
honestly,2 attempting (via his doctoral work) to stitch together an
understanding of various social psychological phenomena that were held
together by vague theories built on weak evidence:
I did a lot of experiments, but not all of them worked. Sometimes I
didn’t quite get the result I’d expected, and sometimes I got nothing
at all. When the latter happens, the best thing to do is to cut your
losses and walk away; it didn’t work, never mind, on to the next idea.
But when I really believed in something—when I really wanted what
I’d thought up, or dredged up from the bowels of the literature, to be
true—I found it hard to give up, and tried one more time. If it seemed
logical, it must be true.3
Even at this early career stage it is clear that Stapel was far from the ideal
model of a scientist. Like many psychologists he wanted positive results,
and moreover he wanted the positive results he expected. He struggled (as
any researcher would) to achieve such aims within a culture where low
statistical power shrouds most results in uncertainty; where publication
bias buries negative results or necessitates p-hacking to convert negatives
into positives; where failed replications or inconsistencies are nearly always
explained by differences in methodology; and where most published findings
are regarded (falsely) as true positives. As we’ve already seen, the out-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
100 | Chapter 5
come of such biased and weak empiricism is complexity and chaos—a world
in which experimenting decomposes into an art. Stapel was keenly aware
of this problem.
Everything had a lot of critical dependencies and the final result
could depend on thousands of little things. It was a science of “on the
one hand, on the other hand” models and theories, which were successively
refined to specify in what situations, under what conditions,
at what moment, using what measurements certain effects
could be detected. Nothing was always true; everything was conditional
on something, and there were always exceptions. For every
“rule,” it turned out that there was some kind of qualification, some
“but” or “perhaps” or “sometimes it works, sometimes it doesn’t.”4
Such a climate makes it difficult for accounts to be falsified—as Chris
Ferguson and Moritz Heene have pointed out, the literature quickly becomes
a “graveyard of undead theories” that can neither be confirmed nor
refuted.5 Stapel writes that, in his early (more honest) years, some of his
peers regarded him as a researcher who formulated overly complex theoretical
explanations. In a vain attempt to make sense of the chaos he saw before
him, he layered his theories on top of verbose summaries of the literature
while his colleagues seemed able to synthesize simpler and more elegant
accounts of their data.
I regularly heard, via the rumor mill, that a senior Dutch academic
regularly mocked in his classes what he called my “pinball psychology,”
because it needed a ridiculous amount of crazy features to keep
the ball rolling. A famous American psychologist—one of my heroes—called
me a “grinder,” by which he meant that I put all the
relevant, reliable, robust effects in a pot along with all the edge cases
and exceptions, and ground them to a featureless, incomprehensible
pulp. He considered that my theories and models were highly sophisticated,
but also too complicated to understand. . . . My stories
weren’t elegant enough, and my results weren’t exciting enough, to
get published in the top journals. Even my best efforts were seen as
mediocre. Not impressive enough, not interesting enough, not innovative,
much too complicated. I wasn’t good enough.6
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 101
Faced with his own inadequacy, Stapel appeared to have three choices.
The first and most challenging approach would have been to defend the
complexity of his theories as the most rational attempt to understand the
available evidence. He could have taken aim at the weak evidential standards
in social psychology, conducted larger and more conclusive studies
(at the cost of achieving lower output), and even attempted to drive improvements
in robustness and transparency. Alternatively, he could swallow the
insults and carve out a respectable academic career within the boundaries
of the status quo—an approach that would still be a struggle given the “publish
or perish” research culture and his own apparent limitations as a researcher.
This would have placed him on par with many psychologists,
fashioning stable but unremarkable careers by engaging in questionable
practices to generate publishable stories from their data. Finally—his chosen
option—he could immerse himself in the game he believed he needed to
win, offering a performance that from the outside appeared rigorous and
brilliant yet, in reality, was built on lies. And so Stapel seemed to embrace
a kind of scientific nihilism, abandoning the premise that any truth at all
can be gleaned from psychological science, and instead focusing his efforts
on the personal reward of being published in top journals, pulling in grant
funds, and attracting applause. In Stapel’s world, studies often became
tricks, sometimes ingenious ones based on a careful reading of the existing
literature, but nonetheless acts of deception rather than serious attempts to
understand the mind and behavior.
After years of balancing on the outer limits, the grey became darker
and darker until it was black, and I fell off the edge into the abyss. I’d
been having trouble with my experiments for some time. Even with
my various “grey” methods for “improving” the data, I wasn’t able to
get the results the way I wanted them. I couldn’t resist the temptation
to go a step further. I wanted it so badly. I wanted to belong, to be
part of the action, to score. I really, really wanted to be really, really
good. I wanted to be published in the best journals and speak in the
largest room at conferences. I wanted people to hang on my every
word as I headed for coffee or lunch after delivering a lecture.7
Fabricating data wasn’t the first port of call for Stapel. He admits that
before he reached that stage, he would routinely engage in softer acts of
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
102 | Chapter 5
misrepresentation, deliberately exploiting various forms of p-hacking to
produce publishable outcomes—in his words, “techniques that could put
a healthy looking shine on otherwise mediocre results.”8 He would drop
or combine dependent measures, drop entire groups that failed to show
effects, and creatively exclude outliers whose results opposed the study
hypothesis.
I knew that this was at the outer limits of what was acceptable, or
even beyond those limits in some cases, but . . . I wanted the results
so badly. Plus, it worked in an earlier study, and the results were
really almost, almost right, and it must be true because it’s so logical,
and it was such a great idea, and I surely wasn’t the only person doing
this sort of thing. What I did wasn’t whiter than white, but it wasn’t
completely black either. It was grey, and it was what everyone did.
How else did all those other researchers get all those great results?9
But questionable practices alone were not enough to satisfy Stapel’s ambition,
at which point he took the extra step into data fabrication. Following
a thorough analysis of the literature, he would design whole (often clever)
studies and then, along with them, manufacture entire data sets, thus eliminating
the risk of the data failing to confirm his hypothesis. Stapel writes
about the first time he fabricated the results of an experiment within his
office at the University of Groningen.
I looked at the array of data and made a few mouse clicks to tell the
computer to run the statistical analyses. When I saw the results, the
world had become logical again. I saw what I’d imagined. I felt relieved,
but my heart was heavy. This was great, but at the same time
it was very wrong.10
Immediately he began to achieve his goals—the coming years would see
keynote addresses at conferences, high-impact papers in prestigious journals
such as Science, further academic promotion, professional prizes, and a
reputation as an academic rock star who knew how to make experiments
work.
For Stapel, insecure and craving approval within a discipline where the
pursuit of truth was (in his mind) pointless, fraud was the rational choice—a
high-reward strategy with low risk of detection that achieved his ambitions
and attracted little suspicion. Fabricating data brought order to chaos and
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 103
enhanced his career. In a twist of irony, Stapel would advance to senior
positions at the University of Tilburg that included teaching of research
ethics and proposing a new management model that aimed to help staff
escape the “publish or perish” culture.11 Along the way he would see other
researchers successfully replicate his fake results, fueling the delusion that
“[w]hat seemed logical was true, once I’d faked it.”12 For more than a decade,
Stapel successfully ring-fenced his fraud and so lived a double life: by
day a charismatic, high-flying dean; by night an anxious fraudster manufacturing
data sets at the kitchen table.
In August 2011, having invented data for over 50 publications, Stapel
finally achieved the fame he long desired—but not in the way he anticipated.
In an act that curtailed their own careers, Stapel’s fraud was exposed
by a group of three junior researchers—including at least one of his own
PhD students.13 Following a damning interim report of the case published
by Tilburg University,14 Stapel quickly admitted his guilt. “I have failed as
a scientist, as a researcher,” he said. “I have adjusted research data and faked
research. Not just once, but many several [sic] times, and not just briefly,
but over a long period of time. I am ashamed of this and I am deeply sorry.”15
In a statement within the interim report, however, he denied committing
fraud for personal gain at the expense of the careers of his students and
staff. “I must emphasize that the errors I have made were not motivated by
self-interest. I do not identify with the picture that has been sketched of a
man who has attempted to use young researchers for his own gain. I have
made mistakes, but I was and remain sincerely committed to the field of
social psychology, young researchers, and other colleagues.” Stapel was dismissed
by Tilburg University, pursued intensely by the media, and successfully
prosecuted (albeit on minor charges) in the Netherlands; he later relinquished
his PhD from the University of Amsterdam.
The scale of Stapel’s misconduct was unprecedented in psychology and
one of the largest discovered in science—never before had such a senior and
well-known psychologist been exposed as such a prolific fraudster. At the
time of writing, the number of retracted papers involving Stapel has reached
58, making it the third highest worldwide in any area of science.16 Several
of Stapel’s PhD students were found to have unwittingly included fabricated
data in their PhDs, and although the official investigation concluded that
they could retain their degrees, many of their peer-reviewed publications
were retracted. Most of his students either left academia or took up posi-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
104 | Chapter 5
tions at obscure institutions. Stapel, however, continued his role in the education
system. In 2014 he taught social philosophy at Fontys Academy for
Creative Industries, a higher education institution in the Netherlands. During
this time, doubts about his apparent redemption were raised when he
was accused of defending himself under a fake identity, a questionable Internet
practice known as sock-puppeting. In response to concerns about his
returning to teaching, Stapel allegedly masqueraded as a user named “Paul,”
writing on Retraction Watch:
Perhaps, God forbid, Stapel is able to teach his students valuable lessons
and insights no one else is willing to teach them for a
2-hour-a-week temporary, adjunct position that probably doesn’t pay
much and perhaps doesn’t pay at all. The man is a failure, yes, but he
is one of the few people out there who admitted to his fraud, who
helped the investigation into his fraud. . . . Nowhere it is written that
failures cannot be great teachers. Perhaps he points his students to
other frauds, failures, and ridiculous mistakes in psychological science
we do not know of yet. That would be cool (and not unlikely). Is
it possible? Is it possible that Stapel has something interesting to say,
to teach, to comment on?17
Notwithstanding his alleged sock-puppetry, Stapel found support in
some unlikely corners. Psychologist Jelte Wicherts argued that Stapel had
already paid the price for his misconduct and had the right to pursue a
teaching career. “He lost his scientific career,” wrote Wicherts on Retraction
Watch. “He relinquished his PhD title. . . . As far as I know, he did not
threaten to sue anyone. He did 120 hours of community service. As long as
he doesn’t do unsupervised research again, I see no problems with him
teaching a course at Fontys.”18
The final report into Stapel concluded with a number of scathing criticisms
of academic life, condemning Stapel’s nearly untouchable status
within Tilburg University, the failure of senior academics to take two prior
allegations of his fraud seriously, the barriers and dangers faced by the
whistle-blowers, the prevalence of questionable research practices that obscure
fraud, and the failure of scientific criticism to detect his fraud sooner.19
Among a long list of recommendations to improve transparency, the committee
called for an inquiry into research standards in social psychology.
At the time writing, many of these recommendations have yet to be imple-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 105
mented—a lethargic albeit unsurprising response. The extreme nature of
Stapel’s fraud has allowed him to be dismissed by the establishment as a
one-in-a-million aberration rather than the predictable by-product of a culture
that, in Stapel’s own words, rewards “dig and deliver” over “dig and
discover.”20
For many psychologists, the instinctive reaction this case is to believe
that fraud is rare, and that Stapel not only manipulated his data but also
played the academic system in ways that few other scientists would contemplate.
The available evidence, however, suggests that this belief, while
comforting, is naive. The scale of Stapel’s misconduct was among the greatest
known in science, but fraud is by no means as rare or insignificant as
many assume. A 2009 meta-analysis by Daniele Fanelli concluded that
approximately 2 percent of scientists (including a range of fields other than
psychology) admit to engaging in some form of data falsification.21 In 2012
Leslie John and colleagues further estimated that as many as 40 percent of
psychologists have falsified data on at least one occasion.22 If this statistic
is reliable then psychology is a discipline in which there are almost certainly
a substantial number of undetected fraudsters currently in operation,
perhaps on the scale of Stapel or greater. The question then becomes not
if Stapel is the tip of the iceberg but why, and what if anything can we do
about it?
The Thin Gray Line
There is no question that the practices of Diederik Stapel were among the
most extreme acts of academic misconduct known, but where should we
draw the line between fraud and the gray area of legal but questionable
practices that form so-called cultural problems in science?23 To illustrate
this dilemma, consider the following scenarios. Which of the following do
you consider to be scientific fraud?
1) A scientist collects 100 observations in an experiment and discards
the 80 that run counter to the preferred hypothesis.
2) A scientist runs ten experiments then selectively writes up the
two that produced positive results that agree with the preferred
hypothesis.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
106 | Chapter 5
Most would agree that the first scenario is fraudulent. Many (perhaps
most) of us would also view the second scenario as fraud, or at least a degree
of misconduct. Mathematically, the two scenarios are the same: whether
the researcher discards 80 percent of their data or 80 percent of their experiments,
they will end up in the same place—with biased evidence. Morally,
perhaps they are also on par. After all, how is selectively reporting an
experiment based on a desirable outcome any less dishonest than selecting
data within an experiment for the same reason? But now consider a third
scenario:
3) A journal reviews ten papers on the same topic and selectively
publishes the two that reported statistically significant effects.
Things just got complicated. In a breath we’ve ascended from individual
dishonesty to the “cultural” problem of publication bias. We can now assert
a diminished individual responsibility, forgetting that the incentive structure
in scenario 3 (“We publish only positive results”) drives scenario 2 (“I’ll
write up only experiments that produce positive results”), which in turn
encourages scenario 1 (“I need this experiment to show positive results”).
Unless we define scientific fraud narrowly as (only) the kind of extreme
fabrication perpetrated by Stapel, it is difficult, if not impossible, to distinguish
the blackest of fraudulent acts from a continuum of dishonest practices
at the individual and group level. What about analyzing a dataset 100
different ways and reporting only the analysis that “worked”? Or trawling
through methods of excluding outliers to push a p value below .05? Are
these acts fraudulent? If not, are we excusing such practices on rational
grounds or simply because they are so common? What would your next-door
neighbor think?
Stapel himself admits that his journey toward data falsification was gradual,
beginning within the gray zone of questionable but common practices
in scenario #2 (discussed in chapters 1–3) before escalating to scenario #1
and beyond, all prompted by publication bias in scenario #3. This suggests
that deliberate exploitation of questionable practices may provide a gateway
to fraud, which in turn raises the tricky question of whether conscious exploitation
of questionable practices should also be treated as fraudulent. The
case of Dirk Smeesters, introduced in chapter 4, provides an intriguing case
in which these questionable practices themselves were judged as misconduct,
leading to article retractions and Smeesters’s resignation.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 107
We can recall that Smeesters was exposed by Uri Simonsohn after his
results appeared too good to be true. In the official report prepared by Erasmus
University in 2012, Smeesters confessed to selectively removing data
that didn’t fit his hypothesis:
Smeesters admitted that he “massaged” data in three articles . . . to
achieve [statistical] significance and that he did not save the raw data
adequately, and they were lost as a result. The massaging of data was
done using the “blue dot technique” in which subjects who have not
read the instructions properly are identified. Massaging takes place
by first analysing all of the subjects; if the outcome is not strong
enough then the analysis is repeated omitting the subjects that did
not read the instructions properly. Smeesters said that data massaging
using the blue dot technique is common, and that he [is] therefore
part of a culture. He agreed with the committee that this culture
is open to discussion and that this does not detract from the fact that
he applied this technique without stating so in the papers.24
One particularly notable aspect of this case was that the report did not
uncover sufficient evidence to conclude that Smeesters fabricated data. The
most it could conclude was that he deliberately exploited practices that the
majority of psychologists admit to using, and that he also kept bad records.
Was this a sufficient reason to prompt his resignation, which was received
by the university even before the investigation had concluded? Psychologist
Rolf Zwaan from Erasmus University chaired the investigation and believes
it was. “He was right to resign,” Zwaan told me. “I suspect he feared that
further scrutiny would reveal more problems. I don’t think he was redeemable.”25
At the same time, Zwaan acknowledges that they had insufficient
evidence to demonstrate that Smeesters was falsifying his results in a similar
way to Stapel.26 “We didn’t think we could prove that he had fabricated
data,” he explained. “I don’t want to speak for the others [on the committee]
but I certainly suspected he did. Our solution was to say: the results are too
good to be true and since the raw data are missing, we think that these
papers ought to be retracted.” At the time of writing, seven of Smeesters’s
articles in peer-reviewed journals have either been retracted or earmarked
for retraction.27
Smeesters claimed that the practices he used were commonplace, an allegation
that the 2012 survey of US psychologists by Leslie John and col-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
108 | Chapter 5
leagues would later confirm. Zwaan agrees. “It seems plausible to me,” he
said. “I have to admit that before being on the committee I was almost
completely unfamiliar with his area of research. . . . When I started reading
more in that area, I was dumbstruck. During his second interview,
Smeesters dropped a stack of papers on the table and said: these people are
doing exactly the same thing I’m doing.” Responding in a later interview
with a journalist, Smeesters claimed that his behavior was motivated by
publication bias. “Journals almost exclusively publish studies with statistically
significant results, which can indeed lead to massaging your data.” He
also denied that his resignation was linked to the misconduct case, claiming
instead that it was due to “burnout.”28 “I definitely will not miss academia,”
he said. “Frankly, I am relieved that I am no longer part of it. . . . If I look
back now, I took the job because I wanted to do research. But in retrospect
my happiest moments were when I returned from a lecture that went well,
and the students had treated me to an applause.”29 On that point, at least,
Smeesters and Stapel shared one characteristic—the desire to please their
audiences.
The Smeesters case sets an interesting precedent in which deliberately
engaging in questionable practices—routine procedures that many psychologists
admit (anonymously) to exploiting—led to a conclusion of research
misconduct and resignation by a tenured professor. However, this standard
is not always applied consistently. Months later, the case of German psychologist
Jens Förster would show that even when researchers admit to
selectively reporting results that support their hypothesis, dismissal can be
avoided through strenuous denial of fraud.
The Förster saga began in 2012 when an anonymous whistle-blower noticed
that the results in three of Förster’s publications looked too good to
be true. Over 40 experiments, the papers in question examined the effect
of different kinds of perceptual training (termed “induction”) on so-called
creative and analytic thinking. Each paper reported that training people to
focus on perceiving objects “locally” (that is, on a fine-grained scale) induced
more critical, analytic thinking, while training them to focus on
objects “globally” (on a larger scale) induced more creative, open-minded
thinking. For example, upon viewing a large letter H that was composed of
many small Fs, responding to the H trained people to think globally. The
whistle-blower noticed that the results in the experiments varied too symmetrically
around a neutral control condition, leading to patterns that
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 109
seemed too linear to be the result of random sampling (see figure 5.1 for an
example from one of Förster’s papers). A statistical analysis suggested that
the probability of finding such perfectly linear (or more linear) results across
all three papers was just 1 in 508 trillion30—which, put in perspective, is
about 50,000 times less likely than two unrelated people having identical
forensic DNA profiles or equivalent to winning a national lottery several
times in a row.31 The initial complaint by the whistle-blower also pointed
out various other unusual characteristics of the experiments, including implausibly
large effect sizes, a gender distribution that didn’t match the demographics
of the recruited undergraduate population, the complete (and
implausible) absence of a single participant drop-out across more than 2,000
participants, and the fact that Förster claimed to have since “dumped” the
Study 10bStudy 10aStudy 9b
Study 9aStudy 8Study 7
Study 6
3.6
Study 5Study 4
Study 3
3.6
3.4
3.2
3.0
2.8
2.6
2.4
2.2
2.0
3.6
3.4
3.2
3.0
2.8
2.6
2.4
2.2
2.0
3.6
3.4
3.2
3.0
2.8
2.6
2.4
2.2
2.0
4.5
4.0
3.5
3.0
2.5
2.0
5.0
5.5
4.5
4.0
3.5
3.0
2.5
5.0
4.5
4.0
3.5
3.0
2.5
2.0
5.0
4.5
4.0
3.5
3.0
2.5
2.0
5.0
HighMediumLow
4.5
4.0
3.5
3.0
2.5
2.0
5.0
4.5
4.0
3.5
3.0
2.5
2.0
5.0
4.5
4.0
3.5
3.0
2.5
2.0
5.0
3.6
3.4
3.2
3.0
2.8
2.6
2.4
2.2
2.0
3.8
3.6
3.4
3.2
3.0
2.8
2.6
2.4
2.2
2.0
3.8
HighMediumLowHighMediumLow
Study 2Study 1
Figure 5.1. The 12 studies from Förster and Denzler (2012), showing unusually linear patterns
of effects between the local (“low”), control (“medium”), and global (“high”) conditions.
The paper has since been retracted on the grounds that the data must have been
manipulated. Reprinted from the University of Amsterdam internal report.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
110 | Chapter 5
raw data. Following an internal investigation, an executive board at the
University of Amsterdam recommended that Expressions of Concern be
placed alongside the three journal articles. However, according to the
whistle-blower, the board fell short of rendering a verdict of misconduct on
the grounds that, because the raw data had been destroyed, there was insufficient
evidence to conclude that the results were due to fraud as opposed
to questionable research practices.32 Note the contrasting judgments of Tilburg
University in the Smeesters case and the University of Amsterdam in
the Förster case: while Tilburg University considered deliberate exploitation
of questionable practices to be misconduct in the Smeesters case (worthy
of retraction and leading to resignation), the University of Amsterdam
judged that comparable practices by Förster fell short of misconduct.
Förster’s whistle-blower disputed the conclusion of the University of Amsterdam
and escalated the case to the Dutch National Board of Research
Integrity (LOWI). On condition of anonymity, he told me why he decided
to take matters to a higher level. “I didn’t feel [the university’s decision] was
a very smart move because the data were practically impossible and the raw
data were gone, and the only outcome could reasonably be to retract the
papers.” When the case eventually surfaced publicly in 2014, the LOWI had
delivered a more damning verdict than the University of Amsterdam. Following
consultation with additional experts, the LOWI report concluded
that there was sufficient evidence of research misconduct. “According to
the LOWI, the conclusion that manipulation of the research data has taken
place is unavoidable,” it stated. “The LOWI found that the variability in the
scores for the control group is so unlikely small that this cannot be explained
by sloppy science or QRP [questionable research practices]; intervention
must have taken place in the presentation of the results of the experiments.”33
The LOWI implied that the responsibility must lie with
Förster, ruling out “the possibility that the many research assistants, who
were involved in the data collection in these experiments, could have created
such exceptional statistical relations.”
Förster strenuously denied engaging in either misconduct or questionable
research practices. Writing on his blog in 2014, he accused the LOWI of a
“terrible misjudgement” and perpetrating a “witch hunt.”
I do feel like the victim of an incredible witch hunt directed at psychologists
after the Stapel-affair. Three years ago, we learned that Diederik
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 111
Stapel had invented data, leading to an incredible hysteria, and understandably,
this hysteria was especially strong in the Netherlands. From
this point on, everybody looked suspicious to everybody.
To be as clear as possible: I never manipulated data and I never motivated
my co workers to manipulate data. My co author of the 2012
paper, Markus Denzler, has nothing to do with the data collection or
the data analysis. I had invited him to join the publication because he
was involved generally in the project.34
How, then, did Förster explain the unlikely linear trends in his data?
Without providing any clear answers, he appealed to the possibility that the
linearity might be a genuine phenomenon, but that he also couldn’t rule out
fraud by one of 150 research assistants he claimed helped collect the data.
“Let me repeat that I never manipulated data,” he said. “However, I can also
not exclude the possibility that the data has been manipulated by someone
involved in the data collection or data processing.”35 Crucially, he also admitted
to engaging in scenario #2 discussed earlier, selectively publishing
experiments that supported his hypothesis. “Coworkers analyzed the data,
and reported whether the individual studies seemed overall good enough
for publication or not,” he wrote. “If the data did not confirm the hypothesis,
I talked to people in the lab about what needs to be done next, which would
typically involve brainstorming about what needs to be changed, implementing
the changes, preparing the new study and re-running it.”
Could selective publication of experiments that “worked” lead to the
extraordinary pattern of results in Förster’s three challenged publications?
Might they form a subset of a much larger file drawer of unpublished results?
Although the University of Amsterdam committee concluded that
this, among other questionable practices, was a possibility, the whistleblower
disagrees. Even selective publication of 40 experiments among hundreds
or thousands would still leave odds of “just one in millions” that the
published data is genuine. “On top of that, it would require an extreme
amount of effort and an extreme amount of experiments and massive numbers
of students.” Even if such practices were the underlying explanation,
they would still warrant retraction of Förster’s articles because the data
underlying those papers would be heavily biased. More to the point, it is
interesting to note that selectively publishing experiments that supported
the desired hypothesis—a practice Förster admitted to using—was deemed
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
112 | Chapter 5
insufficient to demand a resignation when comparable practices at the level
of the data were enough to see the demise of Smeesters.
Unlike Smeesters, at the time of writing Förster remains in the academic
system, holding a position at Ruhr-Universität Bochum in Germany. In 2015
he returned a prestigious €5 million grant from the Alexander von Humbolt
Foundation that was awarded during the LOWI investigation, writing on
his blog that in the wake of his “conflict” that “the organization of a
[€]5-Million-project including 50 co workers is impossible.”36 Förster claims
to have had an epiphany while hiking in the mountains. “I will spend the
rest of my life on being rather than on having,” he wrote. “Thus, I will
leave the materialistic and soulless production approach in science. And I
want to say ‘Adieu’ to 10 cruel years, in which my life was almost completely
determined by others. I am going my own way now.” Later that year an
independent attempt to replicate Förster and Denzler’s retracted 2012 study
failed rather spectacularly. Author Marc Dressler concluded: “the failed
replication, together with the uncovered discrepancies, are not suitable to
relieve Förster from suspicion of fraud.”37
When Junior Scientists Go Astray
Acts of malfeasance by high-profile professors will always make a splash.
Stapel, Smeesters, and Förster were all successful researchers holding academic
positions at prestigious institutions, and their fall from grace has
been extensively publicized, combed over, even glamorized.38 But what of
lesser-known fraud committed by more junior researchers? Such cases may
be less newsworthy, but they reveal how misconduct penetrates all levels
of the academic hierarchy, and how it can be a tempting strategy for earlycareer
scientists facing uncertain futures and heavy competition.
Few cases exemplify the moral dangers faced by junior researchers more
than the case of PhD student Wouter Braet from the University of Leuven
in Belgium. Braet became the subject of a misconduct investigation in 2013
when members of his lab discovered serious errors while reanalyzing the
fMRI data from one of his previously published papers. The paper in question,
originally published in 2012 in the leading journal Cerebral Cortex, was
subsequently retracted. According to the retraction notice: “[I]t turned out
that the original data analyses by the first author [Braet] included several
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 113
operations which are hard to replicate and which do not fit fully with the
methods as agreed upon with the co-authors and as described in the paper.
Because of this we no longer consider these results trustworthy.”39 The
same problem was subsequently identified in a second paper, published in
NeuroImage, which was also retracted.40
At first glance, Braet’s case might appear as an honest error, or series of
errors, rather than premeditated misconduct. However, according to lab
head Hans Op de Beeck (and confirmed by the official University of Leuven
investigation), while Braet’s actions began as honest exploration of the data,
they escalated to fraud when Braet deliberately allowed errors in his analyses
to remain uncorrected because they generated attractive results.41 “He
never explicitly made the choice to falsify data, in the sense of getting to
work one day and saying ‘now I am going to manipulate my data, all the
way,’” Op de Beeck speculates. “It was more a case of diverting from the
proper procedures here and there, in the beginning in a rather exploratory
way and later on more systematically, then willingly keeping the errors after
finding out that things only worked that way, and then finally not disclosing
this to me or other people in the lab.”42
Like many young scientists, Braet appeared ambitious and strove to publish
as many papers as possible in prominent journals. Op de Beeck believes
that the pressure Braet felt to get the “right results,” together with his lofty
career ambitions, fueled a temptation to cheat that began with subtle bias
and later escalated into outright deception. “At the time that he made the
worst decisions he was getting in data for a control experiment during the
process of revising a paper for a high-impact journal,” Op de Beeck recalls.
“The revision would have been a no-go if he would have analyzed the data
without any manipulation. He really needed this paper for his career. He
wanted to get a faculty position—an ambition that I supported—and he was
even very selective about which places would be good enough for him,
which was an ambition that I was trying to tone down because it was
unrealistic.”
Braet himself claimed that his lurch into fraud was motivated by the
“publish or perish” culture of science. Speaking about his case to the Belgian
newspaper De Standaard, he said: “The publication pressure weighed heavily
on me. My academic career wouldn’t go anywhere if I did not write enough
papers. This pressure has led to the errors.”43 Op de Beeck believes that the
temptation for a junior scientist like Braet to commit fraud can be matched
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
114 | Chapter 5
by pressure on supervisors to cover it up, particularly when the supervisor’s
own career is uncertain. “Imagine if I had been under the same career pressure
as Braet when I found out about what he did,” Op de Beeck explained.
“Imagine if I didn’t have tenure yet but was also on a temporary contract
and would have been in the process of applying for jobs here and there. It
would have been even harder to make this decision [to report Braet’s case]
than it already was.” Op de Beeck is adamant that avoiding conflicts of interest
is essential to preserving research integrity. “The more that important
decisions are made by people for whom less is at stake, the better.”
The Braet fraud shares many characteristics with other cases of junior
misconduct. In the same year as the Braet case it was revealed that Adam
Savine, a PhD student with Todd Braver in the Cognitive Control and Psychopathology
Laboratory at Washington University, had fabricated results
in three published papers and six conference presentations.44 Savine was
found out after Braver noticed discrepancies in Savine’s data files shortly
before he was due to complete the oral defense of his PhD. Like Braet, Savine
swiftly confessed to falsifying his findings. His PhD defense was cancelled,
and he was reported by Braver for violation of academic integrity.45
Both Braet and Savine left academia, but what impact did their actions
have on Op de Beeck and Braver? Op de Beeck says he isn’t aware of any
negative career consequences. “People that know the case well seem to be
convinced that I did what I had to do and even see it as a very positive example
of what should be done more often,” he explained. “The internal
university investigation did not blame me or the lab, and promotion committees
or other internal committees do not seem to look at it negatively.”
Nevertheless, he has changed his lab practices in the wake of Braet to emphasize
collective responsibility. “It is now made more clear that data are
the property of the lab, not of an individual person. We found out about the
earlier fraud by a reanalysis of the data, so that makes it very obvious that
data might be checked later. We now have a more systematic back-up and
archiving system, and the co-workers know that both I and my research
manager have access to these data.” He has also overseen a cultural change
in his lab in the way experiments are planned and analyzed. “As a lab we
tend to be more clear now on what sort of decisions should not be taken by
one person in isolation and should be discussed in a group, or at least by the
main researcher together with me. Decisions such as the number of subjects
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 115
to scan, whether or not a subject should be excluded from the analyses,
whether we should rely upon analysis A or B, can be quite difficult to make
and are prone to experimenter biases of all kinds. We already discussed such
matters before the Braet case, but now we do it more consciously and more
systematically.”
For Op de Beeck, his experience proved how vulnerable senior investigators
can be to dishonest practices under their very noses, committed by their
most junior charges. “The case has taught me that everything we do is based
upon trust,” he explained. “I and principal investigators like me can detect
honest mistakes, but we are easy to fool.” Nevertheless, while Op de Beeck
agrees that greater transparency in research practices could help prevent
fraud, for instance in allowing easier reanalysis of existing data, he is concerned
that knee-jerk reactions to fraud cases could lead to draconian and
ineffective regulation of the scientific community. “Even very tight rules,
which would be cumbersome for the many honest scientists, would probably
still allow the exceptional dishonest scientist to commit fraud if he or
she really wanted to,” he said. “I remain convinced that we have to start
with trust, and implement practices that help the mostly trustworthy scientists
do their job as well and as reproducibly as possible, without frustrating
them with bureaucracy.”
For Braver, the consequences of the Savine fraud have been more profound.
At the time the case was initially publicized in 2013,46 he announced
publicly that he was committing his laboratory to practices that
can better detect and deter acts of misconduct.47 Two years later, I caught
up with Braver to find out what reforms he put in place. “The main changes
have been to better standardize, centralize and make transparent the documentation
and archiving process for each project,” he told me. “Another
change that we are starting to push hard, and transition toward, is adopting
the Open Science Framework and principles for our projects.” Braver
is enthusiastic about open science practices, and in particular data sharing.
“I am strongly encouraging each person in our lab to think about using the
Open Science Framework as a repository for raw data and analysis files, so
that we can make these public when a paper is published. It is a bit more
challenging with neuroimaging and other cognitive neuroscience studies
involving large amounts of complex data, but we are looking into various
solutions.”48
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
116 | Chapter 5
Like Op de Beeck, Braver says that he isn’t aware of any detrimental effects
on his career of having (unknowingly) harbored a fraudulent researcher.
However, soon after the Savine case broke, the Society for Neuroscience
temporarily suspended his membership on the grounds that,
despite being found innocent of any misconduct by the US Office of Research
Integrity, he nevertheless shared a collective responsibility for the
fraud. The society’s ethical policy states that it “reserves the right to impose
sanctions and/or to take corrective actions, regardless of whether intentionality
is demonstrated during an institutional investigation.”49 The policy
does not elaborate the specific ethical justification for censuring innocent
PIs [principal investigators] or coauthors—a move that smacks of collective
punishment. Furthermore, it is difficult to see how such a policy achieves
anything beyond dissuading PIs from transparently exposing fraudulent
behavior in their own labs.50
The cases of Braet and Savine remind us that acts of fraud are not limited
to the most senior professors in science; indeed, the temptation to cheat
may be just as powerful, if not more so, among junior researchers facing
stiff competition for career progression. Had Braet and Savine managed to
avoid scrutiny, it is foreseeable that they would have followed the footsteps
of Stapel or Smeesters, so it is ultimately beneficial that they were detected
early. At the same time, it is clear that for Braver and Op de Beeck, discovering
fraud within their own labs was one of the deepest betrayals that can
happen in academic life. In 2013, Braver described his feelings at uncovering
Savine’s fraud as “shocked, angry, and incredibly disappointed . . . the
worst kind of violation of my trust, time investment, and efforts towards
Adam’s mentoring. Being a graduate supervisor is in some ways akin to
parenting, so it was like finding out that your child had just skipped town
after robbing you and your neighbor’s house.”51 Op de Beeck offers a similar
sentiment: “On the day I made the final decision to retract the papers and
(the inevitable consequence) to go public, I was so upset that I got my car
into an accident in the parking lot. The scratches and dent are still very
obvious. Nevertheless, the car still drives—a very appropriate metaphor of
what happened to me.” The silver lining is that Braver and Op de Beeck’s
swift and transparent responses left them with no lasting career damage.
On the contrary, their experiences led to renewed laboratory practices that
not only reduce the likelihood of fraud but also help reduce everyday sources
of research bias.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 117
Kate’s Story
Fortunately for both Braver and Op de Beeck, their discovery of fraud
within their own ranks left their own careers shaken but intact. However,
when this power relationship is reversed—and it is the junior scientists that
uncover misconduct by their seniors—the consequences are more extreme.
As we saw earlier, the three young researchers who exposed Stapel were
crushed by his fall, either leaving academia or taking up positions at relatively
unknown institutions. In other cases, junior whistle-blowers have
faced excommunication while the high-ranking perpetrators they identify
escape virtually untouched.
The story of Kate illustrates the risks faced by scientists who are courageous
enough to identify and challenge senior misconduct. Kate, whose
identity has been changed to safeguard her anonymity, spoke to me about
what happened when she became involved in a fraud investigation directed
at her line manager, and the profound consequences it had for her life and
career. The following is based on Kate’s account of what happened.52
At the time her story begins, Kate was a postdoctoral researcher working
for a PI in psychology, based at a major university. Her work with the PI
focused on behavioral changes in people who had suffered brain damage
and the first year of a three-year position with the PI had gone smoothly.53
“He was really nice at first, really wonderful,” Kate told me. But problems
began to emerge during the second year of the job. At the time, the PI was
in the process of conducting a single case study in a patient with an interesting
type of brain damage. Even though Kate herself wasn’t involved in
the experiment, the PI had involved several junior members of the research
group, including three assistants who were junior to Kate.
Together, the first assistant and the PI had tested the patient using a
customized set of equipment, which needed to be positioned in a certain
orientation for the experiment to provide a fair test. After it was finished,
however, the assistant realized that the equipment wasn’t set up in a way
that would have allowed such a test, which meant that the results the PI
produced (and which happened to perfectly fit his hypothesis) didn’t make
sense. In the meantime, as part of writing up a scientific paper based on
those results, the PI had separately instructed the second assistant to prepare
a visual graphic of the equipment. The second assistant duly prepared
an illustration of how the equipment was actually oriented in the experi-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
118 | Chapter 5
ment (i.e., misaligned) and, upon seeing this, the PI instructed her to instead
draw it as it ought to have been positioned for the results to be interpretable.
When the first and second assistants then pointed out this error
to the PI, Kate says that he brushed their concerns aside.
At this point all three assistants decided something fishy was going on.
One by one they approached Kate for advice. “They each came to me separately
and said he was doing something they felt uncomfortable with,” she
explained. “I told each of them that they needed to talk to each other to get
a full picture of what happened. So they had several meetings together, and
then with me again, and it became clear that the PI was indeed being dishonest.
There was a lot of anger that this could happen because we had
never come across anything like it before. I naively thought that if we simply
took it to a higher authority then it would be taken care of. So I advised
them to see the head of our department.” Kate now doubts whether that
was the right decision. “In retrospect, this was stupid” she adds. “With everything
that happened, what I should have said was drop it, resign, go
away.”
When the three assistants approached their department head, he told
them that if they explained in detail what had happened he would have to
raise it with an even higher authority. “After the meeting, they came back
to me and said that he had listened,” Kate recalls. “But he had also said, are
you sure you want to go ahead with this? He wasn’t trying to talk them in
or out of anything—he was just giving them the options. They decided to
pursue it, and to his credit, he raised it with the head of the university.”
Together the assistants submitted a report to the head of the university that
documented their concerns, and, with Kate’s help, they were able to provide
additional evidence, gleaned from the data itself, that the equipment had
been wrongly oriented. A short time later the PI was informed that he was
under official investigation for misconduct.
It was at this point that Kate’s relationship with her PI deteriorated
sharply, despite the fact that she was not one of the official complainants.
“He directed all his anger at me. He accused me of orchestrating a campaign
against him, of encouraging the assistants to make the complaint. At the
same time, he decided he was going to terminate my employment and give
my job to a new employee.” Kate is convinced that the PI had kept her at
arm’s length throughout the experiment so that he could more easily get
away with fabricating the results. “I was part of several experiments, but
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 119
I’m sure he kept me out of this one because he wouldn’t have been able to
fudge anything if I had been involved. He only included the more junior
people—the ones he could manipulate more easily—and even then he was
very careful, only giving each one a limited view of the overall picture. That
way he could exert his authority more easily. But he underestimated them,
and he never anticipated that they would come to me.”
Kate recalls that, while the complaint was being considered, the junior
team were also placed under a gag order by the head of the university. “He
told them that it was now a legal matter and they must not talk to anyone
about it, not even privately. As a half-joke, I used to say that he hasn’t specifically
told me that. But of course I remained silent too because I didn’t
want to bring the department into disrepute. We knew that if we spoke out,
people who had nothing to do with the case would suffer. And we believed
that everything would be ok because there was going to be procedural justice.
We would keep our mouths shut and in return the university would
stop him from doing this again.”
Kate’s faith in the system was misplaced. When the university eventually
reached an official decision about the PI’s behavior, it fell short of labeling
it misconduct, concluding only that he had engaged in “sloppy research
practices.” He faced no formal reprimand, and no mention of the investigation
or its outcome was published. The university’s only requirement was
that he repeat the disputed experiment. “We were utterly shocked and appalled,”
Kate says. “By this time each of the junior workers had left to pursue
different paths and careers. This left just him and me, that is until he could
get rid of me.” With the departure of his junior team, the PI chose to rerun
the experiment alone. “Given the evidence we presented, I found it extraordinary
that the university administration was happy for him to do this,”
Kate said. “Nobody else was there to monitor what he said to the patient.
Nobody else inspected the data.” When the results of the experiment
emerged, they conformed perfectly to the hypothesis. The manuscript was
swiftly accepted for publication in a well-respected journal, where it
remains.
In the final months of her second year, Kate felt under constant attack.
“I got a lot of bullying from the PI, a lot of petty behaviour. He wouldn’t let
me finish experiments. He didn’t want me back in the lab. In his mind he
had done nothing wrong—he was completely and utterly justified and completely
innocent of any wrongdoing. His retribution against me was merely
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
120 | Chapter 5
his way of restoring justice.” The PI sacked Kate at the end of her second
year, and her department was unable to safeguard her employment. With
few labs conducting similar research, and few jobs available, she was faced
only with hard choices. “The case had taken a great toll, both emotionally
and professionally. My choice was to cut and run and only get one publication,
or to try and hang on and get more. So I decided to try to go for as long
as possible, but obviously that was making it really hard for me.”
With the help of a colleague, she was able to take up a temporary research
position at a small nearby university to continue her research. Meanwhile
she applied on several occasions for faculty jobs back in her original department
and each time was unsuccessful. Eventually, after several attempts,
she learned that a new head of department had repeatedly blocked her from
being shortlisted. Kate is sanguine about this revelation. “At the time it was
hard to accept, but in retrospect I can understand this because as a head
you have to try to keep a lid on things. Looking back, I was extremely naïve
to apply for jobs back there.”
After more than seven years of intermittent short-term contracts—and
no career progression—Kate finally steered her career back on track by taking
a position at a different university. A few years later she then moved to
a larger institution, where she currently holds a tenured position. “It took
me ten years to get back to where I was,” she explains. “And I was even relatively
fortunate compared to how it might have played out, because I published
several papers during my time with the PI, which means that that
period in my career doesn’t stand out as an unproductive hole. And I was
lucky to know enough senior researchers who could write job references for
me, so I didn’t have to rely on the PI for those either.”
As for the PI, Kate later learned that he was encouraged by the department
to move on, and was nudged quietly toward the door by being demoted
from a tenured (permanent) position to a probationary one. Within two
years of the investigation concluding he had left the university for a post
overseas and to this day remains employed as a senior professor in a major
institution. In the years to come Kate came to believe that the PI’s behavior
was not an isolated incident. While she admits she has no hard proof, she
is certain he is a repeat offender. “I have very strong suspicions that he
fudged a paper in a major generalist science journal and there was a later
paper he was working on with another researcher, who happened to be a
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 121
close friend of mine, where the data was strange too. She [the researcher]
and the PI had sent off the paper and had it reviewed when she pointed out
her concerns. They were told by the journal to re-run the experiment and,
sure enough, the PI produced a perfect experiment with identical data. But
the researcher herself didn’t collect the data. She was closely involved with
every part of this project but has no idea who re-ran that experiment—it
just popped out of thin air. I advised her to leave it alone, which she did,
because her contract was ending.”
What would Kate’s advice be for junior scientists who suspect fraud by a
senior researcher in a power relationship? “I would just say, resign, go away,
let it go. If I could go back and do it again, I would have finished my own
experiments and then I would have resigned and started looking for different
jobs.” At the same time she is adamant that senior malfeasance must be
reported. “There is a difference between reporting it and following it
through. The key is that this person—the junior person—should resign.
They must resign.” The whistle-blower in the Förster case offers similar
sentiments. For him, the process took a heavy personal toll, and he offers
little to recommend his choices to others. “In hindsight you would say that
if you have really clear evidence of misconduct then go for it, but if you think
the data are fake but lack definitive evidence, given the system as it is now
it is a waste of time.”
Even though methods for exposing fraud have advanced since Kate’s experience,
she believes psychology has a long way to go in adequately addressing
the problem. “There have been improvements in stopping it but
there is still too much at stake for everyone. Until whistle-blowers are fully
protected and institutions are fully transparent, powerful people will continue
to get away with it. If they are going to be stopped then it can only be
those above them—their line managers and those in the highest positions
in the university—that can do it.” She also believes that the best way to save
junior scientists from her ordeal may be to eliminate fraudsters earlier, as
in the cases of Braet and Savine. “We need to be detecting these people
when they’re young, before they have too much power and before their
exposure as frauds causes so much collateral damage. We need to train
senior researchers to weed them out from the pool, from postgraduate levels
onwards. And we need to be honest when giving references. This would
take a whole generation and a major culture shift.”
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
122 | Chapter 5
The Dirty Dozen: How to Get Away with Fraud
If fraudulent research psychologists were to compile a playbook on how to
beat the system, how would it read? We can glean at least a dozen “top tips”
for the aspiring cheat.
1. Take time to read the literature. It contains the material that will
enable you generate clever hypotheses and fake data that are
both plausible enough and striking enough to impress journal
editors and reviewers.
2. Switch raw data between conditions as needed to fit the pattern
you need—for example, if you need the measure for condition A
to be significantly larger than condition B, make sure that subjects
who show large opposite results are switched. Switching is
a preferable method to fabricating because the raw data structure
is preserved and will carry fewer signs of mishandling. But
don’t switch every data set that runs against your hypothesis (see
#3)—the results need to be believable.
3. When fabricating or switching, never make the results too perfect.
Make them good enough to clearly support your hypothesis
with low p values that survive stringent multiple comparison
correction (this will help create the illusion that you are a rigorous
and trustworthy researcher who understands statistics). Remember
that squeaky-clean data screams of manipulation and
could attract the attention of “data detectives”; passably good
data looks more genuine and will usually go unnoticed.
4. Before using your fabricated data, be sure to test it against the
fraud detection tools developed by Uri Simonsohn and others.
Remember not to get caught out by basic human fallacies such
as the law of small numbers. Check the subconditions that are
independentofyourhypotheses(e.g.,menvs.womenoryounger
vs. older) for strange trends that could expose manipulation—
it’s easy to forget these in haste.
5. Closely guard access to your data. Never share it with outsiders
unless forced to do so. Send data to collaborators only if they
specifically ask for it, and even then be sure to attach conditions
of use. Regardless of who it is, never send more data than abso-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 123
lutely required—remember that many cases of fraud are discovered
by the fraudster’s close colleagues or students.
6. Use outright fraud sparingly—remember that the best lies are
mostly true. Fabricate only the results you need in order to secure
a high-impact paper or grant. Fraudsters get caught when
they get greedy or complacent. If an honestly conducted experiment
disconfirms a fraudulent one, appeal to a hidden moderator,
or better still, bury the study in the file drawer where it
belongs.
7. Where possible, avoid fabricating results that are likely to be the
target of an exact replication. Replications are fortunately rare
so this won’t be a common problem, but nevertheless be aware
that nonreplications by independent groups can draw negative
attention and risk exposing you. If, as is sometimes required,
you must fabricate results to secure a high-impact paper, make
sure the finding is exactly what everyone would have predicted
but shown in a clever way—that way, everybody will believe
your results and nobody will bother trying to replicate them.
8. Keep your published method sections nebulous on precise detail
and ignore e-mails or respond only vaguely to questions from
peers about specific methodological practices. Maintaining a
certain degree of technical ambiguity will make it harder for
other scientists to attempt exact replications of your work, furnishing
you with a convenient cover story in case independent
researchers fail to reproduce your fake results. If what they find
conflicts with your study, you can always discredit them by
pointing to some critical aspect of their method that was different
from and inferior to yours, even if it wasn’t stated explicitly
in your paper. Politely decline any attempts to collaborate with
“replicators,” especially adversarial collaborations where methods
are preregistered. These are extremely dangerous, and you
should simply say that you are too busy with your own research
program to engage in such projects. On the other hand, be sure
to note where someone appears to successfully replicate your
fraudulent studies (whether a direct replication or, as is more
likely, an indirect one). Provided your findings are sufficiently
prominent and in line with what others want to see, confirma-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
124 | Chapter 5
tion bias by other researchers will ensure that at least some of
your fake findings are replicated.
9. Take on a lot of interns and unpaid research assistants, dividing
their responsibilities so that no one individual sees sufficient
workings behind a single experiment to detect any fakery. Keep
as few records as possible about who these people are and when
they worked in your lab. That way if your data is ever questioned
you can pin the fraud on previous lab members, but at the same
time your poor record keeping will protect you from having to
name anyone in particular.
10. To create an additional smokescreen, write the occasional paper
on research practices and, as your career progresses, be sure to
take up positions of influence in earnest domains such as ethics
and methodology. Don’t spend too much time on these (they are
only a cover), but building a reputation in sober topics will help
frame you as an unlikely exponent of fraud, which in turn will
help dissuade potential whistle-blowers from exposing you. At
the same time, be sure to oppose greater transparency in research
practices on the grounds that science is working just fine,
thank you, and has for hundreds of years. Transparency is a
threat to your way of life.
11. If ever accused of misconduct, immediately delete your raw data
to impede any investigation and dismiss the competence and
motivations of your accusers. Where possible, threaten your institution
with legal action should they decide to go public. Bully
any junior whistle-blowers into silence, and try to draw the sympathy
of your superiors and contemporaries by throwing around
terms such as “witch hunt,” “envy,” and “dark forces.” Point out
that your work has been repeatedly replicated by many others,
but without giving specific examples that could be undermined.
By making you and your data as untouchable as possible, there is
a better chance your university will try to minimize or bury the
allegations. And whatever happens, always deny everything.
Confess and you will end up on the academic scrap heap.
12. In moments of doubt when you wonder why you are doing all
this, remember that science is nothing more than a competition
for status in a field of storytellers. You are doing what the system
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Corruptibility | 125
requires of you, and, in any case, you have heard of others who
are committing far worse infractions than you. After all, it’s not
as though you are really hurting anyone. Does it really matter if
any of your specific results are true or false? Who honestly cares?
All findings are to some extent wrong—as a colleague once
wrote, the best we can hope for is to be “interestingly wrong.”
Science always self-corrects in the long run.
Were psychologists determined to combat fraud then as a community
we could do it. But even speaking about the F word, let alone tackling it, is
close to taboo. In one hand we brush off fraud as the actions of a few bad
apples—a distraction that threatens the public image of science. And in the
other hand, we hail the benefits of trust within an honesty system, but
without establishing robust systems for detecting fraud and without protecting
trustworthy and honest researchers from serial offenders.
How can we bring fraud under control? In chapter 8 we will consider
some of the proposed solutions to fraud, including standardized open data,
random data audits by an independent authority, and the promise of criminal
sanctions for identified perpetrators. We will also return to the ultimate
method for weeding out all scientific error, including fraud: direct replication
by independent researchers.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
C H A P T E R 6
The Sin of Internment
Publish means “make public.”
—Mike Taylor, 2012
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
As scientists we celebrate our willingness to share new ideas—after all,
science as we know it wouldn’t exist without a rich culture of exchange.
Yet in chapter 4 we saw how many psychologists reject a key element
of this philosophy, refusing to make their data available for scrutiny
or reuse by other scientists. Intermingled with this “sin of data hoarding”
is an equally worrying, albeit more complex, culture of concealment. Not
only are the data supporting research experiments usually unavailable, but
the articles that psychologists publish in peer-reviewed academic journals—articles
that report the conclusions and implications of their research—are
usually published behind subscription paywalls. Publishing
behind a paywall means that the public who funded the research in the
first place can’t read it unless they are prepared to pay again, forking out
$30 or more to view single articles, or paying even more to take out personal
subscriptions.
Rather than insisting that their research articles are made publicly available,
the psychological community has for a long time been content with
this system of barrier-based publishing. Most academic psychologists in
Western countries cope with access restrictions by taking advantage of subscriptions
paid by university libraries (and thus by the public). This allows
the academics, but not the public, to read each other’s articles, although as
we will see not all universities can even afford to furnish their academics
with subscriptions. Most importantly, relying on a barrier-based system
diminishes the act of publishing from being a means to publicize research
outcomes to being an exclusive communication between academics—telegraph
lines between the windows of the ivory tower.
This restrictive system is by no means unique to psychology; barrierbased
publishing also dominates many other scientific fields. Yet psychology
has as much, if not more, reason to oppose such antiquated forms of dissemination.
Psychological discoveries generate substantial public interest,
are relevant to policy making, and are hugely dependent on public funding.
For failing to embrace the culture of open access that the public and nonacademic
users require, psychology is thus guilty of our sixth major transgression:
the sin of internment.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
128 | Chapter 6
The Basics of Open Access Publishing
The Budapest Open Access Initiative defines open access (OA) publishing
as “free availability on the public internet, permitting any users to read,
download, copy, distribute, print, search, or link to the full texts of these
articles, crawl them for indexing, pass them as data to software, or use them
for any other lawful purpose.”1 This mission seeks to democratize research,
making it transparently available to the public whose taxes fund academia.
OA publishing also benefits a host of other societal groups, ranging from
journalists and science communicators who require timely access to the
primary scientific literature, to businesses who can the use latest evidence
to drive innovation, to students and professional bodies (such as medical
practitioners) needing to learn and apply knowledge, to civil servants seeking
to translate scientific discoveries into public policy. When OA is implemented,
it generally follows one of two major routes, known as full OA and
hybrid OA.
Full OA. The simplest and most accessible OA model is one in which all
articles published by a journal are made freely available under a Creative
Commons Attribution license (CC-BY) that allows readers to “copy, distribute,
display and perform the work and make derivative works based
on it only if they give the author or licensor the credits in the manner
specified by these.”2 Most fully OA journals can be found to operate in
one of two modes: either offering an entirely free service to both authors
and readers (generally subsidized by an institution such as a university or
library), or by charging authors a fee for publishing through article processing
charges (APCs)—the so-called gold OA model. In either case, the
cost burden associated with publishing is shifted from readers to authors.3
Examples of fee-based OA journals include PLOS ONE and PeerJ, and
within psychology specifically, Archives of Scientific Psychology and BMC
Psychology.4
Hybrid OA. Under hybrid OA the journal employs a traditional subscription
model, but some articles that it publishes will be OA while others
appear behind paywalls. There are various paths by which an article in a
hybrid journal can become OA. For some outlets, such as the Journal of
Neuroscience and the Proceedings of the National Academy of Sciences USA,
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Internment | 129
articles are initially published behind a paywall but made freely available
to readers after an embargo period (e.g., 12 months), at no additional cost
to authors. Alternatively, upon having their article accepted for publication,
authors can choose to pay an additional fee to the publisher to make
it immediately OA. This gold OA route is analogous to the APC that would
be charged for a fully OA journal, and despite the fact that the publisher
also collects subscription revenues it typically costs at least 50 percent
more. Finally, some publishers allow the author to freely archive their accepted
manuscript in a public repository—the so-called green OA route—
although most publishers permit archiving of only the accepted postprint
version (e.g., produced with word-processing software) rather than the
typeset article that appears in the journal itself.
Just how committed are psychology journals to these OA models? Most
of the more prominent psychology journals—that is, the journals that psychologists
regard as desirable and career advancing—now offer a form of
hybrid OA involving both green and gold options. For instance, journals
published by the American Psychological Association and the Psychonomic
Society offer fee-based gold OA and, in most cases, permit immediate green
OA of the author’s postprint but not the publisher’s typeset version. Other
publishers, such as Elsevier, also offer gold OA together with a more restricted
green OA in which archiving of the author’s postprint within an
institutional repository is embargoed for at least 12 months (e.g., the journal
Cortex).5 Crucially, very few psychology outlets offer a full OA model,
and none of the most prominent and prestigious psychology journals offer
a model that is immediately and permanently free for both authors and
readers.6
Why Do Psychologists Support Barrier-Based Publishing?
Psychology is notable for its tolerance of barrier-based publishing.7 By comparison,
it is standard practice in many physical sciences to archive both
preprints (prior to peer review) and accepted postprints (following peer
review) in the free repository arXiv.org, which has now been established
for 25 years and receives thousands of submissions per month.8 While physicists
still publish in and support peer-reviewed journals (many of which are
linked to learned societies), the policies and copyright licenses of these
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
130 | Chapter 6
journals have been shaped around arXiv.org to serve the needs of the community,
leading to more transparent and accessible publications. In psychology,
however, the practice of archiving preprints and postprints is still rare,
even when doing so is fully permitted under the publisher’s policy. So what
explains the reluctance of psychologists to endorse a more open approach
to publishing?
One possible reason why psychologists tend not to archive preprints (despite
this being permissible by virtually all publishers) stems from the low
reliability of psychological research discussed in chapters 1–3. In psychology
it is not unusual for reviewers and journal editors to encourage (or even
require) authors to change their hypotheses to fit unexpected results and
then, in the interests of generating a clean narrative, to present the revised
hypothesis in the introduction of the manuscript as though it was predicted
from the outset. Although common, this questionable practice of Hypothesizing
After Results are Known (or HARKing; see chapter 1) is frowned
on, and since comparing the hypotheses in the preprint with the accepted
manuscript would swiftly expose any HARKing undertaken during review,
it is perhaps not surprising that few psychologists take the risk of publishing
their preprints.
Explaining the relative lack of fully OA psychology journals is more difficult
but is probably tied to cultural norms in assessing research quality. In
judging the value of published research, psychologists place great weight on
journal hierarchy and prestige—values that are monopolized by a small
number of prominent, subscription-based outlets. This heuristic for assessing
quality isn’t necessarily a good thing as the evidence linking journal
rank with the merit of individual articles is weak to moderate at best.9 But
here science is bad at being scientific: the actual quality of research plays
second fiddle to the perception of quality.
In psychology this system stands in the way of OA because very few of
the journals that are considered to be high ranking are also fully OA. For
instance, in cognitive neuroscience and cognitive psychology, some of the
most desirable journals are considered to be Nature, Science, Nature Neuroscience,
Neuron, Proceedings of the National Academy of Sciences, Current Biology,
Journal of Neuroscience, Psychological Science, Cognitive Psychology, and
the Journal of Experimental Psychology group. A quick glance at the CV of
any top psychologist in this area will probably reveal papers published in
this set of journals. And yet, how many of these journals are fully OA? Not
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Internment | 131
one. At Nature, authors don’t even have the option to pay to make their papers
freely accessible.
The “hierarchy of journals” hasn’t arisen by chance. Sometimes rank
evolves based on history and trust—journals such as the Journal of Experimental
Psychology group are among the oldest in psychology, having earned
a reputation for careful peer review and editing over many decades. Other
major outlets ride on the “bling factor,” in which prestige is engineered
through the selective publication of results that are novel and eye-catching
(publication bias). It is not unusual for high-ranking journals to reject 90
percent or more of the submissions received; and the more papers a journal
rejects, the more prestigious and desirable it becomes. This isn’t necessarily
because the rejected papers contained flawed science but because the results
are deemed unoriginal, boring, or difficult to hang a story on.
Some proponents of OA have argued for mass boycotts of these journals,
calling for scientists (including psychologists) to instead send their work to
PLOS ONE, PeerJ, or one of several emerging (albeit “lower-ranking”) OA
options.10 This has a certain appeal. After all, if all psychologists did this
en masse then the sin of internment would quickly be resolved. But, of
course, no grassroots movement ever happened en masse without a good
incentive, and in this case the incentive for psychologists to abandon traditional
subscription journals leaves much to be desired: it amounts to sacrificing
career opportunities and advancement, including promotions, grants,
and research fellowships, for the good of the cause. Publishing in fully OA
journals could even lead to scientists losing their jobs. At some UK universities
a trend is emerging for the continuing employment of academics to be
based, in part, on the quantity of publications they are able to generate in
“high-quality journals.”11 And because subscription journals, but not fully
OA journals, dominate the list of outlets considered to be “high quality,”
psychologists face active disincentives to publish their work in more accessible
ways.
Concerns about promotion and employability are particularly salient for
younger researchers who lack the security of permanent academic jobs.
However, while senior psychological scientists have more freedom to publish
in less prominent OA journals, the reality is that many continue to
reinforce the status quo, not only to maintain their academic influence but
to protect the careers of their junior protégés. To illustrate why, suppose
you are a professor of psychology supervising a promising PhD student who
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
132 | Chapter 6
you believe could one day become an academic leader. The student has
worked hard on a series of ground-breaking experiments and sends a paper
to PLOS Biology, the only high-ranking journal in your area that is also fully
OA. As often happens, though, even for excellent work, the paper is rejected.
What now? Do you resubmit the work to a lower-ranking OA journal, hoping
that the paper will be recognized for its merits, despite the fact that
many scientists at best ignore these journals and at worst ridicule them? Or
do you send it to a prestigious journal that sits behind a paywall, where the
paper’s merits will be boosted by an environment that is both prominent
among the desired readership and universally respected? Which is more
important, furthering open and transparent science or furthering the career
of the scientist whose fate lies in your hands?
This conflict between what’s best for science versus what’s best for scientists
is a real-life dilemma faced every day by senior psychologists who
run research groups. Anyone in such a fortunate position has already
learned, time and again, that job committees, grant reviewers, and funding
agencies are dazzled by articles in prestigious subscription journals. Who,
then, is surprised that psychologists at all career levels, from PhD students
to professors, refuse to sacrifice their own livelihoods—and those of their
protégés—on the altar of OA?
Hybrid OA as Both a Solution and a Problem
How can we change the incentive structure of academia to reward rather
than punish psychologists who embrace OA? In the last ten years a number
of initiatives have emerged to nudge psychologists toward greater openness.
Since 2005, the Wellcome Trust—the UK’s largest nongovernment funder
of scientific research—has led the way by requiring grantees to publish their
academic papers through either full or hybrid OA under a CC-BY license.
The trust also expects authors to deposit their manuscript in Europe
Pubmed Central, an online database that provides free access to published
biomedical research. For the first seven years, this policy was unenforced
and, predictably, suffered from low compliance rates with more than half
of grantees still publishing their articles behind paywalls.12 In 2012, the
trust clamped down in an effort to boost compliance, with authors who fail
to adhere to the policy now losing 10 percent of their grant funding; this
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Internment | 133
sanction could range from tens to hundreds of thousands of pounds per
grant.13 A review of compliance throughout 2013–14 found that despite
some improvement, still only 61 percent of grantees had fulfilled all aspects
of the agreement. Remarkably, the trust had also forked out nearly £500,000
in APCs that did not eventuate in publication of the manuscript in Europe
Pubmed Central.14 A later analysis in 2015 was slightly more positive: while
the trust had initially withheld funding for noncompliance on 134 occasions,
only 20 cases were unredeemable and subjected to sanctions.15 The
trust’s head of digital services, Robert Kiley, blames the low compliance on
complicated and opaque policies of publishers, particularly in traditional
subscription journals where the intricacies of hybrid OA can be most confusing
for authors. “These problems are particularly prevalent amongst publishers
offering a hybrid OA option, which . . . is also the more expensive
way to comply with our OA policy [they were 64 percent more costly],” he
wrote. “The Trust remains committed to its OA policy, but ultimately if we
are to be successful and get to the point where all Trust funded research is
OA, then the friction in the system—which is all too evident in this analysis—has
to be resolved.”
Recognizing the importance of OA, but also its many barriers and disincentives,
in 2012 the UK government commissioned an official enquiry
into the future of academic publishing. The report, led by Professor Dame
Janet Finch, was unequivocal in its support for the philosophy of OA, declaring
at the outset, “The principle that the results of research that has
been publicly funded should be freely accessible in the public domain is a
compelling one, and fundamentally unanswerable.”16 Notably, however,
it fell short of calling for step-change adoption of OA to address this “fundamentally
unanswerable” axiom, instead arguing for “a balanced package
of measures to be taken to increase access to research publications and to
accelerate the transition to open access publishing.”17 Specifically, Finch
called for widespread adoption of gold OA where authors pay to publish,
rather than green OA (self-archiving), outlining the need for “a clear policy
direction . . . towards support for publication in open access or hybrid
journals, funded by APCs, as the main vehicle for the publication of research,
especially when it is publicly funded.”18 Why the focus on gold
OA? Because, according to Finch, “[p]ublishers, whether commercial or
not-for-profit, wish to sustain high-quality services, and the revenues that
enable them to do so.”19 Indeed, the report clearly supports the commer-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
134 | Chapter 6
cial needs of the publishing industry in gatekeeping public access to
knowledge:
When funders and institutions began to develop policies to promote
open access, especially access via repositories, both commercial and
learned society publishers that publish subscription-based journals
tended to see them as a threat. Many such publishers saw the prospect
of a requirement that articles should be made available through
institutional and subject-based repositories, after what was seen as a
relatively short embargo period [green OA], as a threat to their revenues
and even to the survival of their journals, with the prospect of
sales falling as swift, free access became accessible via repositories.
Learned societies saw a threat to the publishing income that sustains
many of their charitable scholarly and public engagement activities;
and also to their income from members who are often attracted by
society publications as a membership benefit.20
Stevan Harnad, one of the original signatories of the Budapest Open
Access Initiative, issued a scathing assessment of the Finch report as a “successful
case of lobbying by publishers to protect the interests of publishing
at the expense of the interests of research and the public that funds research.”
He argues that the recommendation to focus on expensive gold OA
over free green OA was a costly mistake:
The Finch Report proposes to . . . push “Green” OA self-archiving (by
authors, and Green OA self-archiving mandates by authors’ funders
and institutions) off the UK policy agenda as inadequate and ineffective
and, to boot, likely to destroy both publishing and peer review—
and to replace them instead with a vague, slow evolution toward
“Gold” OA publishing, at the publishers’ pace and price.21
Harnad has long advocated green OA, and his call for its adoption makes
complete sense within the hybrid model where the publisher also collects
subscription fees. However, an exclusive reliance on green OA would also
eliminate the growing market of fully OA journals that charge APCs, such
as the PLOS and BMC family. Therefore, at least in the short term, Harnad’s
proposal to focus on green OA is unsatisfactory because it returns power to
the status quo of the traditional subscription publishers. In addition, Harnad’s
solution would require authors to accede to the embargo restrictions
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Internment | 135
of green OA imposed by those publishers, meaning that much academic
research would be unavailable to the public and nonacademic users until
6–12 months, or more, after it was initially published behind a paywall. In
the long term it is possible that community pressure would lower such embargos
or eliminate them altogether, which appears to be Harnad’s longterm
goal. On the other hand, since green OA is effectively subsidized by
subscription income, any reduction in subscriptions could lead publishers
to apply even greater restrictions on self-archiving in order to protect their
revenue streams.
At the same time, the reliance on gold OA advocated in the Finch report
is just as unsatisfactory because it applies equally to fully OA journals and
hybrid models. Why is gold OA a problem within hybrid models? Recall that
with fully OA journals, authors make one payment via an APC to have their
article published under a Creative Commons Attribution license (CC-BY)
that allows essentially any reuse provided the original work is attributed to
the authors. Under the hybrid OA model, however, gold OA—which is on
average 60 percent more costly in hybrid journals compared with fully OA
journals—adds to existing payments to great an absurd double- or even
triple-dipping into the public purse. Consider the farcical nature of the following
(completely typical) scenario for a hybrid psychology or neuroscience
journal. First a university pays a hefty annual subscription to a publisher,
which can amount to tens of thousands per year, per journal, and millions
across a “package” of journals. Some of these journals in psychology or
neuroscience will then apply additional levies such as submission fees (e.g.,
at the time of writing, the Journal of Neuroscience charges $140), charges for
color figures (reaching as high as $200 each), and per-page charges (e.g.,
$85 per page at the Journal of Neurophysiology). And then, on top of these
fees, the Finch report would call for the publisher to be paid yet again via
a gold OA charge to make the article publicly accessible upon publication—
with this fee alone reaching into the thousands per article. The absurdity
of this model becomes even more apparent when you consider that in psychology
and neuroscience, as in many sciences, the editors and reviewers
who coordinate peer review typically do so for little or no financial compensation.
What other business in society, other than a journal publisher,
is paid three times for a product that is produced almost entirely by someone
else?22 It is easy to see why publishers are content with the Finch report,
but it is difficult to see how it offers a sustainable long-term solution or a
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
136 | Chapter 6
justifiable expenditure of public funds. Indeed, it remains a mystery who,
other than the publishers themselves, would call the Finch proposal any
kind of realistic mechanism for making research openly accessible to the
public.
In 2013, Research Councils UK (RCUK)—the central body that coordinates
all six major UK research councils, including those that fund psychology
research—issued new rules for grantees that cut through some of the
controversy and confusion of the Finch report. Since April 2013, publications
arising from RCUK grants must now be made OA, either by a green
or a gold route.23 By offering authors the choice between self-archiving and
APCs, RCUK bypasses the limitation of green- or gold-only solutions advocated
by Harnad and Finch, respectively. However, while RCUK’s initiative
is well balanced, it is also insufficient on its own to open up psychology
research. For one thing, it tackles only UK-funded science, which makes up
a small fraction of the total global output. And even within the UK, much
psychology research is funded directly by universities (rather than through
grants) or by alternative funding agencies that do not enforce adherence to
OA. In spite of these limitations, there is no doubt that the RCUK model,
together with comparable policies launched by the NIH, NWO,24 the Higher
Education Funding Council for England,25 and European Research Council,26
are making important headway toward universal OA.
Calling in the Guerrillas
Changes in OA policies are vital, but, as always, top-down reforms can be
too slow and bureaucratic to provide immediate benefits for consumers. It
is perhaps ironic that the consumers who suffer most frequently from lack
of OA tend to be researchers themselves. Even working at a major Western
university, I occasionally find that my library will lack a subscription to a
particular journal holding a paper of interest. It might be that the journal
is relatively obscure, or that the PDF is from an old issue and available for
download only under an extended subscription that my library didn’t purchase.
And so I find myself either contacting the authors for a PDF (a very
hit-and-miss approach—in some cases they may no longer be alive) or asking
friends at even better funded institutions to send me the article via their
more comprehensive library subscriptions. To add insult to injury, it is not
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Internment | 137
uncommon for researchers at less well funded institutions to lack access to
even their own published articles. I discovered this in 2003 when, as a young
postdoctoral researcher, I had to pay (personally) to download an electronic
copy of my very first academic paper, published in the Journal of Experimental
Psychology: Human Perception and Performance.
The frustration academics experience in accessing the peer-reviewed
literature has provoked a remarkable campaign of insurrection against
mainstream publishers. In 2011, psychologist Andrea Kuszewski initiated
an underground Twitter movement called #icanhazpdf that harnesses
crowd sourcing to deliver nearly instant access to the paywalled academic
literature.27 A 2014 analysis estimated that #icanhazpdf is rising in popularity,
particularly in the life and social sciences (including psychology),
receiving over 3,000 requests per year.28
Within the emerging “guerilla open access” movement, some scientists
have taken civil disobedience to an even more radical level. In the same
year that Kuszewski invented #icanhazpdf, Kazakhstan-based researcher
Alexandra Elbakyan launched the website Sci-Hub.29 Based in St. Petersburg,
Sci-Hub provides a search engine and free repository of more than 40
million peer-reviewed academic articles, instantly bypassing publisher’s
paywalls using a variety of techniques such as obtained (or possibly hacked)
academic access credentials. Though the legality of Sci-Hub is in dispute,
Elbakyan asserts that it is the corporate publishers who are acting illegally
by denying free access to the scientific literature, thereby standing in violation
of Article 27 of the United Nations Declaration on Human Rights.30
Elbakyan’s goal with Sci-Hub is a simple one: “to collect all research papers
ever published, and make them free,” with a particular emphasis on making
research accessible to developing countries such as India, Iran, and Indonesia
that lack the funds to pay for subscription access.31
Acts of subversion can be convenient and powerful catalysts for OA.
However, they also run the risk of attracting litigation and even criminal
sanctions. For #icanhazpdf, the legality is unclear; since it doesn’t involve
posting copyrighted material directly into the public domain (only the
request for such material is public), it is not obviously illegal or actionable.
The legal status of #icanhazpdf remains to be tested in court, but this is
unlikely to happen anytime soon. The relatively small user-base of #icanhazpdf
presents little threat to the revenues of the major publishers, which
can run into billions of dollars per annum.32 Elbakyan’s Sci-Hub, on the
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
138 | Chapter 6
other hand, provides a more clear and present danger to the academic
publishing industry. By freely releasing millions of peer-reviewed papers,
Sci-Hub provides a route for individuals and institutions to avoid paying
journal subscriptions altogether. In an attempt to neutralize this threat,
in 2015 Elsevier filed an injunction against Sci-Hub and its affiliates under
US copyright law and the US Computer Fraud and Abuse Act, alleging
that the defendants engaged in “piracy” and “have caused and continue
to cause irreparable injury to Elsevier and its publishing partners (including
scholarly societies) for which it publishes certain journals.”33 Four
months later, a New York district court granted Elsevier’s injunction, after
which Sci-Hub and its partner sites were taken down.34 Within a few
weeks, however, the entire Sci-Hub operation reappeared under a different
web domain.35
In some cases, guerrilla open access tactics have triggered even more
than civil litigation, leading as far as criminal charges. In 2011, prominent
activist and Harvard researcher Aaron Swartz was arrested by the US Secret
Service for downloading millions of articles from an online repository of
paywalled journal articles called JSTOR. Swartz had retrieved the articles
through his university’s JSTOR account, using a laptop hidden inside a wiring
closet at the Massachusetts Institute of Technology. After several
months of state and federal investigation he was charged with wire fraud,
computer fraud, unlawfully obtaining information from a so-called protected
computer, and recklessly damaging a protected computer. In 2012,
the prosecutor added another nine felony counts to Swartz’s indictment,
leading him to face up to 50 years in prison and $1 million in fines. Swartz
pleaded not guilty but would never see a verdict.36 On 11 January 2013—two
years to the day after his initial arrest—he was found hanged in his Brooklyn
apartment.
Counterarguments
Despite the wide range of top-down and grassroots movements to drive
forward OA, it nevertheless remains the minority choice in psychology, as
in many other sciences. We saw earlier how one of the major barriers in
psychology is the low number of prestigious journals that are also fully OA,
creating a dilemma for senior scientists striving to promote the careers of
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Internment | 139
their junior apprentices. This disincentive, combined with the high cost of
gold OA in traditional subscription journals, presents a roadblock for many
academics. However, to fully understand the reasons for limited support of
OA in psychology, we need to look beyond these factors and consider some
of the common objections to the OA movement.
Counterargument 1: I have access, so what’s the problem?
In the busy life of an academic, the concern that people other than us lack
access to the peer-reviewed literature is easy to dismiss. Inertia in OA adoption
is fueled, in part, by the fact that the libraries in the richest and most
influential universities (mostly, those in major Western countries) are sufficiently
armed with subscriptions to make published science available to
their own academics. And if influential academics can read the work of
other influential academics, doesn’t that circumvent the need for OA
altogether?
There are at least three problems with this argument. The first is that it
relies on the politics of privilege, ignoring the fact that academics at less
wealthy universities often lack the access they need.37 When I was a postdoctoral
researcher we once received a visit from a researcher based at a
major Russian institution. After two days spent discussing science and what
it was like to work in the former USSR, she turned and asked quietly if we
could please fill her portable hard disk with PDFs from the major psychology
journals. Why? Because her university lacked subscription access to any
journals whatsoever. I found it astonishing. How could she function as an
academic? How was she able to forge a successful career in psychology? Her
solution was a picture of determined industriousness, contacting individual
authors one-by-one, by telephone or e-mail, to request copies of their articles.
I felt ashamed that such a successful scientist had to ask us to provide
access to literature that she had every right to see for herself. I also felt
ashamed for assuming, naively, that everyone could see this literature simply
because I could. The experience was eye-opening and led me to discover
that limited access is a problem experienced to various degrees by many
universities, even in Western countries.38
The second problem with this stance against OA is that it takes a narrow
view on the wider benefits of research for nonacademic groups, including
policy makers. Applications of psychology in public policy are many and
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
140 | Chapter 6
varied, ranging from tackling challenges like obesity and climate change
through to the design of traffic signs, persuading citizens to vote in elections,
and encouraging people to join organ donor registries.39 In the UK,
recognition of psychology in policy making is so prominent that in 2010 the
government established a dedicated Behavioural Insights Team, specializing
in the application of psychological science to policy.40 In 2015, a comparable
service was established in the United States.41
Given the importance of the peer-reviewed literature to policy making,
you might assume that the public service would be handsomely equipped
with journal subscriptions. Not so. In 2012, Adriana de Palma, a PhD student
in ecology at Imperial College London, discovered this for herself
when she undertook a short-term placement at the UK Parliamentary Office
of Science and Technology (POST).42 Writing on her personal blog after
the placement had finished, de Palma said: “One of the interesting things
I learnt while writing my POSTnote [an in-depth scientific briefing for
politicians] was the importance of publishing my work open access. The
level of access to journals was far lower than I had expected (it was actually
shocking)—I ended up using my academic access throughout my
placement.”43
What happens when civil servants lack the means to fall back on academic
access credentials, as de Palma could? Steven Hill, head of research
policy at the Higher Education Funding Council for England (HEFCE) reports
having “zero access to non OA content while working in Defra, RCUK
and HEFCE,”44 an observation confirmed by HEFCE policy researcher Ben
Johnson.45 Johnson even resorted to taking out a personal subscription to
gain access—or at least he tried. When even that turned out to be impossible,
he issued a public plea for researchers to make their work OA: “If we
want more science of science policy, I implore us to deliver the results
openly so that those making decisions can actually read them.”46 On his
blog he elaborated:
If my experiences are the same as others’ in my situation, then this is
a fundamentally stupid state of affairs for publishing and academia
alike. If it is almost impossible for people to subscribe to subscriptionbased
journals—even for honest and willing people like me who (a)
are tired of going through the rigmarole of searching out free copies
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Internment | 141
of articles, (b) might be willing to pay for a personal subscription,
provided it’s easy to set up and not too expensive, and (c) are pretty
savvy when it comes to using technology and the Web—then what
hope can publishers and academics have of their work reaching a
wider audience? I despair.47
Even putting aside the wider benefits of OA, the argument that “I have
access so everything is fine” has a third problem: those with the greatest
access can often find that their ability to use subscribed material is encumbered
with restrictions. In one recent case, Chris Hartgerink, a PhD student
in psychology at Tilburg University, collided with regulatory barriers when
attempting to content mine information in academic articles. Hartgerink’s
research focuses on extracting summary data from published psychology
papers in order to detect signs of data fabrication, an approach that requires
the automated, bulk downloading and analysis of articles from behind corporate
paywalls.48 Despite the fact that his approach is legal and in accordance
with his university’s journal subscriptions, Hartgerink reported that
Elsevier blocked him from conducting his research:
Approximately two weeks after I started downloading psychology research
papers, Elsevier notified my university that this was a violation
of the access contract, that this could be considered stealing of content,
and that they wanted it to stop. My librarian explicitly instructed
me to stop downloading (which I did immediately), otherwise Elsevier
would cut all access to Sciencedirect for my university.49
In response to Hartgerink’s blog post, Elsevier advised him that his
content-mining approach was acceptable only if he used a specific software
program supplied by the publisher.50 Use of this program, however, is conditional
on authors agreeing to various restrictions, such as the way in
which the extracted data can be published by the researcher, and the capacity
for the data to be reused by others. For example, the policy requires that
the published content must contain no more than 200 characters of mined
text per “snippet,” and that it must also be published under a more restrictive
CC-BY-NC license that forbids future commercial applications of the
work, as compared with a more open CC-BY attribution license.51 For these
reasons, Hartgerink decided that he was unable to use the tool, writing that
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
142 | Chapter 6
Elsevier’s policy “harms individual researchers and the impact of research
on society.”52 Thus, even within a well-resourced Western university, restrictions
on the use of legally subscribed content can fail to deliver the
needs of the psychological community.
A broader concern is that, even for the richest universities, university
library budgets are under increasing pressure from ever-rising journal subscriptions.
In 2012, the Harvard University library issued a provocative and
widely publicized memo in which it condemned the growing cost of subscription
publishing and called for researchers to widely support cheaper
OA journals:
We write to communicate an untenable situation facing the Harvard
Library. Many large journal publishers have made the scholarly communication
environment fiscally unsustainable and academically restrictive.
This situation is exacerbated by efforts of certain publishers (called
“providers”) to acquire, bundle, and increase the pricing on journals.
Harvard’s annual cost for journals from these providers now approaches
$3.75M. . . . Some journals cost as much as $40,000 per year,
others in the tens of thousands. Prices for online content from two providers
have increased by about 145% over the past six years, which far
exceeds not only the consumer price index, but also the higher education
and the library price indices. These journals therefore claim an
ever-increasing share of our overall collection budget. Even though
scholarly output continues to grow and publishing can be expensive,
profit margins of 35% and more suggest that the prices we must pay do
not solely result from an increasing supply of new articles.53
As of 2015, all scholarly articles published at Harvard University are now
made fully OA upon publication.54 Similar steps toward universal OA have
been made by the Association of Universities in the Netherlands (VSNU),
the centralized body that represents all Dutch universities. Throughout
2014 and 2015, VSNU was publicly vocal in its intention to boycott Elsevier,
with the aim to make all publications arising from Dutch universities fully
OA. Under the plan outlined by the undersecretary for education, culture
and science, Sander Dekker, Dutch universities will cancel all journal subscriptions
by 2024.55
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Internment | 143
Counterargument 2: Most of the public don’t want OA. Even
if they could read the literature they wouldn’t understand it,
and they might dangerously misunderstand it.
In discussions with colleagues I have found that this is a remarkably common
view that stems from an element of truth. As in other sciences, many
empirical papers in psychology are highly technical and written exclusively
for a specialist audience of like-minded scholars. Theoretical assumptions,
background studies, and technical terms are rarely explained in detail because
they are (quite rightly) assumed to be common knowledge among the
targeted readership. Given that this is the case, it is surely unlikely that basic
psychological science, as communicated in specialist journals, would be
interesting or intelligible to the general public. Prominent OA critic Daniel
Allington has claimed that OA therefore “amounts to no more nor less than
the handing of a publicly-funded megaphone to the UK’s most privileged
scholars. Soon the whole internet will be able to hear them talking to one
another.”56
Although on its face this argument appears sensible, its critical flaw lies
in the loose definition of “the public.” By lumping together all nonacademics
into a uniform mass, we slip into the well-known psychological trap of
“outgroup homogeneity” in which individuals not belonging to a narrow
ingroup (in this case university academics) are seen as more alike than they
really are.57 In fact, there are many varied applications of specialist knowledge
within nonacademic professions that make up a sizable chunk of “the
public.” The website whoneedsaccess.org provides examples of how OA is
important for some of these nonacademic groups—in addition to civil servants
and policy makers (as discussed earlier), the list includes translators,
research organizations, small businesses, people working with the developing
world, doctors and dentists, nurses, teachers, consumer organizations,
patients, patient groups, amateur scientists, Wikipedia contributors, bloggers,
science communicators, unaffiliated scholars working into their retirement,
enterprising schoolchildren, professional nonacademic researchers,
independent researchers, publishers, and even artists.58 Academics who
opine that OA isn’t necessary because nobody outside their narrow circle
cares about their work not only underestimate the value of their own research
but also the intelligence and intellectual vigor of groups beyond their
immediate horizon.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
144 | Chapter 6
What of the concern that OA might lead unqualified readers to misunderstand
the research, possibly even to the extent that they might put themselves
or others in danger? This may be an issue in a limited subset of psychological
fields, but even where this is a potential problem, the answer is
not for the work to be locked away behind paywalls to “protect” the public
but for academics to prepare intelligible lay summaries of their work that
dispel likely sources of confusion. There are now many avenues academics
can use to engage in productive dialogue with public audiences, for instance
by writing articles for outlets such as the Conversation and the Guardian
Science network, by writing on their own personal blogs, or through journalists
in the print and broadcast media.
Finally, it is of course true that a large proportion of the public will have
no interest in reading any particular peer-reviewed article. However, this
provides no escape from the leading axiom of the Finch report: that given
the reliance of the research on public funding, who are we to deny public
access to the output of our research?
Counterargument 3: OA journals don’t meet the same
editorial standards as traditional journals because authors
pay to publish. The journal therefore has a commercial
incentive to accept poor papers.
We don’t need to look far in psychology to find this criticism of OA journals—a
claim that seems to be particularly prevalent among a certain “old
guard” of psychology researchers. One of the most public examples of
hostility toward one OA journal in particular was the 2012 Psychology
Today article by John Bargh, which we discussed in chapter 1. Recall that
Bargh was responding to a failure to replicate one of his previous and
highly influential “social priming” studies, in which he concluded that
showing young participants words associated with elderly people caused
them to walk more slowly. In a subsequently deleted blog post, Bargh took
aim at PLOS ONE—the journal that published the failed replication of his
work—declaring that, as an OA journal, it provides substandard peer
review:
PLOS ONE . . . quite obviously does not receive the usual high scientific
journal standards of peer-review scrutiny . . . instead, the jour-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Internment | 145
nal follows a “business model” in which authors pay to have their
articles published (at a hefty $1,350 per article). . . . If I’d been asked
to review it (oddly for an article that purported to fail to replicate one
of my past studies, I wasn’t) I could have pointed out at that time the
technical flaws, though these might not have mattered to PLOS
ONE—as a for-profit enterprise, PLOS published 14,000 articles in
the year 2011 alone. Fourteen thousand. Something tells me they
don’t turn down many $1,350 checks.59
In fact, contrary to Bargh’s claims, the paper in question did receive indepth
peer review and was edited by Professor Jan Lauwereyns, an internationally
recognized expert in cognitive psychology.60 Furthermore, PLOS
ONE selects which manuscripts to publish based not on the authors’ ability
to pay (which is considered only once manuscripts are accepted, with fee
waivers available), but according to structured publication criteria that are
clearer than at most traditional psychology journals.61 Finally, not only is
there no evidence to support Bargh’s accusation of editorial negligence, his
criticism of PLOS ONE ignores the fact that many prestigious subscription
journals also operate “business models” that charge authors to publish (e.g.,
through page charges) in addition to charging libraries for subscriptions.
In addition to prejudice against OA journals from senior academics, the
OA movement has also come under direct attack from subscription publishers.
In 2013, journalist John Bohannon of Science magazine unveiled the
results of an audacious sting in which he sought to test the quality of peer
review in several hundred OA journals.62 With the help of academic colleagues,
he wrote a bogus paper with obvious flaws (and fake results) that
should have been readily picked up through peer review. He then submitted
the paper to 304 OA journals that charge APCs, including 167 that were
listed on the Directory of Open Access Journals (DOAJ), an online database
that purports to index “high quality, open access, peer-reviewed journals.”63
The results were disquieting: of 304 journals, over half accepted the bogus
paper, often with little or no peer review. Even more worryingly, that number
dropped only slightly, to 45 percent, when considering the OA journals
listed on the (supposedly selective) DOAJ.
Given the results of Bohannon’s sting, aren’t scientists (including psychologists)
sensible to be suspicious of publishing in OA journals? Wasn’t
Bargh right after all? Certainly there is reason to be careful about choosing
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
146 | Chapter 6
where to publish—and this is true for both subscription and OA journals—
however the sting provides no reason for psychologists to turn away from
all OA journals. First, the most prominent OA journals swiftly rejected
Bohannon’s bogus paper, including PLOS ONE and the targeted Frontiers
journal. In the case of PLOS ONE, Bohannon even noted that it “meticulously
checked with the fictional authors that [ethics] and other prerequisites
of a proper scientific study were met before sending it out for review.
PLOS ONE rejected the paper 2 weeks later on the basis of its scientific
quality.” Second, viewed as a research study, the sting also included an elementary
flaw—that by selectively targeting OA journals, and not including
a control group of subscription journals, it is impossible to know whether
OA journals are any more likely than subscription journals to accept bogus
papers. Addressing this error on his personal blog, PLOS founder Michael
Eisen wrote:
We obviously don’t know what subscription journals would have
done with this paper, but there is every reason to believe that a large
number of them would also have accepted the paper. . . . Like OA
journals, a lot of subscription-based journals have businesses based
on accepting lots of papers with little regard to their importance or
even validity. When Elsevier and other big commercial publishers
pitch their “big deal,” the main thing they push is the number of papers
they have in their collection. And one look at many of their journals
shows that they also will accept almost anything.64
Eisen also took aim at the selectively chosen subset of OA journal targeted
in the sting, which he argued was unrepresentative of OA journals
overall: “To suggest . . . that the problem with scientific publishing is that
open access enables internet scamming is like saying that the problem with
the international finance system is that it enables Nigerian wire transfer
scams.” He concluded that the “long con” orchestrated by subscription publishers
far outweighs the actions of the obviously predatory OA journals
who fell for Bohannon’s sting: “Not only do [the subscription publishers]
traffic in billions rather than thousands of dollars and [deny] the vast majority
of people on Earth access to the findings of publicly funded research,
the impact and glamour they sell us to make us willing participants in their
grift has serious consequences.”
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Internment | 147
Counterargument 4: What’s in it for me?
In the competitive world of academic science, researchers naturally seek
rewards for engaging in practices that break from the norm. As with data
sharing (see chapter 4), there is now substantial evidence that OA publication
is associated with increased citation rates. In a key 2005 study, Chawki
Hajje, Stevan Harnad, and colleagues from the University of Quebec reported
that, across a sample of more than 1.3 million articles published
between 1992 and 2003, those that were OA attracted between 25–250
percent more citations than those behind paywalls. In particular, they
found that OA articles in psychology were associated with more than double
the number of citations.65 Since then, the SPARC Europe project has catalogued
this effect across multiple fields and many published reports, finding
that of 70 published investigations, 46 find evidence of an OA citation advantage,
with the magnitude ranging from less 5 percent to over 500
percent.66
One caveat that must be applied to such studies is that they relied on
observational designs in which data from previously published papers were
collated and analyzed. Because of their retrospective nature, such designs
are unable to show that differences in citations between OA and non-OA
papers are caused by the openness of the publishing route as opposed to a
myriad of other possible confounding variables, not least of which may be
the inherent quality of the research. In order to provide causal evidence
that OA boosts citations, we would need to see the results of a controlled
trial in which papers were randomly assigned to either an OA or non-OA
group, much like a clinical drug trial—and no such interventional study has
yet been reported. Nevertheless, the correlational evidence is at least suggestive
of a causal effect, so in the quest for professional rewards, the citation
advantage of OA provides a good reason for authors to prefer more open
publishing practices.67
An Open Road
What does the future hold for OA publishing in psychology? Can the field
break free, once and for all, from the grip of subscription publishers that
can be paid not once, not twice, but three times for a product they didn’t
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
148 | Chapter 6
create in the first place? Although we are some way from an ideal future of
universal OA, the signs are promising that it will eventually come to pass.
In addition to policy-led reforms and subversive grassroots initiatives, we
are also seeing novel publishing models emerge in psychology that are creating
a new arena for open practices. Recent examples include the new OA
journal Collabra, which unlike most journals pays reviewers and editors for
their work, but cleverly provides them with the choice of either collecting
the money directly or paying it forward into an APC waiver fund that can
be used by other authors or their own institutions. With wide uptake, this
model could be transformative. Meanwhile, PeerJ offers a much cheaper
gold OA model than other journals by providing membership plans to authors
rather than charging APCs per paper. Although neither model or
journal is yet regarded as a first choice for most psychologists, the field is
positioned for new ideas to take hold.
As we will see in chapter 8, there is an even more radical solution to the
sin of internment—one that is not as distant as it might appear. We could
abolish journals altogether.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
C H A P T E R 7
The Sin of Bean Counting
Not everything that counts can be counted, and not
everything that can be counted counts.
—William Bruce Cameron, 1963
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Science is all about measurement. Whether it’s the speed of a neutrino
or the reaction time of a person, quantifying natural phenomena is
what scientists do. Yet, for all our ingenuity, one question has proven particularly
challenging: can science measure itself? And, assuming that the
quality of science is something that is in fact quantifiable, can scientists
themselves be boiled down to a series of numbers that reflect their past
achievements and future potential? In psychology, as in many other sciences,
there is a growing push toward weighing up the worth of individual
academics and their research contributions based on various “metrics,”
and to then use those metrics to award jobs and funding. Metrics such as
the “impact factor” of the journals in which we publish, the quantity and
monetary value of grants we receive, the number of papers we publish, our
author position on those papers, and the number of citations we receive,
have all become so-called key performance indicators. And because academics,
and the bureaucrats who oversee us, view these indicators as synonymous
with high-quality science, the measures themselves have become
career targets.
There are reasons to be skeptical about reducing the quality of science
to numbers. Goodhart’s law of economics warns us that when a measure
becomes a target, it ceases to be a good measure.1 This is because those of us
who stand to win or lose immediately seek ways to game the system in order
to achieve the target, sacrificing other important (and unmeasured) goals
along the way, and thereby compromising the value of both the measure
and our wider mission. On top of this, many such measures hold no true
value in the first place.
For basic scientists, the decisive indicator of our empirical achievements
is the advancement of theory that accurately predicts and explains future
observations. As we saw in chapter 3, this path eventually leads us to refine
theories into such razor-sharp explanatory instruments that they become
laws. For applied psychological scientists, the goal is more practical, but the
logic is the same: the greatest signifier of success is that we changed something
in society, whether addressing a problem such as obesity or mental
illness, using psychological evidence to develop a successful policy, or building
something that in a demonstrable sense “works.” Yet in our obsession
to quantify quality—to turn a thing from words into numbers—we have
allowed these aspirations to become overshadowed by a culture of metrics-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bean Counting | 151
based performance management. In doing so, psychology commits our seventh
and final transgression: the sin of bean counting.
Roads to Nowhere
In psychology, the sin of bean counting manifests in several different ways.
We reduce the skills and knowledge of individual scientists to their citation
metrics and other simplistic surrogates of “impact.” We reward academics
disproportionately for winning research grants, summing scientific inputs
(funding) to outputs (discoveries) rather than weighing them against each
other, and in the process favoring more expensive research. And we tally
the authorship order of scientists on research papers while clinging to an
antiquated system of attribution that muddies their individual contributions.
Some of these problems are more severe than others—and none are
unique to psychology—but they all stem from the same root cause. To borrow
the words of Oscar Wilde, we have become obsessed with the price of
everything and the value of nothing.
Impact Factors and Modern-Day Astrology
The first hint of the metrics revolution came in 1955 with the publication
of “Citation Indexes for Science,” by Eugene Garfield in Science magazine.
Prior to this there had been no systematic way for scientists to track which
academic papers were cited by which, or for counting the number of citations
received by individual articles or journals. Garfield’s ambitious plan
was to bring coherence and efficiency to science by creating a massive referential
database of citations. Among his core suggestions was the creation
of a citation metric called the impact factor—a way of quantifying the influence
of a paper on others by counting the number of citations it received.
Garfield wrote:
In effect, the system would provide a complete listing, for the publications
covered, of all the original articles that had referred to the
article in question. This would clearly be particularly useful in historical
research, when one is trying to evaluate the significance of a
particular work and its impact on the literature and thinking of the
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
152 | Chapter 7
period. Such an “impact factor” may be much more indicative than
an absolute count of the number of a scientist’s publications.2
Since then the impact factor has evolved to become one of the most
powerful metrics in science—a proxy for the prestige of journals and the
caliber of the scientists who publish within them. The modern form of the
impact factor is calculated at the level of the journal (rather than article) as
the average number of citations in the current year to articles published
within the previous two years. So, for example, to calculate the 2014 impact
factor for Nature magazine we add together the total number of citations in
2014 to Nature articles published in 2012 and 2013 (71,677) and divide it
by the total number of citable Nature articles during the same period (1,729).
The answer, 41.5, is among the highest in scientific publishing. Journals are
now being routinely ranked by their impact factors, with academics—including
psychologists—striving to publish their papers in so-called high
impact journals.
In the decades following Garfield’s call for science to get organized, he
would go on to create the Science Citation Index (and Institute for Scientific
Information, now owned by Thomson Reuters), as well as ushering into
existence the disciplines of scientometrics and bibliometrics—divisions of
information science that seek to itemize the impact and value of science.
Writing in the inaugural issue of Scientometrics, a journal that Garfield
helped establish in 1978, his coeditor Mihály Beck defended their mission
to quantify quality:
The quantitative evaluation and intercomparison of scientific activity,
productivity and progress seems to be sheer nonsense for many,
perhaps for the majority of active scientists. This attitude is quite
natural, but its source is mainly prejudice, ignorance and/or misunderstanding
of the basic ideas of scientometrics. The skeptics are
scornful of the hopeless aim of measuring the unmeasurable. The
tremendous increase of the scientific production over the last decades
has made the emergence of this new field of science both necessary
and possible.3
Garfield says that he created journal impact factor (JIF) with two purposes
in mind: to help libraries decide which journal subscriptions they
should purchase, and to help authors decide in which journals they should
publish—operating on the (since upheld) assumption that scientists gener-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bean Counting | 153
ally view journals with higher JIFs as being more prestigious and desirable.4
Garfield’s assumption that citation statistics tell us anything about the quality
of published work is widely disputed, but not nearly as contentious as
the now routine practice of evaluating the quality of individual articles, and
individual scientists, by the impact factors of the journals in which they
publish.5
Why is it a problem to assess the value of individuals based on journallevel
citation statistics? The first concern lies in the statistical properties of
citation distributions. Most of the citations in a journal are generated by a
minority of articles, a phenomenon that Garfield himself has referred to as
the “80/20 rule”: across science as a whole, around 80 percent of citations
are attracted by around 20 percent of papers. This leads to a right-skewed
distribution of citations in which most articles receive few citations while
a scattered minority receive a lot. The JIF is therefore a poor indicator of
central tendency because it represents so few individual samples in the
population. Figure 7.1 shows this graphically by plotting the citation distribution
for Open Biology, a journal published by the Royal Society. Only 1 in
16 articles attracted the same number of citations as the JIF, and many
received few to no citations at all. Because of the skewness and spread of
citation distributions, a second concern is that JIF only barely, if at all, predicts
the citation rates of individual articles within the journal.6 This means
that publishing in a so-called high-impact journal provides no promise that
an article will be highly cited, just as publishing in a lesser-known journal
is far from condemning it to obscurity.7 Troublingly, however, JIF does correlate
with one metric: the rate of articles retracted owing to fraud or suspected
fraud.8 Rather than referring to outlets such as Nature and Science
as “high impact” journals, it would, if anything, be more accurate to call
them “high retraction” journals.
A third problem is that the JIF is easily massaged by editors and publishers.
One tactic is for the journal to publish papers that are predicted to be
highly cited (such as review articles) toward the beginning of the year, giving
those articles more time to accrue citations over the two-year period
that feeds into the JIF calculation. For the same reason, it is not unheard of
for psychology journals to strategically delay publication of potential “heavy
hitters” until the start of the following year. More insidious tactics include
journals publishing editorials that excessively cite the journal’s own articles,
or editors nudging authors of submissions to do the same. Such practices
are rightly frowned on but usually fly below the radar of wider scrutiny.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
154 | Chapter 7
Percentageofarticles
14
12
10
8
6
4
2
0
Number of citations
504642383430262218141062 54
Average (JIF)
Figure 7.1. Citation distributions highlight the meaningless nature of the journal impact
factor (JIF). Here we see the distribution of citations used to calculate the 2014 JIF for the
Royal Society journal Open Biology (the results shown here were extracted from http://
rsob.royalsocietypublishing.org/citation-metrics). The chart plots the percentage of articles
published throughout 2012–2013 (y axis) according to how many citations they received
in 2014 (x axis), with the JIF calculated as the average of this distribution (vertical
dotted line). Note the long rightward tail, or right skew, in which most articles received
few citations, with about 9% attracting none at all. Open Biology is actually less severely
skewed that many scientific journals—here, the top 20% most highly cited papers attracted
56% of the total citations. Even so, the degree of skewness is sufficient to ensure
that two-thirds of articles had fewer citations than the JIF, and the JIF itself represented
only about 6% of individual articles published. As these distributions become broader and
more heavily skewed, JIF becomes increasingly meaningless as an indicator of individual
article citations.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bean Counting | 155
Even putting aside its statistical flaws and vulnerability to manipulation,
the integrity of JIF is undermined by the fact that it is arbitrarily negotiated
rather than objectively calculated. Journals can lobby the company that
determines the JIF (Thomson Reuters) to reduce the number of citable articles
in the JIF calculation and thereby inflate the score. In 2006, the editors
of PLOS Medicine described the remarkable series of face-to-face meetings,
phone calls, and e-mails as part of negotiating their 2005 JIF.
Depending on which articles were deemed to be citable, their resulting JIF
could vary between 3 and 11—a range that would dramatically affect the
journal’s published JIF ranking. The editors were less than enamored with
the opacity and capriciousness of the process, concluding that “science is
currently rated by a process that is itself unscientific, subjective, and secretive.”9
In 2013, Björn Brembs, Kate Button, and Marcus Munafò further
revealed how the JIF of Current Biology jumped by 70 percent, from 7.0 in
2002 to 11.9 in 2003, not because the number of citations increased by a
significant amount, but because the number of citable items during the 2001
period was quietly reduced from 528 to 300.10
Even if JIFs were indicative of article-level citations (which they are not),
would it be safe to assume that the number of citations an article receives
reflects the value of its contribution to science? In other words, to what
extent is the quality of science synonymous with short-term popularity? To
test the idea that reproducibility correlates with JIF—and therefore that
article prestige predicts at least one potential measure of study quality—
Björn Brembs and colleagues examined the relationship between JIF and
statistical power. They found that the statistical power of 650 neuroscience
studies was unrelated to the JIF of the outlets in which they were published
(see figure 7.2). Not only do such findings call into question the typical
methods used to evaluate researchers in academic departments, they also
cast doubt on Eugene Garfield’s claim over ten years earlier that JIF “is not
a perfect tool to measure the quality of articles but there is nothing better.”11
Whether there is any metric better than JIF is unclear, but it is hard to
imagine anything worse.
Despite all the evidence that JIF is more-or-less worthless, the psychological
community has become ensnared in a groupthink that lends it value.
As a junior postdoctoral researcher, I had above my desk a printed list of
“top journals” I aimed to publish in, rank ordered by JIF. Science, Nature,
Nature Neuroscience, Neuron, PNAS, Current Biology, and so on. I would send
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
156 | Chapter 7
most of my papers to The List, starting near the top and working my way
down, one-by-one, until the paper was either accepted in a “high-impact”
outlet or I had to settle for publishing in a specialist journal. In retrospect
it was a colossal time sink, and not only for me. Consider the waste I produced
in the review system—the hours of reviewers’ time I consumed by
requiring other scientists to repeatedly read and then reject my submissions,
not (in most cases) because my studies were scientifically flawed but because
the research question or results were not important or definitive enough
for such prestigious outlets. Multiply this behavior across many thousands
of scientists worldwide, and the stupidity of our collective action is nothing
short of staggering.
Fast-forward 15 years and psychology remains hooked on JIF. When I applied
for promotion at my own university in 2013, the promotion criteria—
which were sensible in many other respects—stated that “[e]ach publication
in the [applicant’s] list should be accompanied by the impact factor of the
Figure 7.2. Brembs, Button, and Munafò (2013) found no discernible correlation between
the statistical power of individual studies in neuroscience and the JIF of the journals in
which those studies appeared. These results therefore provide no evidence that journals
with higher JIFs publish more reproducible science.
Statisticalpower(%)
100
80
60
40
20
0
Impact factor
30101.00.1 60
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bean Counting | 157
journal.” Why? What possible information could such a discredited statistic
possibly tell a promotion committee about my academic achievements or
future prospects? Two years later when I examined a PhD at Maastricht
University in the Netherlands I was confronted with an assessment form in
which the student not only had to count the number of publications she had
published (a questionable practice in itself), but was also required to add up
the JIFs of those publications to produce a “summed impact factor.”12
Evaluation of PhD thesis of: xxxxx xxxxx
by: [Prof. dr. C. Chambers]
PhD Length 3.5 years
OUTPUT STATISTICS
Number of empirical chapters not published 2
Number of empirical chapters published 4
Number of first authored chapters 6
Summed impact factor of publications 20.124
Please rate output, skills, theoretical thinking and formal presentation
(1 = below average; 2 = average; 3 = very good; 4 = exceptional; most
scores in most theses will hover around 2)
OUTPUT (SCORE 1–4)
Overall productivity/quantity of output
Evaluation of summed impact factor
Summed score
SKILLS (SCORE 1–4)
Experimental design
Data acquisition methods
Level of statistical analysis
Computational approaches, modeling
Summed score
THEORETICAL THINKING (SCORE 1–4)
Originality of ideas in all articles chapters
Scholarly depth in all articles/chapters
Scholarly depth of General Introduction
Scholarly depth of General Discussion
Summed score
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
158 | Chapter 7
FORMAL PRESENTATION (SCORE 1–4)
Quality of academic English writing
Quality of figures
Summed score
TOTAL SCORE
Total summed score
Please judge which individual publications are of high impact (High
impact corresponds to the top 5% of journals in the applicable research
field. If applicable, write down in empty field the start page number[s] of
the high impact publication[s].)
HIGH IMPACT PAPERS
Number of publications of high impact N/A
Psychology’s ongoing obsession with JIF is revealed no more blatantly than in this perplexing
PhD assessment from Maastricht University in the Netherlands—a university that has
joined 13 other Dutch institutions in signing DORA. The form shown here, from 2015,
requires the student to add up the JIFs for all articles they have published during their
PhD to calculate a “summed impact factor.” The examiner is then asked to evaluate the
summed impact factor.
As an examiner I was then expected to offer an “evaluation of the
summed impact factor.” I found this requirement astonishing given that, in
the previous year, all Dutch universities had signed the San Francisco Declaration
on Research Assessment (DORA). DORA commits signatories to
“not use journal-based metrics, such as Journal Impact Factors, as a surrogate
measure of the quality of individual research articles, to assess an
individual scientist’s contributions, or in hiring, promotion, or funding decisions.”13
Soon after I submitted my assessment form, in which I pointed
out the contradiction between signing DORA and using JIF to assess students,
I received a reply from a senior university administrator explaining
that the evaluation of PhD students according to their summed impact
factors was a new mechanism being trialed.
please note that this evaluation form has still a “pilot” status (it is not
[university] broad, not faculty, but [departmental] policy), and the
impact and publication rank issues will be evaluated /adjusted. . . .
We also take care that critical issues detected in this pilot will not
influence the evaluation of the thesis in any negative sense.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bean Counting | 159
Despite the DORA serving as a national repudiation of JIF, it appears
that individual departments within Dutch universities are free to create
mechanisms that disregard it.
The JIF remains a popular metric despite no good evidence that it captures
anything meaningful about the quality of science or scientists. If our
attachment to JIF tells us anything, it is only how the sin of bean counting
overwhelms our sense of reason. Brembs and colleagues conclude that “[m]
uch like dowsing, homeopathy or astrology, journal rank [including judgments
according to JIF] seems to appeal to subjective impressions of certain
effects, but these effects disappear as soon as they are subjected to scientific
scrutiny.”14 Structural biologist Stephen Curry, from Imperial College London,
goes even further in condemning our collective foolishness:
[A]stonishingly for a group that prizes its intelligence, [we] have acquired
a dependency on a valuation system that is grounded in falsity.
We spend our lives fretting about how high an impact factor we
can attach to our published research because it has become such an
important determinant in the award of the grants and promotions
needed to advance a career. We submit to time-wasting and demoralising
rounds of manuscript rejection, retarding the progress of science
in the chase for a false measure of prestige.
Curry offers the following warning to anyone who uses JIF as an indicator
of pretty much anything at all:
—If you include journal impact factors in the list of publications
in your cv, you are statistically illiterate.
—If you are judging grant or promotion applications and find
yourself scanning the applicant’s publications, checking off
the impact factors, you are statistically illiterate.
—If you publish a journal that trumpets its impact factor in
adverts or emails, you are statistically illiterate. (If you trumpet
that impact factor to three decimal places, there is little
hope for you.)
—If you see someone else using impact factors and make no
attempt at correction, you connive at statistical illiteracy.
The stupid, it burns. Do you feel the heat?15
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
160 | Chapter 7
Unfortunately, it seems that most psychologists have yet to even get the
message let alone feel any heat. Of the 840 organizations that have signed
DORA in officially renouncing JIF, the Association for Psychological Science
(US) is the only major professional group to have done so in psychology.
Conspicuously absent are the American Psychological Association, the Psychonomic
Society, and the British Psychological Society, as well as numerous
other national and international psychological groups.
Even the Association for Psychological Science offers a strangely conflicted
rejection of JIF. Four months after signing DORA, the then chief
editor of its flagship journal, Psychological Science, Henry L Roediger III,
wrote an insightful critique of JIF but concluded by saying: “I have been
hard on [J]IF, but let me play devil’s advocate and admit that [J]IF, even with
its flaws, captures something real about journal quality.” Contrary to DORA,
Roediger even falls short of dismissing JIF as an indicator of individuals,
leaving room for it at the table provided other factors are also considered:
“Even if [J]IF values have some uses, when judging job candidates, tenure
candidates, and grant proposals, the focus should be on the individual (or
proposal), not just on [J]IF of journals where the candidate published.”16 To
further muddy the waters, Psychological Science is promoted by its publisher
as sporting “a citation ranking/impact factor that consistently places it in
the top 10 psychology journals worldwide.”17 Even within one of the few
psychological organizations to have signed DORA, it is clear that JIF retains
a tight grip.
Wagging the Dog
Throughout my career I have been reasonably fortunate in attracting research
funding, which means I sometimes find myself sitting in Kafkaesque
meetings about university strategies to maximize so-called grant capture.
Through the fog of management speak, a familiar scenario descends in
which everyone in the room transforms into caricatures from the popular
Irish sitcom Father Ted:
Professor A: “Our application success rates are too low!”
Professor B: “That’s right, and we don’t pull in as much grant funding
as these other universities, here look at this list.”
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bean Counting | 161
Professor C: “We need more grants!”
Professor A: “Yes, more grants! But how?”
Professor C: “We must apply for more grants! Then we will get more
grants.”
Professor B: “Brilliant! And let’s also be sure to apply for more expensive
grants. Then we’ll rank higher on these arbitrary league tables
that show which universities spend the most public money.”
Professor A: “Oh yes, it’s very important to spend more public money.
How else will the world know how brilliant we are?”
[Everyone nods sagely]
Of course, the one thing that tends to be lacking from such strategy meetings
is any talk of science, along with any recognition that grants are a means
to an end rather than an end in themselves. And if I find myself rather wearily
raising this issue, again, my comments are often met with an uncomprehending
look or even an eye roll. Who invited this guy? Such is the nature of
grant strategy—its purpose is not to ruminate on our scientific mission but
to focus on how to hit funding targets, attract more money than our “rivals,”
hone the art of grantsmanship (the fabled skill of successful grant writing),
and capitalize on the strategic priorities of funding agencies to ensure that
we choose research questions that the agencies care most about and so will
win us more money. Carl Bergstrom, a professor of biology at Washington
University, sums up this mindset aptly: “I’m told that long ago, scientists
pursued grants in order to do research, rather than vice versa.”18
Why are research grants so important? At the level of departments and
institutions, grant capture has become a banner of prestige and status—a
way of showing the world that our particular organization is leading the
way, ahead of the pack. At the level of individuals within institutions, the
number of grants we receive (and their monetary value) feeds into decisions
about who to hire and who to promote. At University College London, for
instance, staff who apply for promotion are expected to show a “strong record
of support through grants for research and travel,” and in order to be
promoted to the more senior academic ranks of reader or professor they are
required to have “significant [i.e., lucrative and repeated] and sustained [i.e.,
long-term] success in obtaining research grants (and plans for further grant
submissions).”19 These requirements are not unusual among British and
American universities. Despite the fact that, in most cases, grants are inputs
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
162 | Chapter 7
that pay for specific research projects, they have become regarded as outputs—fiscal
quality indicators that are added to the assessment of research
publications in determining the value of individual researchers. As psychologist
Dorothy Bishop has observed, grants have become trophies on the
shelf rather than investments to be balanced against returns:
Many people seem to perceive a large grant as some kind of “prize,”
a perception reinforced by the tendency of the Times Higher Education
and others to refer to “grant-winners”20. Yet funders do not give
large grants as gestures of approval: the money is not some kind of
windfall.21
But hang on, you might ask, is it entirely wrong to assess individual scientists
or their institutions based on grant capture? Surely what makes an
awarded grant an indicator of research quality is the fact that someone other
than you felt your research was important enough to fund, right? While
this is partially true, it is also misleading. What counts in science is the
quality of the research that is actually conducted and published, not the
hypothetical quality of research we plan to conduct, regardless of how impressive
that plan may sound.22
Our system of academic promotion highlights the folly of rewarding scientists
for plans rather than outcomes. Consider a scenario in which two
psychologists are competing for one promotion opportunity. Jane applies
with no major grants but a track record of high-quality publications. She
can also show that her research has had a significant influence on fundamental
theory, despite being unsupported by grant funding.23 Now let’s pit
her against Mary, a researcher in the same field and at the same career level.
Mary has a comparable publication record and research influence, but unlike
Jane her research was supported by large public grants. Who do you
think should be promoted? If you asked your next-door neighbor (or any
economist for that matter) they would probably choose Jane and praise her
efficiency and ingenuity—after all, conducting impactful projects without
major funding requires a special degree of resourcefulness.
In our academic promotion system, however, Mary would wipe the floor
with Jane.24 Why? Because in judging a researcher’s achievements we sum
inputs (money spent) with outputs (research produced) rather than dividing
them. And this, in turn, means that our system would value Mary over Jane
for the sole reason that Mary’s research was more expensive. Being awarded
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bean Counting | 163
a research grant is a fine achievement and something for which a researcher
can feel justly proud. But grant capture should be completely irrelevant for
assessing the contribution of researchers to science because a grant is not
a contribution to science. Rewarding academics for acquiring grants is like
deciding the winner of a football tournament based on which team spent
more on boots.
Every so often the sin of bean counting reaches extraordinary heights.
Since 2012, a number of major UK universities have set minimum funding
targets in order for academics to merely retain their jobs. While none of
these cases occurred specifically within psychology, they nevertheless hold
important lessons for our field. In 2012, Queen Mary University of London
was the first to announce that avoiding redundancy would require staff to
acquire £50,000 per year of grant income as a lecturer, £65,000 as a senior
lecturer, £80,000 as reader, and £100,000 as a professor.25 At the Institute
of Psychiatry at Kings College London, the criteria were even higher, revealed
in a leaked internal document to reach £200,000 for professors.26
At Queen Mary, these new rules were compounded by also requiring staff
to publish a minimum number of so-called high-quality publications over a
four-year period, with quality defined solely as publishing within outlets
that have JIFs above 7.0. Such bare-faced bean counting prompted two
Queen Mary academics, John Allen and Fanis Missirlis, to declare their
university’s policy “a triumph of vanity as sensible as selecting athletes on
the basis of their brand of track suit.”27 Both researchers subsequently lost
their jobs.
The Murky Mess of Academic Authorship
A successful career in psychology, as in others sciences, depends on more
than simply chalking up a long list of grants and publications. Equally important
are our individual contributions to published work and the resulting
credit we claim, a point highlighted in section 6.1 of the position statement
of the Committee on Publication Ethics, Responsible Research Publication:
International Standards for Authors: “The research literature serves as a record
not only of what has been discovered but also of who made the discovery.”28
As part of every piece of research there is a crucial negotiation between
parties for attribution. Who was the main progenitor of the idea?
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
164 | Chapter 7
Who did the most work? Who analyzed the data? Who wrote (or will write)
the manuscript? Did person X do enough to justify authorship as opposed
to earning a thank-you in the acknowledgements? If X is included, do we
have to also include Y? The list of questions goes on.
Careful negotiations about author attribution are vital because greater
credit rightly belongs to researchers who made a greater scientific contribution.
Beyond this point, however, everything about our system of attribution
fails us miserably. The biggest problem is that rather than listing and
quantifying the contributions made by researchers to an academic paper,
we instead reduce this rich information to a crude heuristic: the published
order of authorship. In psychology, order of authors works (mainly) like
this—and because the rules are so murky and capricious the following
points are highly caveated. The first-named author is usually the researcher
who made the greatest intellectual contribution to the study and, again
usually (but not always), the person who took responsibility for data analysis
and much of the interpretation. Typically the first author is also expected
to take the lead in writing the paper and coordinating the drafting
process with the other coauthors. After the first author, the next most
important position is the last author. The last author, or senior author, is
usually the principal investigator—the top dog who made the study possible
either by holding the grant that funded it or by supervising the student who
conducted it.29 If academic papers were movies, the first author would be
the director while the last author would (usually) be both the producer and,
ideally, the assistant director. In between the first and last positions lie the
various grips and gaffers, and attribution here becomes even more nebulous.
Second author is typically the next most coveted spot, held by someone
who contributed significantly but slightly less than the first author.
Between second and last lies the vast tundra of middle authors, whose
actual contributions can vary enormously without anyone—even the authors—really
knowing why they are in a particular position. For example,
the fourth author of a seven-author paper might be a specialist statistician
who designed and implemented a vital part of the analysis, or it might be
an intern who collected some data on a quiet weekend, or even a collaborator
who shared a bit of code early on and never saw the paper again. Most
psychology journals don’t list the specific contributions that each author
made to the design, implementation, analysis, interpretation, and writing
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bean Counting | 165
of published papers; they simply list the order and leave it for the community
to guess.
The second major problem is that factors other than scientific contribution
routinely influence the degree of attribution. In some areas of neuropsychology
and psychiatry, it is standard practice for the clinician who
supplied patient volunteers to be offered coauthorship as compensation,
despite this person making no contribution to any other aspect of the research.
Similar customs apply for principal investigators (lab leaders), who
claim authorship on papers emerging from their lab even when they had
little intellectual or practical involvement in the work. Even more egregious
are various forms of “gift authorship.” Sometimes this happens by falsely
inflating the contribution of an existing team member on the grounds that,
in order to provide an immediate career boost, one coauthor needs a higher
level of credit than another. Other times, authorship is offered to colleagues
either as a social gesture (to maintain a working friendship or in anticipation
of quid pro quo) or in the case of heavyweight colleagues as a form of
celebrity endorsement to bias the journal editor into accepting the paper.
All such practices violate established guidelines for authorship outlined by
the American Psychological Association (APA) and the International Committee
of Medical Journal Editors (ICMJE). The APA specifically notes that
authorship is not necessarily warranted for “providing funding or resources,
mentorship, or contributing research but not helping with the publication
itself.”30 The ICMJE further outlines four necessary and sufficient conditions
that must be met by any researcher wishing to claim coauthorship:
• Substantial contributions to the conception or design of the work;
or the acquisition, analysis, or interpretation of data for the work;
and
• Drafting the work or revising it critically for important intellectual
content; and
• Final approval of the version to be published; and
• Agreement to be accountable for all aspects of the work in ensuring
that questions related to the accuracy or integrity of any part of
the work are appropriately investigated and resolved.31
A third problem is that attribution by rank order makes it impossible
for authors to make an equal contribution, despite this being a frequent
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
166 | Chapter 7
possibility in any group-led scientific enterprise. Under a ranking system,
the only way to represent equal attribution is through a clumsy asterisk
next to two (or more) names in the byline, with a footnote stating that
“these authors contributed equally to the work.” Such footnotes are easily
missed when reading a paper and are not stated at all when a paper is
cited. Rank order requires that one author is always listed before another
and, for better or worse, that author will be seen as the more important
contributor.
The ethics of academic authorship are impossible to police and are routinely
flouted because the culture of bean counting in psychology rewards
academics who publish a high volume of first- and last-authored publications.
If you are a junior researcher in psychology you will want to tally up
a high quantity of first-authorships, ideally in journals that are prestigious
or with high JIFs. As you progress into research leadership and emerge as a
“lab head,” it becomes more important to cultivate a rich crop of last authorships
as an indicator of seniority. In this way, we can see how the sin of bean
counting manifests not simply as a ledger of articles published, but as the
quantity of articles published with specific attributions, over and above the
scientific quality and importance of the work, and ignoring the murky imprecision
of rank authorship.
I myself am guilty of fueling this obsession with authorial bean counting.
In 2012, I wrote a blogpost that offered 25 tips for early career researchers.
Regrettably, some of my advice emphasized the quantity of publications
with particular attributions rather than the quality of science being
performed:
Aim to publish every year. . . . The rule of thumb in psychology and
cognitive neuroscience is to publish four good papers per year, but
this can vary. Whatever happens, be sure to publish something
every year. If your experiments are slow or not yet producing publishable
data, then publish a review paper. . . . Aim for as many firstauthorships
as possible.32
In response, Dorothy Bishop quite rightly called me out for bean counting,
writing in a comment below my article:
I am uneasy with the recommendation to publish 4 papers per year,
and to write up as much as possible of your thesis. I think you should
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bean Counting | 167
publish when you have done something important enough to communicate
to the rest of the world. . . . I’d rather see a c.v. with one
really carefully-done, thoughtful, original paper every year, than one
with ten papers per annum, none of which is memorable in any way.
It’s a balance that’s not always easy to get right, but my advice is, remember
that if you write stuff, editors and reviewers have to read it,
and they may form a negative impression of you if it’s trivial or pointless
stuff, or if you have clearly been chopping up a decent study into
minimal publishable units.33
This gave me pause for thought. Why had I offered this advice? In my
reply, I diverted the blame to decision makers, as many researchers who
offer such suggestions do. It wasn’t my fault, I implored, because I was told
so by the powers that be. Don’t shoot the messenger, and so forth. It was a
weak defense:
This was the advice I received time and again from the BBSRC fellowships
committee; in fact I remember a slide from the induction
session of my David Phillips fellowship in 2006 which stated “Publications:
ensure a balance between Nature and potboilers.” And when
I went for my mid-term assessment, they made a point of warning me
that because I was transferring institutions, I should guard against a
drop in the *quantity* of publications. . . . from my experience, at
least, quantity of output is still regarded as an important gauge of
individual success in science. I’m not saying it should be.34
It is a curious experience to reread one’s own words and see them as
fundamentally flawed. What I realize now is that in any resource-limited
system, quantity and quality lie at opposite ends of a seesaw. This means
that by weighing the value of scientists in any way according to the quantity
of journal articles they produce, we necessarily devalue quality. At its extreme,
valuing quantity encourages salami slicing of research into what are
known as “minimal publishable units”—the smallest amount of research
that can be accepted in a journal and thus count in a CV—which in turn
floods the peer-reviewed literature with an overabundance of hurried, small,
and oversold studies. Instead we should be rewarding scientists for publishing
a fewer number of higher quality studies that are more theoretically
important, larger, and more methodologically rigorous. Developmental psy-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
168 | Chapter 7
chologist Uta Frith has referred to this revisionist philosophy as “slow science,”
drawing an analogy to cooking.
Slow Food and Slow Science is not slowing down for its own sake,
but increasing quality. Slow science means getting into the nitty
gritty, just as the podding of fresh peas with your fingers is part of
the production of a high quality meal. Science is a slow, steady, methodical
process, and scientists should not be expected to provide
quick fixes for society’s problems.35
Frith’s argument is compelling, but what about the counterargument
that a complete deemphasis on the quantity of publications would invite
exploitation by researchers, and even laziness? How could two researchers
be compared if their work was judged of similar quality but one had simply
produced a greater amount than the other? Should they be judged equivalently?
The answer is that assessments of quality can take into account
quantity, but only indirectly: producing a larger volume of research would
increase the overall quality of a researcher’s contribution to science only if
the contribution to theory or practice rises in proportion with the amount
of work. On the other hand, it is possible that for some researchers, quality
may increase only slightly, or not at all, with quantity—for instance, a collection
of multiple papers reporting low-powered experiments could be
judged as the same or lower quality than a single paper describing a rigorous,
high-powered investigation. To avoid perverse incentives, the key requirement
in any quality-based system is that the assessment of individuals,
or comparisons between them, must be made independently of the sheer
number of publications.
Roads to Somewhere
In this chapter we have considered three ill-judged ways of assessing the
quality of psychological science and scientists—journal impact factors,
grant awards, and authorship rank. So are metrics and quality indicators
completely useless? Should we abandon them entirely?
No. It is natural to be drawn toward citation metrics. We all want our
work to be influential, and we take pleasure in it being cited because it
means someone read it and, with luck, found it useful.36 But there is a criti-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
The Sin of Bean Counting | 169
cal difference between things that make us feel good and the criteria we
should use to assess and value other scientists. Recognizing that difference
and applying it requires mental discipline. Metrics like JIF are meaningless
as indicators of scientific quality, but citations themselves at the level of
individual articles—while not indicators of quality—are suggestive of influence
and interest-value.37 There is no shame in paying attention to these
features of our academic landscape provided we see them for what they are
and, crucially, for what they are not. The lazy worship of metrics should
never replace informed expert judgment. Only expert appraisal can render
authoritative judgments on scientific quality (and even then it can often be
wrong). We allow metrics to supplant our subjective powers at our own
peril.
The same principle applies to grant funding. It is perfectly understandable
for scientists to celebrate the award of a grant that enables them to
pursue research that would otherwise have remained out of reach. Being
awarded funding carries its own intrinsic rewards—as researchers we now
get the resources to address an exciting and important question, and the
money in the grant can allow us to buy our time out of teaching and administrative
responsibilities within our university departments. As with citation
metrics, however, grant income should never be used to assess the quality
or potential of researchers because a grant is a mission plan, not a completed
mission. If grants are considered at all, they should be weighed against research
outputs: assessors should ask whether the scientific contribution of
this work was in proportion with the resources the researcher consumed.
Assessors should not simply reward the act of consumption.
Finally, what do we do about the convoluted mess that is academic authorship?
In the final chapter we will consider an alternative contributorship
model that borrows certain features from the physical sciences. Under a
contributorship model, the order of authorship is irrelevant—all that matters
is the work that authors actually undertook to earn their positions on
a paper. Authors thus earn credit for the extent of their contribution to different
roles, whether originator of the theory or hypothesis, developer of
methods, gatherer of data, and so on.
The sin of bean counting exists because we lack a robust way of assessing
the quality of scientific output in the short to the medium term. How do
we weigh up the scientific contributions of different individuals in the competition
for jobs and funding? How do we decide if a completed grant was
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
170 | Chapter 7
successful? Do we check whether independent scientists replicated the
funded studies, or at least that the studies were conducted rigorously and
are therefore replicable? Until psychology develops a proper mechanism for
attributing credit and assessing genuine research quality then bean counting
will continue to reign supreme.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
C H A P T E R 8
Redemption
The method of science, as stodgy and grumpy as it may
seem, is far more important than the findings of science.
—Carl Sagan, 1995
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Throughout the last seven chapters we have seen how psychological science
falls prey to a number of cultural sins, including bias, hidden analytic
flexibility, unreliability, lack of data sharing, vulnerability to fraud,
failure to make published work publicly available, and an obsession with
bean counting. Left unchecked, these sins not only hold back discovery
and practical applications of psychology; they also pose an existential threat
to the discipline itself. In an increasingly competitive world—a world in
which public spending on science is under intense pressure—psychology
can ill afford to bury its head in the sand and hope these failings will be
ignored.
We can anticipate two objections to this conclusion. On the one hand,
defenders of the status quo will argue that none of these failings are unique
to psychology. Questionable research practices and indifference toward replication
are present across much of the life and social sciences, and publication
bias is ubiquitous to varying degrees across all science. Lack of data
sharing, barrier-based publishing, fraud, and the worship of impact factors
threaten every field, not just psychology, and not even just science but the
humanities and social sciences too. So, why should we focus on psychology?
And why should it fall on psychologists to “fix” science?
The answer is simply this: as psychologists we must take responsibility
for the quality of the research we produce, regardless of what may be happening
in other fields. I focus on psychology because psychology is where
I live. If the roof of my house is leaking and threatening to collapse, I don’t
look at my neighbor’s house and say “They have the same problem so I don’t
need to worry about mine.” I fix my roof, and, in doing so, maybe I can
develop a solution that helps fix other roofs as well. But if everyone passes
the buck then repair never happens. The diffusion of responsibility is a key
reason why reform in psychology has been so slow, despite many of the flaws
identified here being known for decades.
At the opposite end of the spectrum lie those who will take these sins as
proof that psychology isn’t a science and should therefore be abandoned.
This is equally false. Precisely what makes an area of study a science is a
vexed question that has provoked strong and varied reactions from philosophers
for over a century. To adopt a Popperian perspective, a scientific discipline
can be said to be one in which the phenomenon under investigation
is quantifiable, the hypothesis testable, the experiment repeatable, and the
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 173
theory falsifiable. Much psychological research clearly meets all these conditions
even when its practitioners—psychological scientists—fail to meet
the highest standards in every area.1 For example, most studies in quantitative
psychological fields (such as psychophysics, cognitive psychology, experimental
psychology, and social psychology) include objective measurements
of behavior that are repeatable in principle. It’s just that few
experiments are exactly repeated and many findings fail to replicate when
they are, which can be explained by practices that are weak but nevertheless
scientific. In the same way, most hypotheses are precise enough to be
falsifiable even if researchers succumb to confirmation bias or hindsight
bias, but again these failings fall within the spectrum of scientific practices.
The susceptibility of psychology to these problems is comparable with other
areas of biomedical research, which have been shown to suffer from low
reproducibility, research bias, and publication bias.2 And lest the reader is
still tempted to cast psychology into the fire, we should remind ourselves
that in spite of the severity of these problems we have nevertheless assembled
a remarkable vault of knowledge about the mind and brain. Over the
last century, hundreds of robust psychological phenomena have been established
and fed into the development of theory. These insights have changed
the way we think about human cognition, behavior and society.
Advances aside, there is no denying that psychology is less robust than
many other sciences, in particular compared with the physical sciences.
Fields such as physics and chemistry are less prone to bias, place more emphasis
on replication, offer more precise and falsifiable theories, and manifest
more open practices. Psychology instead shares characteristics with
Richard Feynman’s famous definition of “cargo cult science.” Returning to
his 1974 essay, which we discussed in chapter 3:
In the South Seas there is a cargo cult of people. During the war they
saw airplanes land with lots of good materials, and they want the
same thing to happen now. So they’ve arranged to imitate things like
runways, to put fires along the sides of the runways, to make a
wooden hut for a man to sit in, with two wooden pieces on his head
like headphones and bars of bamboo sticking out like antennas—he’s
the controller—and they wait for the airplanes to land. They’re doing
everything right. The form is perfect. It looks exactly the way it
looked before. But it doesn’t work. No airplanes land. So I call these
things cargo cult science, because they follow all the apparent pre-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
174 | Chapter 8
cepts and forms of scientific investigation, but they’re missing something
essential, because the planes don’t land.
The good news for psychology is that the planes do land. The bad news
is that they don’t come as often as they should because we have failed to
heed Feynman’s warning that “you must not fool yourself, and you are the
easiest person to fool.” In this final chapter we will discover how psychology
can meet Feynman’s challenge and, in doing so, fortify and enrich its
mission.
Solving the Sins of Bias and Hidden Flexibility
In chapters 1 and 2 we saw how biased research practices weaken the published
evidence base in psychology. Confirmation bias sabotages various
stages of the scientific process, from publication bias in favor of positive and
novel results to the reliance on “conceptual replication” that can confirm
but never refute previous findings. We also saw how hindsight bias manifests
through the questionable practice of Hypothesizing After Results are
Known (HARKing) in which the researcher either knowingly or unknowingly
presents an unexpected result as though it was predicted all along.
Meanwhile, hidden flexibility in research analyses, including p-hacking,
inflates the rate of false discoveries. We have already discussed some of the
“nudge” incentives that can help reduce bias. Badges for rewarding open
practices, methodological checklists, and disclosure statements are all valuable
and useful initiatives. However there are fundamental limitations in
what they can achieve because, while they incentivize transparency and
unbiased practices, they don’t actually prevent them. We may never be able
to eliminate bias altogether from human nature, but there is one sure way
to immunize ourselves against its consequences, and that is peer-reviewed
study preregistration.
Registered Reports: A Vaccine against Bias
My earliest experience of publication bias came in 1999 with the rejection
of my first scientific paper.3 “The results are only moderately interesting,”
chided an anonymous reviewer. “The methods are solid but the findings
.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
174 | Chapter 8
cepts and forms of scientific investigation, but they’re missing something
essential, because the planes don’t land.
The good news for psychology is that the planes do land. The bad news
is that they don’t come as often as they should because we have failed to
heed Feynman’s warning that “you must not fool yourself, and you are the
easiest person to fool.” In this final chapter we will discover how psychology
can meet Feynman’s challenge and, in doing so, fortify and enrich its
mission.
Solving the Sins of Bias and Hidden Flexibility
In chapters 1 and 2 we saw how biased research practices weaken the published
evidence base in psychology. Confirmation bias sabotages various
stages of the scientific process, from publication bias in favor of positive and
novel results to the reliance on “conceptual replication” that can confirm
but never refute previous findings. We also saw how hindsight bias manifests
through the questionable practice of Hypothesizing After Results are
Known (HARKing) in which the researcher either knowingly or unknowingly
presents an unexpected result as though it was predicted all along.
Meanwhile, hidden flexibility in research analyses, including p-hacking,
inflates the rate of false discoveries. We have already discussed some of the
“nudge” incentives that can help reduce bias. Badges for rewarding open
practices, methodological checklists, and disclosure statements are all valuable
and useful initiatives. However there are fundamental limitations in
what they can achieve because, while they incentivize transparency and
unbiased practices, they don’t actually prevent them. We may never be able
to eliminate bias altogether from human nature, but there is one sure way
to immunize ourselves against its consequences, and that is peer-reviewed
study preregistration.
Registered Reports: A Vaccine against Bias
My earliest experience of publication bias came in 1999 with the rejection
of my first scientific paper.3 “The results are only moderately interesting,”
chided an anonymous reviewer. “The methods are solid but the findings
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 175
are not very important,” said another. “We can only publish the most novel
studies,” declared the editor as he frog-marched me and my boring paper
to the door. The experience was not what I expected. I was 22, a fresh-faced
PhD student in his first year, instilled with the belief that science was a
process of objective truth seeking. The very last thing I expected was that
a scientific journal—and one so highly respected in my field—would reject
a piece of theoretically and methodologically sound research based on the
results. It just didn’t make sense and seemed incredibly unfair. How could
we be expected to control the results? Didn’t this just turn publishing into
a lottery? Surely there had to be some mistake?
My supervisor wasn’t surprised. Although he had approved a draft of the
manuscript before submission, he had harbored reservations about the way
it was presented. “It didn’t really tell a story,” he said. “I could see what you
were trying to do, but you have to give the reviewers a narrative—something
they can get excited about—or they will find your paper boring and reject
it.” The conversation was formative in my training. “But the results are the
results,” I exclaimed. “Shouldn’t we just let the data tell the story?” He shook
his head. “That’s not how science works, Chris. Data don’t tell stories, scientists
tell stories.”
Data don’t tell stories, scientists tell stories.
Over the next decade I heeded those words. We told all kinds of stories
based on data. Big stories led to papers in zeitgeist journals like Nature Neuroscience,
Neuron, and PNAS. Smaller stories led us to solid specialist journals.
We’d carefully design and conduct experiments and then weave compelling
narratives out of the results. It was fun and we published well. We
were awarded grants. I got a career. Everybody won.
Or did they? Every paper had a story, but was that story the most unbiased
representation of the data? Of course not. We would gloss over imperfections
or noisy results that were hard to explain, burying the nuances in
“supplementary information” or not mentioning them at all. I’m certain
that I committed the sin of hidden flexibility many times, and no doubt at
an unconscious level even more than I realize. Every paper had a pithy logline
and a tale to tell. As hard as we worked on the science—and we worked
very hard to design experiments that were theoretically meaningful—we
worked tirelessly on the storytelling.
The strategy worked. In 2006 I was awarded a fellowship that allowed
me to move to the UK and establish my own research team at a major Lon-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
176 | Chapter 8
don university. Initially I practiced the same tried-and-true formula, one
that my new institute already embraced: novel experiment + great results
+ great storytelling = publication in a prestigious journal. To succeed you
needed every element in place; if even one was missing or below par then
your study would end up in a lower-ranking specialist journal or the file
drawer. By now, however, this style of science was starting to grate on me.
I had always found the publish-or-perish culture unappealing, and the pressure
at my new institute was higher than ever. As scientists, the one part
of an experiment we were supposed (in theory) to relinquish control over
was the results. To teach this but to nevertheless pin success on “good results”
was a devil’s temptation toward bias, questionable research practices,
and fraud. I was tired of watching colleagues analyzing their data a hundred
different ways, praying like gamblers at a roulette wheel for signs of statistical
significance. I was fed up with the inexorable analysis and reanalysis of
brain imaging data until a publishable story emerged from an underpowered
design. I also became dubious about the robustness of my own work. As a
friend dryly observed, “You guys do high-impact work, but you’re like magpies.
You do one study and move on to something else without ever following
it up.” He was right, of course. “That’s the game,” I admitted. “Why
would anyone waste time doing the same experiment twice when no funder
will pay for it and no top journal will publish it? And why would you take
the risk of failing to replicate yourself?” Talk about shooting yourself in the
foot.
As my doubts grew, the idealism of my younger self began to reassert
itself. My institute was packed to the rafters with brilliant people, but it felt
like the scientific equivalent of an elite telemarketing company. Walking in
the front door each morning you would face off against a wall-mounted
screen listing this week’s “top sellers”—the roll call of who published what
in which prestigious journal. Nature Neuroscience. Current Biology. Science.
The last-author credit would always belong to one of the professors, with
the first-author slot filled by a tireless protégé who barely left the building.
If your name wasn’t on that list (and mine usually wasn’t) you felt small,
inadequate, an imposter. You pushed yourself harder. You pushed your staff
and students harder. As the director of the institute at the time—and a
valued mentor—once told me: “This place is powered by appetite. You keep
the young researchers hungry. You keep them on fixed-term contracts with
uncertain futures, and you place them in competition with the world and
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 177
each other. The result is high-octane science.” I was never convinced that
he really believed in this philosophy, but it was the reality within those
walls.
After two years in London I left for a more stable academic career, and
admittedly a more relaxed professional lifestyle. For several years I continued
cranking the handle before a series of events reawakened my idealism
with a jolt. In 2011 we had a paper rejected by the Journal of Cognitive Neuroscience
for the main reason that because one of the experiments was a
close replication of the other, the study and results weren’t considered sufficiently
novel or important to be publishable. One of the reviewers even
told the editor in a private comment (obtained by us only after we had
unsuccessfully appealed the rejection): “The methods are strong, but the
outcome produced results that aren’t particularly groundbreaking.”
A year later I snapped. In September 2012 we received a rejection letter
from the specialist journal Neuropsychologia, in which our paper was declined
because one of the analyses (and not even a test that was important
for the conclusions) returned a statistically nonsignificant outcome. We
hadn’t glossed over it; we hadn’t massaged it away by p-hacking or HARKing;
we had simply reported it honestly and transparently. It was like 1999
all over again. Unpublishable results. Boring study. Reject. A few days later
I sat down and wrote a letter to the editors of Neuropsychologia, which I also
posted on my personal blog.4 I thanked them for their work on our paper
and made it clear that what I was about to say wasn’t personal. I had reviewed
for the journal and published in it many times, but I was now severing
all ties. “My real problem is that by rejecting papers based on imperfect
results, Neuropsychologia reinforces bad scientific practice and promotes
false discoveries. . . . For this reason, my main purpose in writing is to inform
you that I will no longer be submitting manuscripts to Neuropsychologia
or reviewing them. I will be encouraging my colleagues similarly.”
In the meantime I began to form an idea for a solution. I had been following
the work of blogger Neuroskeptic, who for several years had been
proposing study preregistration as a way to improve transparency and reproducibility.5
In his articles, and in the comments below them, you could
always find a vigorous debate about the potential benefits and drawbacks
of preregistration. Requiring authors to prespecify their hypotheses and
analysis plans held great promise for reducing a wide range of questionable
research practices, such as p-hacking and HARKing. Furthermore, if jour-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
178 | Chapter 8
nals could somehow be convinced to accept preregistered studies without
knowledge of the results then it would also prevent publication bias and
eliminate the incentive for authors to engage in questionable practices in
the first place. It seemed to me a brilliant solution to many problems. Suddenly
there would be no further need for excessive storytelling, no need to
gloss over inconsistencies and “messiness” of data. But how could you persuade
authors and journals to actually do this? Should it be mandatory?
Would the process be too cumbersome and bureaucratic? There were dozens
of unanswered questions, but I found the debates fascinating, and I had an
idea. I just needed a journal to try it in.
An opportunity soon came. A few weeks after publishing my open letter
to Neuropsychologia, the chief editor of Cortex, Sergio Della Sala, invited me
to join his editorial board. I accepted and immediately set to work on a
proposal for a new type of empirical article called a Registered Report in
which study protocols (including introduction and methods) would be reviewed
before authors gathered their data.6 Registered Reports stem from
the simple philosophy that the publishable quality of science should be
judged according to the importance of the research question and rigor of
the methodology, and never based on whether or not the hypothesis was
supported. This wasn’t a new idea and certainly wasn’t mine. In addition to
Neuroskeptic’s blog posts calling for preregistration, there had been several
proposals within the last 50 years for such a publishing mechanism. As far
back as 1966, psychologist Robert Rosenthal wrote:
What we may need is a system for evaluating research based only on
the procedures employed. If the procedures are judged appropriate,
sensible, and sufficiently rigorous to permit conclusions from the
results, the research cannot then be judged inconclusive on the
basis of the results and rejected by the referees or editors. Whether
the procedures were adequate would be judged independently of
the outcome.7
A few years later, in 1970, G. William Walster and T. Anne Cleary from
the University of Wisconsin offered the same idea:
In proposing this alternative policy, we argue that all decisions involving
the treatment of data should be considered design decisions.
Then, since the decision to publish the results of a study is particular
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 179
treatment of data, it follows that the same limitations should be imposed
on publication decisions as are imposed on all designs. When
one views publication in this way, it becomes immediately clear that
a specific change should be made in current policy. There is a cardinal
rule in experimental design that any decision regarding the treatment
of data must be made prior to an inspection of the data. If this
rule is extended to publication decisions, it follows that when an article
is submitted to a journal for review, the results should be withheld.
This would insure that the decision to publish, or not to publish,
would be unrelated to the outcome of the research.8
Neither Rosenthal’s nor Walster and Cleary’s proposals were ever implemented,
but perhaps Registered Reports could be our chance to finally do
so. In the model I had in mind, protocols considered scientifically important
and robust and that met strict guidelines for prospective rigor would be offered
“in principle acceptance.” The journal would then commit to publishing
the outcomes regardless of how the results turned out, provided the
authors adhered to their preregistered protocol, that various prespecified
quality checks were passed, and that the conclusions were based on the evidence
obtained.
Three days later my proposal was ready for feedback. I distributed it to
the editorial board, and at the same time I also posted it as an open letter
on my blog.9 I knew that publishing an open letter would be controversial
as journals are accustomed to considering such proposals behind closed
doors rather than in sight of the public. But I had resolved to take an open
route for two reasons. My foremost concern was that the proposal might
have a glaring flaw that I hadn’t considered, and the best way to find that
out was to crowd source critical feedback from the wider community, including
those who had commented on Neuroskeptic’s earlier proposals. And
secondly, I wanted the journal—and me—to be held accountable for whatever
decision it reached about Registered Reports. I had heard on the grapevine
that similar ideas had been mooted in the past at other journals and
binned by conservative boards amid concerns that accepting papers before
results could force the journals to publish negative or inconclusive outcomes.
And even though I had known Sergio Della Sala for many years and
respected him, the wider Cortex editorial board was an unknown quantity.
I had no idea how the board would react in private, but going public pro-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
180 | Chapter 8
vided a crucial test of the idea. If the journal decided to reject Registered
Reports then its reasons would need to be sufficiently defensible to survive
public scrutiny. And if the journal adopted them then the community, having
been involved since the beginning, could play a role in shaping the
initiative. In one case we would learn something important; in the other
we would hopefully do something important.10
Within a few days my open letter had attracted thousands of views and
dozens of comments below the line. The feedback contained many constructive
suggestions for improvement, most of which I integrated into the proposal.
However my strategy divided the Cortex editorial board. Some editors
were supportive of Registered Reports, but many were not, even though
none of the strongest opponents ever expressed their views to me personally.
Several board members also felt that I had risen above my station in
proposing the initiative in such a public way, just a few days after being
invited to join the editorial board.
Their response was understandable. I knew that the open letter would
seem like a coup to some, but I decided that the ire of a few editors was a
small price to pay for exposing Registered Reports to the crucible of public
opinion and giving it the best possible chance of a fair hearing. Fortunately,
the chief editor was strongly in favor. A month later, in November 2012,
Cortex became the first journal to approve Registered Reports. We assembled
an editorial subcommittee to handle submissions and prepared for the
launch in May 2013.11
How do Registered Reports work? Unlike conventional papers where
peer review happens after the entire study is finished and written up, here
the review process is split into two stages (see figure 8.1). At Stage 1, authors
submit an introduction, proposed methods and the analysis plan before they
have collected data. These are initially triaged for scientific significance,
clarity, and adherence to specific Stage 1 review criteria. I took a lot of time
to shape these criteria based on ideas and feedback from Neuroskeptic and
many others. Editors and reviewers at Stage 1 assess:
1. The significance of the research question(s)
2. The logic, rationale, and plausibility of the proposed hypotheses
3. The soundness and feasibility of the methodology and analysis
pipeline (including statistical power analysis)
4. Whether the clarity and degree of methodological detail would
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 181
be sufficient to replicate exactly the proposed experimental procedures
and analysis pipeline
5. Whether the authors provide a sufficiently clear and detailed description
of the methods to prevent undisclosed flexibility in the
experimental procedures or analysis pipeline
6. Whether the authors have considered sufficient outcomeneutral
conditions (e.g., absence of floor or ceiling effects; positive
controls) for ensuring that the results obtained are able to
test the stated hypotheses
Let’s take a moment to explore these points in more detail. The first two
criteria are designed to test the scientific credibility of the proposal. Is the
research question important? Do the hypotheses arise logically from the
literature? The third criterion tests whether the proposed methods are
rigorous and realistic, with particular emphasis on statistical power. As
we saw in chapter 3, underpowered experiments are common in psychology
and neuroscience, increasing the rate of false negatives and false positives.
At Cortex we decided to set a minimum statistical power of 0.9,
meaning that the experiment must have a sufficiently large sample that all
statistical hypothesis tests have a 90 percent chance of correctly rejecting
a false null hypothesis. We also invite authors to consider alternative Bayesian
sampling and hypothesis-testing methods in which statistical power is
irrelevant and data are simply acquired until a sufficiently decisive answer
is reached.
The fourth and fifth criteria focus on the extent of detail in the proposed
methods. Are the study procedures elaborate enough to provide a replication
“recipe” that could be repeated by other researchers? This takes into
account the major cause of unreliability in psychology (as discussed in chapter
3), that method sections are often too vague to permit replication. In
addition, are the proposed analyses sufficiently precise to prevent “wriggle
room” that could allow authors to either consciously or unconsciously exploit
researcher degrees of freedom such as p-hacking? Finally, the sixth
criterion requires authors to specify, in advance, what quality checks and
positive controls are required in order for the study to provide a fair test of
their hypothesis.12 An experiment might produce meaningless results because
equipment was wrongly calibrated, or because behavioral performance
from the participants was at ceiling or floor levels, or because a
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
182 | Chapter 8
condition with a known effect failed (i.e., a reality check or “positive control”).
To avoid publication bias, any such tests would have to be outcomeneutral,
which is to say they must be independent of the study hypotheses.
Keeping these tests independent of hypotheses prevents the common bias,
seen in status quo publishing, of authors and reviewers using positive or
negative outcomes to decide whether or not an experiment “worked,” and
to reject or accept it accordingly.
Manuscript withdrawn
Revision invited
Authors
decline
to revise
Authors
decline
to revise
Authors revise and
resubmit (stage 2)
Revision invited
Manuscript rejected
Manuscript withdrawn
Withdrawn registration is published
Manuscript rejected
Authors revise and
resubmit (stage 1)
Authors withdraw paper
In-principle acceptance (IPA)
Editorial triage
Stage 1 reviewers invited
Authors conduct study
Manuscript withdrawn
Withdrawn registration
is published
Manuscript rejected
Full manuscript acceptance
and publication
Stage 2 reviewers invited
Stage 1 Registered Report
Peer review of introduction, method,
proposed analyses, and pilot data (if applicable)
Stage 2 registered report
Peer review of introduction, method,
results, discussion
Figure 8.1. The submission pipeline and workflow for Registered Reports at Cortex and
several other journals.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 183
If the submitted manuscript passes editorial triage, it is then it is sent for
in-depth peer review where experts assess the proposed study rationale and
methods, testing it against the Stage 1 criteria. Following a favorable assessment,
which might require revision of the protocol, the journal can then
offer the paper “in-principle acceptance” or IPA. Only once IPA is obtained
can researchers then implement the preregistered experiment. Following
data collection and analysis, they resubmit a Stage 2 manuscript that includes
the introduction and methods from the original submission plus the
results and discussion. The results section of the completed manuscript
must include the outcome of all preregistered analyses. Crucially, however,
it can also include any additional unregistered analyses provided they are
clearly distinguished and identified as exploratory (post hoc) rather than
confirmatory (preregistered). The authors are also required to deposit their
data in a publicly accessible archive, addressing the sin of data hoarding. At
Stage 2 the reviewers, who are ideally the same as at Stage 1, consider the
following criteria:
1. Whether the data are able to test the authors’ proposed hypotheses
by passing the approved outcome-neutral criteria (such as
absence of floor and ceiling effects)
2. Whether the introduction, rationale, and stated hypotheses are
the same as the approved Stage 1 submission (required)
3. Whether the authors adhered precisely to the registered experimental
procedures
4. Whether any unregistered post hoc analyses added by the authors
are justified, methodologically sound, and informative
5. Whether the authors’ conclusions are justified given the data
The first of these criteria assesses whether the experiment passed the
preapproved quality checks and positive controls. For example, if the study
was on the effect of alcohol on cognitive function, did the authors confirm
that the alcohol had the necessary intoxicating effect through an established
questionnaire or other measure? If this positive control were to fail, then
this would suggest that the intervention was administered incorrectly, and
thus that the hypothesis was not properly tested. The second and third criteria
check for consistency between the introduction and methods sections
of the preregistered protocol with the same sections in the Stage 2 manuscript.
Finally, the fourth and fifth criteria check that any additional explor-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
184 | Chapter 8
atory analyses, and the overall study conclusions, are sensible. The finished
article is published only after this process is complete.13
Note what is missing from the Stage 2 review criteria. There is no assessment
of impact, novelty, or originality of results. There is no consideration
of how conclusive the results happen to be. There is no weight placed on
whether or not the experimental hypothesis was supported. Such considerations
may be relevant for scientists in deciding whether a study is exciting
or newsworthy, but they tell us nothing about scientific quality or the longerterm
contribution of the study. For Registered Reports, the outcomes of
hypothesis tests are irrelevant to whether a piece of scientific research meets
a sufficiently high standard to warrant publication.
At around the time that Cortex launched Registered Reports in May 2013,
similar initiatives began popping up at Perspectives on Psychological Science
and Attention, Perception, and Psychophysics.14 My impression was that preregistration
was fast transforming from theory into reality, but I also became
aware that some journals were quietly shelving the idea without transparent
debate. Those of us working to promote Registered Reports were
hearing that chief editors, petitioned behind closed doors to adopt the initiative,
were rejecting it on the grounds that it would lead to the publication
of negative findings that might be cited less than standard articles, leading
to a consequent drop in the journal’s impact factor. I found this worship of
impact factors deeply disappointing, and it reinforced my belief that closeddoor
politics are the antithesis of rational decision making (and are probably
the reason that Registered Reports had gone nowhere since the 1960s).
My colleague Marcus Munafò and I decided that something needed to
be done or the initiative could be killed by the sin of bean counting before
it got started. Three days after the Cortex launch we met at a pub in Bristol
and devised a plan. Over the next month we would assemble dozens of scientists
and members of journal editorial boards, and, in June 2013, we
published a joint open letter article in the Guardian headlined “Trust in
Science Would Be Improved by Study Pre-registration.”15 The article, which
was eventually signed by more than 80 senior academics, called for all life
science journals to offer Registered Reports as a new option for authors.
Registered Reports were not going to go quietly into the night.
The storm that followed was astonishing. The publicity of the Guardian
opened the debate to a much wider community of researchers, and two opposing
camps took shape. On one side were the reformers pushing for
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 185
greater transparency and reproducibility; they were generally in favor of
Registered Reports being adopted, if only to find out how the articles it
produced compared with standard publishing. On the opposite side was a
rearguard of (often powerful) traditionalists who argued that preregistration
would “put science in chains.”16 I found these criticisms puzzling.
Many were based on what appeared to be simple misunderstandings or
elementary errors about the initiative. Had these people even read what we
had written? Other reactions struck me as disingenuous misrepresentations
that appeared to have no purpose except to derail the initiative and preserve
the status quo. What follows are some of the main objections that emerged,
my personal interpretation of the objection, and a longer explanation in
each case:
• Registered Reports prevent exploration of data and curb scientific creativity.
Verdict: False. This is perhaps the greatest misconception
of the initiative. Authors of Registered Reports are welcome to
perform unregistered exploratory analyses with as much creativity
as they can muster, just as they would with standard publishing.
The only requirement is that such analyses are labeled transparently
as exploratory rather than being presented as confirmatory. I
was particularly surprised at how many traditionalists clung (and
still cling) to this argument, despite exploratory analyses being
formally invited as part of Registered Reports policies at all adopting
journals. It appears that some traditionalists not only want the
freedom to conduct exploratory analyses (which Registered Reports
explicitly welcome), but also want to be able to present those
exploratory analyses as confirmatory hypothesis testing. I reached
the rather unsettling conclusion that the traditionalists who continued
to oppose Registered Reports on these grounds simply
wanted the freedom to HARK, and because HARKing is socially
unacceptable the opponents had no choice but to argue that Registered
Reports would somehow prevent exploration.
• Registered Reports could lead to the denigration of exploratory, observational
research. Verdict: Unsubstantiated and probably false.
This is a subtler version of the first counterargument. The concern
here is that developing a more robust mechanism for hypothesis
testing would somehow lead to exploratory research (that is, re-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
186 | Chapter 8
search that doesn’t involve a priori hypotheses) being sidelined and
seen as second-class. However this concern is illogical and speaks
to a peculiar insecurity held by those who value exploration in science.
All that Registered Reports do is to clarify the distinction
between confirmatory hypothesis testing and more exploratory
forms of analysis and hypothesis generation, both of which are
valuable to science and welcomed as part of the format. But if the
mere act of distinguishing one from the other denigrates exploratory
research then what does that say about the value our community
places in exploratory research in the first place? High rates
of HARKing in psychology, as identified by Norbert Kerr and Leslie
John (see chapters 1 and 2), show that exploratory research is
indeed regularly crammed into an ill-fitting confirmatory framework,
shoehorning Thomas Kuhn into Karl Popper for the sole
purpose of achieving publication. Rather than criticizing Registered
Reports for adding clarity to a system that champions obfuscation,
why weren’t the traditionalists developing parallel publishing
initiatives that celebrate purely exploratory science? Indeed, if
exploration mattered so much to them, why hadn’t they done so
years ago?17
• Registered Reports can be gamed by “preregistering” a study that is
already completed. Verdict: True only for fraudsters. It’s a curious
sociological phenomenon that many otherwise reputable psychologists
genuinely believe this is a possibility. Do they really hold
their colleagues in such low regard? In any case, for Registered
Reports such a strategy would not only be fruitless but is impossible
without committing fraud. When authors submit a Stage 2
manuscript it must be accompanied by a laboratory log indicating
the range of dates during which data collection took place together
with a certification on behalf of all authors that no data (other
than pilot data in the Stage 1 protocol) was collected prior to the
date of IPA. Time-stamped raw data files generated by the preregistered
study must also be deposited in a public archive, with
the time-stamps postdating in-principle acceptance. Submitting a
Stage 1 protocol for a study that had already been completed would
therefore require complex falsification of laboratory records and
data time stamps. Even putting aside the fact that such behavior
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 187
would be clearly fraudulent, traditionalists who raise this concern
overlook a major problem with such a strategy: based on the comments
of reviewers, editors usually require changes to the proposed
experimental procedures following Stage 1 review. Even
minor changes to a protocol would be impossible if the experiment
had already been conducted, and would therefore defeat the purpose
of preregistration. Unless the authors were willing to engage
in further dishonesty about what their experimental procedures
involved—a degree of fraud that is beyond redemption—“preregistering”
a completed study would be a highly ineffective publication
strategy.
• Registered Reports won’t stop fraud. Verdict: Straw man. This was
a common reaction and another straw man because Registered
Reports are not designed to stop fraud. No publishing mechanism,
least of all the status quo, can protect science against complex
and premeditated acts of misconduct. What Registered Reports
achieve, above all, is to eliminate publication bias along with
the pressure to massage data, reinvent hypotheses, or behave dishonestly
in the first place.
• Registered Reports lock authors into publishing with a specific journal.
Verdict: False. Authors are free to withdraw their Registered Report
submissions at any time—there is no binding contract with
the journal. The only requirement is that study withdrawal after
IPA leads to the publication of a Withdrawn Registration, which
includes the abstract from the Stage 1 submission together with a
reason for the withdrawal. This ensures that the process is transparent
to the scientific community.
• Registered Reports fail to lock authors into publishing with a specific
journal. Verdict: Red herring. Some traditionalists have criticized
Registered Reports for exactly the opposite reason: that it could be
gamed by authors precisely because there is no binding contract
with the journal. Their argument goes like this. Suppose a researcher
has a Stage 1 protocol accepted in principle with a specialist
journal. They conduct their study but find something amazing
and unexpected in the results that they feel could be sold to
Nature or Science. What would stop them from withdrawing their
paper from the specialist journal and resubmitting it as a conven-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
188 | Chapter 8
tional (unregistered) article to a more prestigious outlet? The answer
is: nothing. But there is a catch. Authors would do so knowing
that their choice would be transparent to their peers because
withdrawing a paper after IPA triggers publication of a Withdrawn
Registration. It would interesting to see how an author’s peers
would react to a reason for withdrawing a Registered Report on
the grounds that: “After finding something remarkable and unexpected,
we decided we could publish this in a more prestigious
journal.” In many ways, such transparent careerism would be
refreshing.
• Registered Reports are not suitable for exploratory science or for developing
new methods where there are no hypotheses. Verdict: Red
herring. This is a common objection but irrelevant because the
format isn’t designed to be applicable to anything other than
hypothesis-driven science, which makes up a large bulk of published
research in psychology and beyond.
• Registered Reports are suitable only for one-shot experiments, not a
series of sequential experiments where the outcomes of one experiment
feed into the design of the next. Verdict: False. At many of the adopting
journals, authors can register experiments sequentially, each
time adding to the previous set. At each stage in the cycle the previous
version of the paper is accepted, eliminating the risk that the
addition of later registered experiments could jeopardize publication
of the earlier ones.
• Reviewers of Stage 1 submissions could steal my ideas and scoop me.
Verdict: Possible but highly unlikely. Scooping is the bogeyman
of science. Everyone knows someone who overheard someone
talking with someone about a story in which someone got scooped.
Maybe. In fact such cases are very rare.18 Concerns about being
scooped do not stop researchers applying for grant funding or presenting
ideas at conferences, both of which involve releasing ideas
to a much larger group of potential competitors than would typically
see a Stage 1 Registered Report (which usually isn’t published
until the study is completed). It is also noteworthy that once inprinciple
acceptance is awarded, the journal cannot reject the
Stage 2 submission because similar work was published elsewhere
in the meantime. Therefore, even in the unlikely event of a re-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 189
viewer rushing to complete a preregistered design ahead of the
authors, such a strategy would bring little career advantage for the
perpetrator, and would very possibly backfire.19
• If Registered Reports were mandatory or universal they would . . . Verdict:
Straw man and slippery slope fallacy. However this sentence
ends, the objection is irrelevant because we never proposed
that Registered Reports should be mandatory or universal—indeed
quite the opposite. The argument that Registered Reports
should be a universal option for hypothesis-driven research is quite
different to the argument (proposed by nobody) that it should be
obligatory across all science.
• A major previous discovery (e.g., mirror neurons) would never have
been possible under a Registered Reports model, therefore Registered
Reports would hold back science. Verdict: Arguably false but irrelevant
even if true. This concern is beside the point because Registered
Reports have never been suggested as a replacement for
exploratory science, only as an enhancement of hypothesis-driven
science. But even putting that fact to one side, how can we be sure
that discoveries such as mirror neurons wouldn’t have emerged
serendipitously from a Registered Report? A major misconception
of Registered Reports is that they hinder serendipity when in fact
they protect it. To illustrate this point, suppose you conducted a
standard (unregistered) study and found something serendipitous
that you believe is surprising and groundbreaking. What do you
suppose will happen when you submit your findings to a journal
through a conventional (unregistered) publishing route? Because
the results are surprising the reviewers are likely to be skeptical
about them, holding up publication for months or years while you
argue your case or run additional experiments. You might even
find that the barriers are too great and give up, dumping the results
in a file drawer. Now suppose you conducted exactly the
same study as a Registered Report. At Stage 2, reviewers can’t recommend
rejection on the basis of the results; therefore your serendipitous
finding is protected.20 This prompts us to turn the tables
and ask: how many serendipitous results such as mirror
neurons might have even been reported even sooner if they were
revealed within Registered Reports?
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
190 | Chapter 8
• We don’t need Registered Reports because we have replication. Verdict:
False. This argument ignores the fact that direct replication
in psychology is extremely rare and associated with many disincentives,
not least of which is the contempt shown by journals for
replication studies. Registered Reports, however, provide a perfect
avenue for replications by provisionally accepting papers before
authors invest the resources into conducting them. What better
incentive could you have for persuading authors to consider conducting
replication studies?
• We don’t need Registered Reports because protocols are already assessed
through grant reviews. Verdict: False. The first time I heard
this criticism I couldn’t believe the person was serious. Any psychological
scientist who has reviewed or applied for a major grant
knows that such applications contain nothing close to the level of
technical detail that is required for a Stage 1 Registered Report.
Grant applications propose the bigger picture of a planned program
of work; they rarely drill down into the specific details of
individual experiments to a degree that is required for a specific
Stage 1 protocol. And even for the occasional cases where a protocol
is sufficiently detailed, funded grant applications are almost
never published so who would know whether the researcher did
what they said they would?21 A private preregistration that is never
published is worthless to the scientific community.
• With publication virtually guaranteed, authors of Registered Reports
will conduct their experiments poorly, leading to meaningless results.
Verdict: False. This rather cynical objection emerges fairly regularly
from traditionalists. As one put it: “If you’re a young researcher
and you get your good idea preaccepted based on the
question and design, then it’s just more efficient to do a quick,
sloppy analysis and damn the results—after all, who cares? The
paper was already accepted. Time to move on to the next one.”22
Aside from portraying early-career scientists as little more than
ladder climbers, this argument ignores the fact that Stage 1 submissions
must include outcome-neutral tests and quality checks
for ensuring that the proposed methods are capable of testing
the stated hypotheses. Stage 2 submissions that fail any critical
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 191
outcome-neutral tests can be rejected, providing an inherent safeguard
against sloppy science. This objection also disregards the
fact that Stage 1 review involves stringent and detailed assessment
of the proposed analyses. It is no more possible for authors to conduct
a “quick, sloppy analysis” as part of a Registered Report than
for a standard conventional article—indeed it may be a lot less
likely for a Registered Report.
• The case for Registered Reports assumes that scientists act dishonestly,
and sends the message that there is no trust in the scientific community.
Verdict: Non sequitur and red herring. This argument
rests on the false premise that bad practice is synonymous with
deliberate deceit. As we have seen, however, bias and questionable
research practices can be unconscious or stem from ignorance
without implying any dishonesty. At a deeper level, the objection
misdirects us to place a greater emphasis on how psychological
science is perceived externally, and how researchers feel, than on
how the research is actually conducted. Regardless of whether bias
and questionable practices are conscious or unconscious, the solutions
are the same.
• Registered Reports are based on a naive view of the scientific method.
Verdict: False. Registered Reports provide a way to protect the
integrity of the deductive scientific method, but they do not elevate
deductive science above alternative exploratory approaches.
One might just as easily argue that a better drug for treating cancer
is “naive” because it doesn’t treat hepatitis. There is also a curious
inconsistency inherent in this viewpoint. Some traditionalists
may well believe that the hypothetico-deductive model is the
wrong way to frame science, but if so, why do the very same researchers
routinely publish articles that report p values and purport
to test a priori hypotheses? Are they merely pretending to be
deductive in order to get their papers published? Registered Reports
ensure that when researchers are truly engaging in deductive
science that they do it in as unbiased a way as possible and that
they are rewarded appropriately for doing so. Those who criticize
Registered Reports on these grounds are not actually arguing
against Registered Reports. They are criticizing the fundamental
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
192 | Chapter 8
way research is taught and published in the life sciences, despite
supporting that very system themselves and without proposing
any alternative.
• Registered Reports will overload the peer-review system. Verdict:
Unknown but probably false. It is true that Registered Reports
involve two stages of peer review at the same journal, each of
which is likely to involve at least one round of manuscript revision.
However, this is offset by the fact that authors are much less
likely to be successively rejected by multiple journals, as pointed
out nicely by neuroscientist Molly Crockett in response to my Cortex
open letter:
[T]he value of this system is that a given manuscript will
(ideally) only go through a single review process—so in
terms of collective hours spent reviewing papers, your proposal
may actually reduce the burden on the scientific community.
Consider the process we have now. Papers often
face a string of rejections before getting published (and
often rejections are based on data, not methods—e.g., null
findings). A given paper may go through the review process
at 3 or 4 different journals before getting published—so anywhere
from 6 to 12 (or more) reviewers may take the time to
review the paper. This is extremely inefficient both for reviewers,
and for authors, who must spend a substantial
amount of time re-formatting the manuscript for different
journals. None of this is time well spent. In contrast, the
extra time involved for authors and reviewers in your proposed
system *is* time well spent—the steps you outline
guard against all sorts of problems that are rife in the scientific
literature.23
• Registered Reports will lead to researchers bombarding journals with
protocols that have no funding or ethics and will never happen. Verdict:
False. As one critic said: “Pre-registration sets up a strong
incentive to submit as many ideas/experiments as possible to as
many high impact factor journals as possible.”24 Armed with IPA,
the researcher could then prepare grant applications to support
only the successful protocols, discarding the rejected ones. How-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 193
ever, this entire objection is beside the point because Stage 1 Registered
Reports must include a statement confirming that all necessary
support (e.g., funding, facilities) and approvals (e.g., ethics)
are already in place and that the researchers could start promptly
following IPA. Since these guarantees could not be made for unsupported
proposals, the concern is moot.
As we can see, many of the critical reactions to Registered Reports were
based on misunderstandings, logical fallacies, or ideological objections parading
as rational counterarguments. In the wake of the Guardian letter we
also faced a remarkable intensity of ad hominem attacks. Through various
channels we were accused of being “self-righteous,” “sanctimonious,” “fascists,”
“a head prefect movement,” “Nazis,” “Stasi,” “crusaders” on a “witch
hunt,” and worse. In one widely circulated e-mail, a professor who I happened
to know went so far as to belittle the 80 scientists who signed our
Guardian letter, stating: “Looking at the Chambers letter, I was struck by
the lack of scientific weight of the signatories.”25 That the mere suggestion
of a new type of article in science could provoke such aggression was telling
about what the aggressors sought to protect. As Niccolò Machiavelli wrote
over five centuries ago, “the innovator has for enemies all those who have
done well under the old conditions.”
Despite the fact that most of the critical reactions to Registered Reports
were in my view deeply flawed, several points did have merit. One concern
was that the time taken to review Stage 1 protocols could be incompatible
with short-term student projects, where students would usually lack the
time to wait for peer review and provisional acceptance before starting data
collection. There are at least two possible solutions to this problem. The
first is to accept that operating within such a rigid schedule is incompatible
with Registered Reports and either conduct such studies as unregistered or
instead preregister protocols for student projects in a database without peer
review, such as the Open Science Framework. A more radical solution
would be to reorganize undergraduate student projects into a daisy-chain
system where students work for several months on a Stage 1 protocol while
simultaneously implementing the provisionally accepted protocol from a
previous year’s student. Under this system, students would never implement
the specific protocol that they submitted for peer review but they would
nevertheless gain intensive training in all aspects of deductive science.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
194 | Chapter 8
A second concern is that the sin of bean counting (see chapter 7) could
make Registered Reports an unattractive choice for younger scientists. Because
Registered Reports set rigorous methodological standards, requiring
large sample sizes and higher statistical power, researchers can find that
their experiments take longer to complete than they otherwise would under
the status quo. As we saw in chapter 3, psychology and neuroscience are
endemically underpowered, which permits researchers to publish a higher
volume of lower-quality papers, rich in post hoc storytelling but making
only a limited and biased contribution to knowledge. Registered Reports
turn this equation upside down. Researchers who publish Registered Reports
are likely to publish a fewer number of larger and more credible papers;
however, as long as the community values quantity over quality then
a more credible publication record is not guaranteed to provide young scientists
with more secure careers in science—and this is a problem that can
be fixed only by senior scientists changing the way they assess junior researchers
for jobs, grants, and career progression. A related concern is the
conservatism of the most prestigious journals, which despite claiming to
publish the highest-quality research nevertheless rely on publication bias to
select which papers to accept. To compete in the current academic system,
young scientists need to be strategic in where they send their work, so if
Registered Reports were offered only within specialist journals their reach
and appeal would hit a glass ceiling. In the shorter term, the solution to this
problem is the central goal of the Registered Reports initiative: to see the
format launched within all journals that publish hypothesis-driven science,
regardless of prestige. In the longer term, the solution—as we will discuss
later—is to do away with journals altogether, rejecting the premise of “prestigious”
outlets and allowing the quality and contribution of individual studies
to speak for themselves.
A third limitation of Registered Reports is that it is unclear to what extent
it can be applied to analyses of preexisting data. We saw in chapter 4
how analysis of existing data archives can be used to answer important
questions that may not occur to the investigators who conduct the original
studies. This raises the question of whether analysis of preexisting data
could be preregistered under a Registered Reports mechanism without the
process being biased by the authors having prior knowledge of the outcomes.
A potential solution to this problem—at least for sufficiently “big” data—is
to consider such existing data sets as split-half discovery and replication
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 195
samples. The system would be analogous to a card game in which one player
deals while the other cuts the deck: After vowing on the record that they
had never viewed the dataset in question, authors would submit a proposed
analysis to the journal. If the proposal passes prestudy review and is provisionally
accepted, then the journal would then decide (using a random algorithm)
which half of the data is the discovery sample and which is the
replication sample—that is, a random cutting of the deck. Under this system,
a registered secondary analysis would be considered to produce a finding
of note only if it replicated in each subsample.
Finally, one important question is whether there is any evidence that
Registered Reports will be effective in reducing publication bias and questionable
research practices. As with any new initiative, prior evidence of
effectiveness cannot and does not (yet) exist.26 From a logical point of view,
barring failure of the peer-review process, prespecification of hypotheses
and analysis plans is guaranteed to eliminate the practices of p-hacking and
HARKING. Similarly, unless Daryl Bem was right all along about the existence
of precognition, accepting papers before results are known renders
them immune to publication bias. But whether eliminating these practices
will lead to more reproducible science is an empirical question that has yet
to be answered. In 2014, the International Journal of Radiation Oncology, Biology,
Physics launched a randomized trial of Registered Reports, results of
which are pending. Among other outcome measures, the editors are testing
whether the rate of positive, negative, and indeterminate results will differ
between studies that are assigned randomly to either a standard unregistered
format or to a Registered Report.27 There are also signs in the clinical
trials literature that preregistration, in general, may reduce publication bias
and/or researcher bias. A 2015 analysis of medical trials in the prevention
of heart disease found that, since the advent of mandatory clinical trial
registration in 2000, the percentage of published trials showing a statistically
significant effect dropped from 57 percent (17 out of 30) to just 8 percent
(2 out of 25).28 Preregistration might therefore help put the brakes on
false discoveries.
Three years after the launch of Registered Reports the tone of the discussion
has shifted. The storm of personal attacks has subsided, and the opposition
from many traditionalists appears to have softened. The debate is
far from over, but important progress has been made. Some who initially
resisted the initiative have become supporters, and we are seeing completed
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
196 | Chapter 8
examples of Registered Reports now appearing in the literature.29 At the
time of writing, more than 40 journals have adopted the format, and not
just in psychology and neuroscience but also in cancer biology, empirical
accounting, nutrition research, political science, and psychiatry. In addition,
by the time this book is in print, a number of “high-impact” journals
are likely to be offering them, including Nature Human Behaviour.30 The
initiative has been heralded by the UK Academy of Medical Sciences as one
of several promising solutions to improving research transparency and eradicating
publication bias.31 In parallel, more than 750 journals across the full
range of sciences have agreed to review their adoption of open science as
part of the Transparency and Openness Promotion (TOP) guidelines, a process
that involves the consideration of Registered Reports.32 In March 2014
we also established the Registered Reports Committee at the Center for
Open Science. This committee, which I currently chair, aims to develop
and promote Registered Reports as a new way to improve the credibility of
published research.
Perhaps the most significant step forward for Registered Reports came in
November 2015. On the 350th anniversary of launching the world’s first
scientific journal, the Royal Society officially launched the initiative within
its journal Royal Society Open Science. This is a major development not only
because it is the first endorsement of Registered Reports by a learned society,
but because it establishes the format well beyond psychology to cover the
full spectrum of more than 200 physical and life sciences.33 From here a
future beckons in which Registered Reports may become a popular format
for science in general, offered at all journals that publish the outcomes of
hypothesis testing. If our goal is to establish a reproducible knowledge base,
then the literature must include at least some papers that are published based
on theoretical value and methodological rigor, independently of results.
Preregistration without Peer Review
Registered Reports are rapidly becoming available across a range of scientific
fields including psychology; however in many cases journal-based preregistration
may not be possible for researchers. The authors’ preferred journal
may not yet offer the format, or, as with student projects discussed
earlier, the authors may need to preregister and commence data collection
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 197
immediately, with no time to go through the Stage 1 review process. Even
when Registered Reports are not an option, it is still possible to preregister
studies in one of several available online registries.34 Protocols deposited in
registries are not peer reviewed, but they offer authors the opportunity to
prespecify their hypotheses and analysis plans, which can help assure skeptical
reviewers that a particular hypothesis or analysis was preplanned
rather than post hoc. As psychologist Leif Nelson puts it: “In a world of
transparent reporting, I choose preregistration as a way to selfishly show
off that I predicted the outcome of my study.”35
Unreviewed preregistration has pros and cons compared with Registered
Reports. Because there is no peer review, it has the advantage of allowing
researchers to proceed sooner with data collection. On the other hand, the
lack of review means that a protocol published in a registry has no assurance
of leading to a peer-reviewed publication. Furthermore, unreviewed preregistration
is vulnerable to the criticism that it allows authors the opportunity
to publish vague protocols that are not replicable and that permit all the
usual researcher degrees of freedom. Therefore, peer-reviewed articles eventually
arising from unreviewed protocols could still contain p-hacking or
other forms of selective reporting. Evidence from the clinical trials literature
suggests that unreviewed preregistration is indeed prone to bias. A
2009 analysis found that of 147 adequately preregistered medical trials, 31
percent later changed their primary outcome measures, and 13.9 percent
were even “preregistered” after the completion of the study.36 A more recent
analysis of 89,204 trials found that 31.7 percent had changed their primary
outcome measures.37 The same team later found that, consistent with
p-hacking, alteration of primary outcome measures was associated with the
reporting of statistically significant outcomes.38 In the wake of these findings,
Ben Goldacre and his team established the compare project to monitor
the extent to which medical researchers adhere to their preregistered
protocols.39 The results so far have not been encouraging: at the time of
writing, 58 of 67 trials published since 2014 have either failed to report all
preregistered outcome measures or have silently added new “primary” outcomes
that were not preregistered.
Although preregistration is uncommon in psychology, limited evidence
points to similar problems with unreviewed protocols. In 2015, Annie
Franco and colleagues compared the preregistered protocols with final published
articles for 32 studies entered in the Time-sharing Experiments for
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
198 | Chapter 8
the Social Sciences registry, an initiative of the National Science Foundation.40
They found that 41 percent of published articles reported fewer experimental
conditions than in the associated protocol and that 72 percent
reported fewer outcome measures. Consistent with selective reporting practices,
including p-hacking and HARKing, effect sizes for outcomes that were
reported were consistently larger than for those that went unreported.
The warning for psychology here is clear. While unreviewed preregistration
offers greater protection against bias than no preregistration, the lack
peer review of the protocol—and the lack of comparison between the preregistered
protocol and the final report—means that the credibility of preregistered
studies will vary significantly. Unreviewed preregistration is
therefore likely to be less robust than Registered Reports, where peer review
and continuity between protocol and outcome is inbuilt.41
Solving the Sin of Unreliability
How can we make psychological science more robust and reproducible? In
chapter 3 we considered in detail the sin of unreliability, caused by indifference
(and in some cases hostility) toward replication, lack of statistical
power, publication of vague and nonreplicable methods sections, misuse
and misapprehension of frequentist statistical methods, and failure to retract
clearly unreliable papers from the scientific record. We also discussed
some of the possible solutions, including a greater emphasis on replication
and high statistical power, adoption of Bayesian inferential methods to
allow more meaningful hypothesis testing, increased adversarial collaborations
to overcome ego-driven biases, and the implementation of higher
methodological reporting standards at journals. Registered Reports have
the potential to provide a natural antidote to many of these problems. Because
they are reviewed and accepted in advance of data collection, authors
can rest assured that their efforts in replicating a previous study will not
end up in the file drawer. Registered Reports improve reliability by setting
minimum a priori power requirements and welcoming Bayesian inferential
methods. However, despite the great promise held by Registered Reports,
the format is currently too embryonic to provide a one-size-fits-all solution
for unreliability; moreover because it focuses on publishing (scientific outputs)
it may never provide an all-encompassing solution that addresses the
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 199
full range of the scientific process, including exploratory research and inputs
from funders. This raises the question of what other initiatives could
be instituted, both in psychology and in wider science, to incentivize robustness
and boost reliability. Here are some hypothetical initiatives that do not
yet exist but that could provide additional advantages:
Development of a Reproducibility Index. As noted in chapter 7, science is
knee deep in citation metrics that provide only limited value. One potentially
useful metric, however, may be the extent to which a previously published
finding has been replicated. The idea for a Reproducibility Index
was suggested in 2013 by Adam Marcus and Ivan Oransky, who proposed
it as a complement to the journal impact factor: “Rather than rate journals
on how often their articles are cited by other researchers, let’s grade them
on how well those papers stand the most important test of science: namely,
does the work stand up to scrutiny?”42 There are several ways such a metric
could be developed. One strategy would be to continuously monitor all
literature following the publication of a key article, with the paper’s score
regularly updated either positively if the central finding is replicated, negatively
if the central finding fails to replicate, or neutrally if no replication
attempts are published.43 The Reproducibility Index would thus provide a
running gauge of the known reliability of particular findings and papers. If
such a measure were rolled out widely across psychology, indexes at the
level of specific results could be clustered to indicate the evidential reliability
of subdisciplines, journals, and institutions, though as always the
reliance of any one metric would not be recommended.
The greatest barrier in developing a Reproducibility Index would be the
initial workload. Establishing the foundation of such an initiative would
require a careful retrospective analysis of most (if not all) of the psychological
literature, which amounts to hundreds of thousands, if not millions, of
published papers. Since the definition of what constitutes a “replication” is
in many cases subjective, it would be impossible to fully automate such a
project44—it would instead require the coordinated action of a large number
of experts within different fields to chart the “replication profile” of the
entire evidence base. Starting from the oldest paper, each individual assessment
would be based on an automated forward-search of all peer-reviewed
literature that either cited or otherwise undertook a sufficiently similar
methodology to the original study. With the original study and a manage-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
200 | Chapter 8
able subset of subsequent papers in hand, the assessor would then judge
whether (a) each of the newer studies performed a replication of the original
study, either direct (close) or indirect/conceptual—with the nature of any
variation noted; and (b) the extent to which that replication, if present,
succeeded. Successful and unsuccessful replications could lead to the original
study being scored positively or negatively accordingly, with greater
scores applied for closer replications and greater degrees of success or failure.
To maximize reliability, many different experts would ideally assess
each paper, with the overall score determined by an aggregate analysis of
the distribution of individual scores. Once the profile of the current literature
was in place, the project could then shift into a rolling mode, scanning
new entries to the literature as they appear. By taking advantage of machinereadable
tags in published papers, it is possible that rolling updating of the
Reproducibility Index could be at least partially automated.
Of course, the key question underlining such a project is who would
perform the daunting task of establishing such a mechanism? There is no
easy answer to this question, but the solution—if there is one—must lie in
crowd sourcing. According to records held by the International Union of
Psychological Science, it is estimated that there are approximately one million
psychology researchers worldwide, represented by 87 national professional
bodies.45 By distributing the workload across many thousands of
members of this population (much like peer review) we could make the
project feasible if additional funding could be attracted to provide centralized
administrative support.46 Achieving such a project would nevertheless
require an unprecedented unity of purpose within the psychological
community.
Pottery Barn rule by journals and funders. One innovative solution to unreliability
discussed in chapter 3 was Sanjay Srivastava’s Pottery Barn rule in
which journals would commit to publishing replication attempts of novel
findings that they previously published. This idea could also be extended
to include funding agencies, with a percentage of total science spending
ring fenced to support replication attempts of previously funded projects.
Although this would in the short term reduce the net funds available for
original research, in the medium term it could improve reproducibility
and also bring significant cost savings. Across the whole of biomedical sci-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 201
ence, it has been estimated that lack of reproducibility costs the United
States over $25 billion annually.47 Diverting a relatively small amount of
public money to support reproducibility and replication early in the scientific
process has the potential to reap significant benefits downstream,
both into terms of discovery and economic efficiency.
Regular multicenter replication initiatives. In any field where direct replications
are only rarely undertaken, it is vital that there are regular group-led
initiatives to monitor reproducibility. Brian Nosek and the Center for
Open Science have led the way in championing such approaches in psychological
science. In chapter 3 we covered the 2014 special issue in the
journal Social Psychology that reported replication attempts across a wide
range of studies, covering recent influential papers to seminal work of the
1950s. In 2015, Nosek and a team of 270 researchers followed up with an
even larger reproducibility project, this time attempting to replicate 100
major studies. They found that just over 1 in 3 prior results could be reliably
reproduced.48 Until replication and minimally biased practices are
standard in psychology, such initiatives will be crucial for assessing the
credibility of the evidence base. They must, therefore, continue to be supported
and expanded to cover more subfields.
Registered Reports funding schemes. To date, Registered Reports have been
offered only by journals, with authors required to have all necessary funding
and approvals in place before they submit a protocol. However, it is
possible to perform protocol review even sooner, integrating the review of
protocols by journals and funders to maximize efficiency and impact. According
to the proposed system, authors would submit their research proposal
before they have funding.49 Following simultaneous review by the
both the funder and the journal, the strongest proposals would be offered
financial support by the funder and in-principle acceptance for publication
by the journal. This model carries a range of potential benefits for all
stakeholders. The journal stands to benefit by publishing work that is both
cutting edge and reproducible—criteria that are difficult to achieve in
single papers published through conventional means. At the same time,
the funder stands to benefit by supporting a set of projects that are guaranteed
to be published in a respected journal, eliminating publication bias
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
202 | Chapter 8
and maximizing transparency. This ensures that the specific work supported
by the funder is undertaken; and there may also be administrative
efficiencies for the funder where the review process is managed, at least in
part, by the journal (with the funder of course maintaining a key role in
deciding which work is supported). Authors also stand to benefit by having
their publications accepted in a respected journal before they start
their research, and through a mechanism that not only minimizes researcher
bias but also eliminates the incentive for authors to engage in
biased research practices in the first place. The papers they produce from
such a mechanism are therefore likely to be impactful and well cited. Finally,
the scientific community, more widely, stands to benefit from increasing
the stock of research that is timely, important, transparent, and
reproducible.
Solving the Sin of Data Hoarding
In chapter 4 we witnessed the widespread lack of data sharing in psychological
science. Despite clear advantages for science as a whole, few researchers
deposit their data in public repositories, and most even fail to
share data with other researchers when requested to do so. Central among
the emerging reforms are the Peer Reviewers’ Openness (PRO) initiative
and the Transparency and Openness Promotion (TOP) guidelines. These
initiatives are very different but highly complementary. PRO mobilizes
the democratic power of peer reviewers to insist that archiving of data
and materials, or a publicly stated reason for not archiving, is a critical
prerequisite for the peer to provide an in-depth review. The TOP guidelines,
on the other hand, seek to encourage greater transparency at the
level of policy by asking journals and organizational signatories to evaluate
and publish their adherence to various transparency standards and
levels of adoption. Although the TOP guidelines do not require journals
to adopt open practices, the fact that their degree of compliance will be
published provides an incentive to adopt as high a level of transparency
as possible.
Both the PRO and TOP initiatives are tremendously promising because
they seek to ensure visibility and accountability of open science practices.
But they are not enough. For all publicly funded psychology research, it
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 203
TOP Guidelines Adopted by More Than 750 Journals and 60 Organizations
Level 0 Level 1 Level 2 Level 3
Citation Standards Journal encourages
citation of data,
code, and materials,
or says noth-
ing.
Journal describes
citation of data in
guidelines to authors
with clear
rules and exam-
ples.
Article provides appropriate
citation
for data and materials
used consistent
with journal’s
author guidelines.
Article is not published
until providing
appropriate citation
for data and
materials following
journal’s author
guidelines.
Data Transparency Journal encourages
data sharing, or
says nothing.
Article states
whether data are
available, and, if so,
where to access
them.
Data must be
posted to a trusted
repository. Exceptions
must be identified
at article sub-
mission.
Data must be
posted to a trusted
repository, and reported
analyses will
be reproduced independently
prior
to publication.
Analytic Methods
(Code) Transparency
Journal encourages
code sharing, or
says nothing.
Article states
whether code is
available, and, if so,
where to access it.
Code must be
posted to a trusted
repository. Exceptions
must be identified
at article sub-
mission.
Code must be
posted to a trusted
repository, and reported
analyses will
be reproduced independently
prior
to publication.
Research Materials
Transparency
Journal encourages
materials sharing,
or says nothing.
Article states
whether materials
are available, and,
if so, where to access
them.
Materials must be
posted to a trusted
repository. Exceptions
must be identified
at article sub-
mission.
Materials must be
posted to a trusted
repository, and reported
analyses will
be reproduced independently
prior
to publication.
Design and Analysis
Transparency
Journal encourages
design and analysis
transparency, or
says nothing.
Journal articulates
design transparency
standards.
Journal requires
adherence to design
transparency
standards for review
and publica-
tion.
Journal requires
and enforces adherence
to design
transparency standards
for review
and publication.
Study Preregistration Journal says noth-
ing.
Journal encourages
preregistration of
studies and provides
link in article
to preregistration if
it exists.
Journal encourages
preregistration of
studies and provides
link in article
and certification of
meeting preregistration
badge re-
quirements.
Journal requires
preregistration of
studies and provides
link and
badge in article to
meeting require-
ments.
Analysis Plan Prereg-
istration
Journal says noth-
ing.
Journal encourages
preanalysis plans
and provides link in
article to registered
analysis plan if it
exists.
Journal encourages
preanalysis plans
and provides link in
article and certification
of meeting registered
analysis
plan badge require-
ments.
Journal requires
preregistration of
studies with analysis
plans and provides
link and
badge in article to
meeting require-
ments.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
204 | Chapter 8
is in the public interest for archiving of study materials, analysis code, and
anonymized data to be mandatory, with the mandate enforced by universities
and funders, and possibly through legislation. Of course, in some cases
it may not be possible to allow full public access to such archives; some
materials may be copyrighted or subject to other legal restrictions, and
data might be potentially attributed to specific individuals even when the
researcher believes it is sufficiently anonymized. In balancing the ethical
risks of sharing data, researchers can conduct a risk assessment that takes
into account two factors: the sensitivity of the data acquired and the estimated
likelihood of anonymization being thwarted to reveal a participant’s
identity. In special cases, researchers may also have good reasons
to apply a time-limited embargo on data access, on the grounds that they
plan to publish several papers arising from a single, indivisible data set.
Such considerations should rightly influence the freedom of access to archived
data—questions of who can access it and when. However it is unacceptable
that such considerations determine whether data is archived in
the first place. All data should be archived because, sooner or later, data
that are not archived are lost to science and to future generations of scientists.
The vast majority of everyday studies in cognitive, experimental,
and social psychology could immediately release simple anonymized data
with minimal to no ethical concerns, and for those where risks are higher,
models of gated access—either temporarily or in the long term—can be
applied.
Replication Journal discourages
submission of
replication studies,
or says nothing.
Journal encourages
submission of
replication studies.
Journal encourages
submission of
replication studies
and conducts
results blind review.
Journal uses
Registered Reports
as a submission
option for
replication studies
with peer review
prior to observing
the study
outcomes.
At the time of writing, the TOP guidelines have been adopted by more than 750 journals and 60 organizations.
For their policy on data and materials to also be compliant with the PRO initiative, the journal or
organization would need to achieve Level 2 or above for the Data Transparency and Research Materials
Transparency standards. Table adapted from Nosek et al. (2015).
TOP Guidelines Adopted by More Than 750 Journals and 60 Organizations (cont.)
Level 0 Level 1 Level 2 Level 3
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 205
Solving the Sin of Corruptibility
In chapter 5 we journeyed beyond the common gray area of questionable
research practices to study the outer reaches of fraud. What we find is a
field that is powerless at preventing fraud and incapable of protecting
whistle-blowers. Of all the deadly sins, the sin of corruptibility is the least
forgivable because it perpetrates such grave injustices against the most vulnerable
researchers in our community. It is also the most challenging to
reform because many of the solutions will require coordination with external
organizations and lawmakers.
Random data audits. Acts of fabrication often leave a statistical footprint in
the data, as we saw with the cases exposed by Uri Simonsohn and alleged
by the Förster whistle-blower. However, the exposure of such fraud cases
should not depend on self-regulation by lone “data detectives.” What psychology
(and science more widely) needs is an external regulator to systematically
audit research data. With additional investment of resources,
this monitoring could be accomplished by existing mechanisms, including
the UK Research Integrity Office and the US Office of Research Integrity.
Monitoring could be dovetailed with data archiving: automated algorithms
could search available data in published articles and archived repositories,
either systematically or randomly depending on the breadth
and depth of the analyses, and with the potential for additional human
auditing in cases flagged as potentially suspicious. Like drug testing in
sport, rigorous data audits conducted in accordance with due process (including
the presumption of innocence) could be a powerful mechanism
for detecting and deterring acts of fraud.
Profiling of researchers. Could personality profiling of researchers help prevent
cases of fraud before they happen? In 2010, psychologists Kevin Williams
and colleagues from the University of British Columbia reported
that, of several personality measures, plagiarism in psychology undergraduates
was correlated most strongly with psychopathy.50 Assuming that this
relationship is causal and extends to data fabrication, this raises the question
of whether students and staff who score high on psychopathy may be
more likely to commit academic fraud. Should such individuals even be
weeded out of the system, as Kate suggested following her ordeal (see
.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
206 | Chapter 8
chapter 5)? Although prevention is innately appealing, such an approach
raises grave ethical and practical concerns. Williams and colleagues explain
why:
It seems unlikely that school boards and university senates would approve
of mass prescreening of students for psychopathy. Any attempt
to determine probability-of-expulsion in advance suggests an unsavory
“guilty until proven innocent” approach. . . . Even if prescreening
were to be approved, there is no established cutoff score for psychopathy
in nonoffender populations. . . . Even if scores were kept
confidential, labeling could be extremely harmful to the student.
The surveillance of high scoring individuals would be highly problematic
ethically and practically.51
Even though profiling is premature and arguably quite sinister, there is
clearly a need for more research on the personality characteristics of academic
fraudsters. At the same time, selection panels for PhD programs and
academic appointments should favor applicants with a proven track record
in honesty, rigor, and openness—characteristics that are likely to provide
a safeguard against fraud.
Criminalization of academic fraud. One of the most surprising facts about
academic fraud is that it is seldom prosecuted as a criminal offence. While
perpetrators face dismissal from their jobs or student candidature, and
their academic careers are usually terminated, they are unlikely to face
fraud charges. Diederik Stapel’s case stands as an exception to this rule—
recall from chapter 5 that he avoided trial in the Netherlands by pleabargaining
to 120 hours of community service, albeit a sentence usually
administered for minor offences. The prosecutor’s office concluded that
Stapel had “acted against scientific integrity” while also ruling, oddly, that
he had “not misused public funds, because they were given for doing research
that was performed,” even though the research was fraudulent.52
Had Stapel embezzled funds for direct personal gain, he would no doubt
have faced more serious criminal charges.
Is it acceptable that academic fraud is treated at most as a minor offence,
if an offense at all? Given how easy it is to get away with data fabrication,
the career advantages it brings, the public resources its wastes, and the collateral
damage it causes to students and colleagues, there is an argument
.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 207
that academic fraud should be added to the criminal statutes and prosecuted
as a more substantial offence. In 2014, such a mechanism was put to the
test in Australia in the case of Parkinson’s disease researchers Bruce Murdoch
and Caroline Barwood from the University of Queensland. Following
an internal investigation triggered by a whistle-blower, Murdoch and Barwood
resigned, and two of their coauthored papers were retracted under
allegations that the studies never took place. The university then referred
the case to the Queensland Crime and Corruption Commission, which,
following its own investigation, charged Murdoch with three counts of
fraud, seven counts of fraudulent falsification of records, and five counts of
general dishonesty.53 Barwood was also charged with six counts of fraud or
attempted fraud.54 In March 2016, Murdoch pleaded guilty to all charges
and was handed a two-year suspended jail sentence, with Barwood’s trial
still pending. This case may provide a basis for establishing similar legal
mechanisms in other countries.
Protection and support of whistle-blowers. Our academic system fails abysmally
to protect whistle-blowers. In the case of Diederik Stapel, the three
junior researchers who exposed him would almost certainly have had
more successful academic careers had they remained silent. And in Kate’s
case, the fact that she assisted and advised junior whistle-blowers led to
her being sacked by a vengeful line manager, followed by ten years of academic
exile. What can we do to better protect those who put their careers
on the line to expose misconduct? The first principle must be to ensure
that speaking out is as minimally disadvantageous as possible for whistleblowers.55
Institutions should adopt research integrity policies that commit
them to supporting the careers of whistle-blowers. For a postdoctoral
researcher like Kate, this could be extending a current salary so as to provide
researchers with the opportunity to continue their work, either independently
or under new line management. For a PhD student it could
mean offering extended candidature and the chance to make up lost
ground with a new project and supervisor. It must also be enshrined in law
that whistle-blowing presents no barrier to promotion or retention. Finally,
we must ensure that institutions have in place independent and
transparent mechanisms for dealing with allegations of misconduct because
it is rarely clear who whistle-blowers can safely approach with their
concerns.
.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
208 | Chapter 8
Replication. Direct replication by independent researchers is the ultimate
method for rooting out error in science, whether caused by fraud, bias, or
other mistakes.56 Nonreplications alone should never be considered diagnostic
of fraud, but they are the best way to ensure that the scientific record
self-corrects in the long run. A culture in which independent replication
of key findings is routine would also reduce the incentive to commit
fraud in the first place, because it would ensure that no single finding is
heralded as a “discovery” until it is has been independently repeated.
Solving the Sin of Internment
In chapter 6 we saw how much psychological science is “published” behind
paywalls that make it inaccessible to the public, including nonacademic
users. Of all the deadly sins, this is perhaps the easiest to remedy through
individual actions and the growing momentum toward open access (OA)
publishing by governments and funders. One simple action that all researchers
can take is to ensure that they publish in journals designated as “green”
by the Rights MEtadata for Open archiving (RoMEO) project.57 RoMEO
green journals allow posting of papers on a personal website or freely accessible
repository immediately after acceptance (i.e., without an embargo).
In the medium term, funders, institutions, and governments should further
advance policies that not only mandate the publication of publicly funded
research through OA routes, but which do so via fully OA journals rather
than hybrid OA.
In the longer term, the debate surrounding OA publishing gives us pause
to consider whether the current journal-based system is fit for purpose. Is
it right that corporate publishers generate billions in profits on the back of
free labor provided by scientists? Given that scientists perform all the essential
work—the research, the peer review, and the editing—why do we
even need publishers? The litmus test for any system facing obsolescence is
whether we would invent it, as it is now, if didn’t already exist. In their 2013
article (see chapter 7), Björn Brembs, Kate Button, and Marcus Munafò call
for a radical reinvention of the academic publishing system to place it under
the administration of university libraries, dropping the need for journals
and supporting the totality of the peer-reviewed literature at a fraction of
the current cost:
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 209
We therefore would favor bringing scholarly communication back to
the research institutions in an archival publication system in which
both software, raw data and their text descriptions are archived and
made accessible, after peer-review. . . . This reputation system would
be subjected to the same standards of scientific scrutiny as are commonly
applied to all scientific matters and evolve to minimize gaming
and maximize the alignment of researchers’ interests with those
of science.58
Neuroscientists Sam Schwarzkopf and Dorothy Bishop have proposed
similar ideas—a new platform for peer review and publishing that is divorced
from journals.59 Under Schwarzkopf’s proposed system, authors submit
their manuscript to a centralized review platform where it is considered
by editors and reviewers before publication. However, unlike most journals,
the editorial decision is based on scientific rigor rather than the usual criteria
of novelty or impact. Schwarzkopf’s system also opens up the possibility
of different types of review, including not only the traditional review of
the entire paper but also various forms of partial review including ratingonly
reviews where a larger number of reviewers score a paper without providing
any written feedback, specialist statistical review where the reviewer
focuses solely on the statistical validity of the analyses, and data review
where reviewers are tasked with checking the authors’ results by repeating
their data analyses using the supplied analysis scripts. Following favorable
reviews—and revision where necessary—the article, data, and materials
would be published under a Creative Commons Attribution license (CC-BY),
and each item would be separately citable to enable transparent reuse and
assignment of credit. To assure maximum transparency, the peer reviews
themselves would also be published alongside the final article, which could
either be signed or anonymized.
Both Schwarzkopf and Bishop propose a model in which study protocols
could be reviewed prior to results being gathered, in the same way that
journals currently consider Stage 1 Registered Reports. Under Bishop’s
proposal, the protocol could be submitted before funding has been acquired,
and even before a sufficient team has been assembled to conduct
the research. Publication of the protocol could then be used to attract collaborators,
similar to the Registered Replication Reports initiative offered
by Perspectives on Psychological Science. Following peer review, protocol ac-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
210 | Chapter 8
ceptance, and the formation of a suitable team, the published protocol could
then be used as a basis for a funding application. This process may be attractive
to funders because a large part of the peer review (which the funder
would normally coordinate) has already been undertaken. The system
would also be open to various other kinds of specialized article formats,
including review articles, exploratory (non-hypothesis-driven) empirical
articles, and methods development.
Once a paper is formally accepted on the platform, this new system turns
the tables on traditional publishing. In addition to gating publication
through prepublication review, the platform would also invite postpublication
review, where other scientists are able to comment on and rate the
accepted article. Those articles that are rated highly or that generate the
most traffic would rise to the surface and become more prominent on the
platform. Then, rather than authors submitting their work to zeitgeist journals
such as Nature or Science, these journals (and others) could instead bid
for articles on the platform. For example, they might invite authors to submit
a shorter version for their journal that emphasizes wider implications
(linked to the full version of the paper); alternatively, specialist science
writers based within journals could publish overviews of the most popular
or highest-rated papers.
By separating peer review and publishing from journals, this system has
numerous advantages over traditional scholarly publishing. As Brembs and
colleagues point out, it would be much cheaper than maintaining expensive
journal subscriptions and OA publishing costs and would thus permit the
redirection of university library budgets toward supporting the coordination
and administration of the platform. The platform would also ensure that
publication of academic research is based on the theoretical importance and
methodological rigor of a study, with factors such as novelty and impact relevant
only in determining the newsworthiness to secondary users after publication.
And finally, it would forever solve the sin of internment by ensuring
that fully OA publishing is an intrinsic product of the research process.
Solving the Sin of Bean Counting
In chapter 7 we considered the various ways in which psychology is failed
by an academic system that idolizes superficial citation metrics, treats re-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 211
search grants as outputs, and remains in the grip of an archaic system of
academic authorship where assumptions about individual contributions are
based on the crudest of measures: authorship rank.
A potential solution to the problem of authorship rank would be to replace
it with a more transparent contributorship model that identifies the
specific involvement in the research made by the different authors.60
Under a contributorship model, for every publication, authors would be
listed in alphabetical order, with authors listing their percentage contribution
to each part of the paper.61 Potential categories of contribution could
include: theory and background, study design, data acquisition, provision
of analytic tools/reagents, data analysis, interpretation of results, and
manuscript preparation. Percentages would be allocated in two ways: relative
to the sum of work completed by all authors on the paper, and relative
to the totality of each individual’s contribution. Each of these statistics
would provide useful information to different users. On the one hand, the
percent contribution relative to the paper would tell the community which
authors contributed most to different aspects of a research project. If we
take an example of a cognitive neuroscience paper led by a postdoctoral
researcher and principal investigator, together with advice offered by a
statistician, it might be the case that the postdoc did 100 percent of the
data collection, and that the postdoc and statistician divided up the data
analysis between them 80 percent to 20 percent. The principal investigator,
however, might have had no involvement in the data acquisition or
analysis but might have had the most substantial contribution to the theory
and background and the study design, dividing this 60 percent to 40
percent with the postdoc. To comply with established conditions for authorship,
all authors would also be expected to make a contribution to
manuscript preparation.
On the other hand, the percent contribution to different domains relative
to the author’s total contribution could provide useful information for assessors
in constructing profiles of individual researchers. By averaging contributions
across publications within the different categories, a researcher
might develop a profile as an all-rounder who takes a hand in all aspects of
their research, or as an implementer who conducts studies and interprets
the results but plays less of a role in theory and study design. Alternatively
a researcher might emerge as an analyst who specializes in statistical methods,
or as a theoretician and designer—as would be typical of many princi-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
212 | Chapter 8
pal investigators. Profiles are also likely to change over time: as researchers
become more senior and progress from being postdocs to principal investigators,
their profiles may also switch from being all-rounders or implementers
to being theoreticians and designers.
Replacing traditional authorship with a contributorship model has the
advantage of achieving maximum transparency, but as with all systems that
generate quantitative outputs, it is vital that such information is not overinterpreted.
One risk of a contributorship model is the community drawing
a false sense of precision from the percent contributions, which are likely
to be uncertain in many cases.62 This presents a significant danger for individual
profiling, where, for example, funding agencies or university departments
might decide to use contributorship profiles to triage applicants
for grants or jobs. While this is a genuine threat, clinging to the traditional
authorship model does little to protect us from such bean counting, as job
applicants are already routinely triaged according to their quantity of firstauthored
publications or grants.
In the long run, to solve the sin of bean counting we must answer a fundamental
question: how do we judge the quality of research and researchers?
There is no metric, no simple heuristic, no set of calculations or algorithms.
Perhaps we can take a cue from the physical sciences, which have
long embraced subjective quality assessments by independent experts. This
unattributed quote provided by pharmacologist David Colquhoun shows
how one chemistry department assesses research excellence:
The greatness of a faculty member is not judged simply by the members
of the Department but rather by letters we collected, typically 10 to 15,
from experts outside the Department, nationally and internationally.
The question we ask of these experts is whether the research of the candidate
has changed the community’s view of the nature of chemistry in
a positive way. It is not based on how much funds the candidate had
brought to the University in the form of grants. It is not based on the
number of published papers. It is not based some elaborate algorithm
that weighs publications in journals according to the impact factor of
the journal. It is based simply on establishing new knowledge. As a Department
we do not discuss h-index metrics and we do not count publications
or rank them as to who is first author. We just ask, has the candidate
really changed significantly how we understand chemistry.63
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 213
Concrete Steps for Reform
Having reviewed the seven deadly sins and a range of possible solutions,
what can each of us to do to safeguard the future of psychological science?
What follows are series of suggested actions that we can take as researchers,
organizations, and nonacademic citizens.
For junior and senior researchers:
1. Be aware of the dangers of confirmation bias, low statistical
power, and p-hacking. For studies with clear a priori hypotheses,
submit your work as a Registered Report or preregister it on an
available registry such as Open Science Framework.
2. Be aware of hindsight bias and Hypothesizing After Results are
Known (HARKing). For work that is truly exploratory, don’t
write up the paper so that it presents an a priori hypothesis in
the introduction.
3. When writing the method section of an article, declare your degree
of transparency by stating the “21 word solution” by proposed
by Simmons, Nelson, and Simonsohn (2012): “We report
how we determined our sample size, all data exclusions (if any),
all manipulations, and all measures in the study.”64
4. Sign the Peer Reviewers’ Openness initiative.65
5. Adopt Bayesian hypothesis testing. Become familiar with the interpretative
limitations of p values and conventional null hypothesis
significance testing.
6. Avoid the trap of assuming that every new result published in
the psychological literature is a true positive and a definite fact.
Science doesn’t proceed one fact at a time; it is more like the
process of reclaiming land. If a particular finding of interest
hasn’t been directly and independently replicated then be cautious
about extending it and applying it in ways that depend on
it being credible.
7. Publish your research in RoMEO green or blue journals, prioritizing
outlets that allow immediate (embargo-free) archiving of
accepted papers. Never publish in journals that require subscription
access and that don’t permit archiving.66
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
214 | Chapter 8
For senior researchers:
1. Participate in joint replication initiatives in your field for the
common good.
2. Implement policies within your laboratory, department, or institute
for standardized publicly archiving of data, code, and
materials.
3. Establish a “data partner scheme” with other labs for mutual
checking of data analyses and curation strategies.67
4. Take signs of research misconduct seriously. Assume responsibility
for dealing with fraud before it becomes someone else’s
problem.
5. Avoid judging job applicants on the basis of journal impact factors,
grant income, quantity of first-authored publications, or
any other superficial metric. Instead, ask applicants and their
referees to describe their best papers, their contribution to the
work, and what the research contributed to theory or practice.
Value any proven commitment to open science practices and
principles.
6. Sign the San Francisco Declaration on Research Assessment
(DORA) as an individual.68
7. Above all, remember that junior scientists will learn by your example.
If you indulge in the deadly sins, your protégés will too.
And, in time, so will theirs.
For journals, funders, learned societies, or universities:
1. Sign the Transparency and Openness Promotion guidelines.69
2. For journals: If your journal publishes empirical papers that report
the outcomes of hypothesis testing then offer Registered
Reports as a new article option.70
3. For journals: Offer badges for open practices and require authors
to complete methodological checklists at submission.71
4. For journals and funders: Introduce a policy for sharing of data
and study materials that complies with the Peer Reviewers Openness’
Initiative.
5. For universities and funders: Never judge the track record of applicants
based on their record of grant income. Income should
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 215
be either ignored or balanced against the contribution of the
published outputs to theory or practice.
6. Sign DORA as an organization.
For journalists and citizens:
1. For journalists and readers: Adopt a critical mindset when writing
or reading news stories reporting the latest psychological
science. Has the study been replicated, either independently or
by the authors? Was the sample size large enough to justify the
conclusions that are being drawn? What measures were put in
place to counteract sources of research bias? Remember that extraordinary
claims require extraordinary evidence.
2. For journalists: Dig deeper in investigating the basis for statistical
claims in psychological research. For instance, if researchers
are claiming that a particular measure is the same between different
groups, are they basing this interpretation (incorrectly)
on a statistically nonsignificant p value, or (correctly) on a Bayes
factor? If the paper reports p values, what was the statistical
power of the study? Ask the authors to estimate the chances that
any obtained positive effects in their paper are true positives
(i.e., the positive predictive value).
3. For journalists: As well as reporting the findings of published
research, take into account its degree of transparency. Did the
authors make their data publicly available? Did they preregister
their study predictions before data collection? Did they publish
their work through an OA pathway? Journalism has an important
role to play in ensuring that scientists are held accountable
for conducting transparent and reproducible research.
Coda
At the 2014 Association for Psychological Science conference in San Francisco,
I gave a seminar on Registered Reports as part of a symposium called
“The Replication Revolution: One Year On.” It was a talk I had given many
times before, but this time felt different. After over a year of giving presen-
.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
216 | Chapter 8
tations on preregistration—mostly as an isolated voice—I now found that I
was just one of many speakers advocating reform, among a lineup that included
Bobbie Spellman, Brian Nosek, E. J. Wagenmakers, and Jelte
Wicherts. There was also something quite unique about this meeting compared
with the typical humdrum of an academic conference. Just hours
earlier, some senior American psychologists had publicly accused proponents
of open science of being “second stringers,” “bullies,” and “replication
mafia.” As the audience filed in, I wondered if that was really how psychologists
saw us? Had the debate surrounding reform in psychology really become
so politicized?
I have no idea how many psychologists attended that symposium—I
would guess close to a thousand—but the atmosphere was electric. Spellman
began by suggesting that psychology is in the midst of a political revolution,
pushing transparency and reproducibility up the agenda to sit alongside
the traditional aspirations of novelty and creativity. Along the way,
attacks were to be expected; not everyone would agree with the current
course—and in the questions that followed, it was clear that many did not.
One year later, a physicist from CERN approached me after I gave an updated
talk in London at a meeting of the Academy of Medical Sciences.
Physics took this journey long ago, he said, and he was glad to see psychology
moving the same way. “I am convinced that you are doing the right
thing to turn your field into a good area of science,” he later wrote. “I realise
that you will have to fight off some nasty dinosaurs but I am convinced you
will win—just look at what happened to the dinosaurs.”
He was of course referring to the opponents of reform in psychology,
Machiavelli’s winners. But the truth is that the dinosaurs are within all of
us—they are our unconscious biases, fragile egos, and propensity to cut
corners. A dinosaur rears its head every time we are tempted to p-hack or
change a hypothesis; every time we covet data as personal property; every
time we judge each other based on superficial metrics. The dinosaur is the
subjugation of our true mission to the ideals of storytelling, glory hunting,
game playing and bean counting. Therefore, this book should not be seen
as an attack on defenders of the status quo but as an intervention on ourselves,
and a mission plan for self-improvement. As the saying goes, there
is nothing noble in being superior to your fellow person—true nobility is
being superior to your former self.
.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Redemption | 217
If we succeed then the future of psychological science is bright. We can
look forward to deeper integration with computer science, biology, and
physics, led by a culture of transparency that makes our work dependable
and our data and tools reusable. Rather than looking back on psychology
as a flawed indulgence, future generations of scientists will see the current
period as a renaissance—a time when psychology left the cargo cult behind
and became a truly rigorous and open science.
.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
NOTES
Preface
1. For examples of psychology-related REF impact case studies, see “Assessing
the viability of electric vehicles for daily use,” http://impact.ref.ac.uk
/CaseStudies/CaseStudy.aspx?Id=17132; “Facilitating intervention based on
an enhanced understanding the antecedents and outcomes of debilitating
exam-related anxiety,” http://impact.ref.ac.uk/CaseStudies/CaseStudy.aspx
?Id=43758; “Fundamental research on memory enables a robust criminal
justice system,” http://impact.ref.ac.uk/CaseStudies/CaseStudy.aspx?Id
=30209; “Vision science and road Safety,” http://impact.ref.ac.uk
/CaseStudies/CaseStudy.aspx?Id=30208; “Human factors and space exploration,”
http://impact.ref.ac.uk/CaseStudies/CaseStudy.aspx?Id=27640; “Cardiff
research supports the creation of the UK Climate Change Committee,”
http://impact.ref.ac.uk/CaseStudies/CaseStudy.aspx?Id=3480; “Influencing
international tobacco policy on standardised tobacco packaging,” http://
impact.ref.ac.uk/CaseStudies/CaseStudy.aspx?Id=40192.
Chapter 1. The Sin of Bias
1. Daryl J. Bem, “Feeling the future: Experimental evidence for anomalous retroactive
influences on cognition and affect,” Journal of Personality and Social
Psychology 100, no. 3 (2011): 407, http://dx.doi.org/10.1037/a0021524.
2. Peter Aldhous, “Is this evidence that we can see the future?,” New Scientist,
11 November 2010, www.newscientist.com/article/dn19712-is-this-evidence
-that-we-can-see-the-future.html.
3. http://www.newscientist.com/article/dn20447-journal-rejects-studies
-contradicting-precognition.html. French and Ritchie’s nonreplication of
Bem was eventually published in PLOS ONE: Stuart J. Ritchie, Richard Wiseman,
and Christopher C. French, “Failing the future: Three unsuccessful attempts
to replicate Bem’s ‘Retroactive Facilitation of Recall’ Effect,” PLOS
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
220 | Notes to Chapter 1
ONE 7, no. 3 (2012): e33423, http://dx.doi.org/10.1371/journal.pone
.0033423.
4. Eric-Jan Wagenmakers, Ruud Wetzels, Denny Borsboom, and Han L. J. van
der Maas, “Why psychologists must change the way they analyze their data:
The case of psi; Comment on Bem (2011),” (2011): 426, http://dx.doi.org/10
.1037/a0022790. This article is available freely at http://www.ejwagenmakers
.com/2011/WagenmakersEtAl2011_JPSP.pdf.
5. Georg Henrik von Wright, A Treatise on Induction and Probability (Oxfordshire:
Psychology Press, 1951), 7: 86.
6. Peter C. Wason, “On the failure to eliminate hypotheses in a conceptual
task,” Quarterly Journal of Experimental Psychology 12, no. 3 (1960): 129–40,
http://dx.doi.org/10.1080/17470216008416717.
7. Peter C. Wason, “Reasoning about a rule,” Quarterly Journal of Experimental
Psychology 20, no. 3 (1968): 273–81, http://dx.doi.org/10.1080/14640746
808400161.
8. For a definitive review of confirmation bias and its suspected origins, see
Raymond S. Nickerson, “Confirmation bias: A ubiquitous phenomenon in
many guises,” Review of General Psychology 2, no. 2 (1998): 175.
9. Saul M. Kassin, Itiel E. Dror, and Jeff Kukucka, “The forensic confirmation
bias: Problems, perspectives, and proposed solutions,” Journal of Applied Research
in Memory and Cognition 2, no. 1 (2013): 42–52, http://dx.doi.org/10
.1016/j.jarmac.2013.01.001.
10. Alan G. Gross, “The roles of rhetoric in the public understanding of science,”
Public Understanding of Science 3, no. 1 (1994): 3–23, http://dx.doi.org/10
.1088/0963–6625/3/1/001.
11. Joshua Klayman and Young-Won Ha, “Confirmation, disconfirmation, and
information in hypothesis testing,” Psychological Review 94, no. 2 (1987):
211, http://dx.doi.org/10.1037/0033–295X.94.2.211.
12. Hugo Mercier and Dan Sperber, “Why do humans reason? Arguments for an
argumentative theory,” Behavioral and Brain Sciences 34, no. 02 (2011): 57–
74, http://dx.doi.org/10.1017/S0140525X10000968.
13. For early discussions of publication bias, the reader is directed to Theodore
D. Sterling, “Publication decisions and their possible effects on inferences
drawn from tests of significance—or vice versa,” Journal of the American Statistical
Association 54, no. 285 (1959): 30–34, http://dx.doi.org/10
.1080/01621459.1959.10501497. The issue was revisited throughout the
1960s and 1970s, for instance by Robert Rosenthal in 1979: Robert Rosenthal,
“The file drawer problem and tolerance for null results,” Psychological
Bulletin 86, no. 3 (1979): 638, http://dx.doi.org/10.1037/0033–2909.86.3
.638.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 1 | 221
In 1995, Sterling himself reassessed the problem and concluded that the
“practice leading to publication bias have not changed over a period of 30
years.” See Theodor D. Sterling, W. L. Rosenbaum, and J. J. Weinkam, “Publication
decisions revisited: The effect of the outcome of statistical tests on
the decision to publish and vice versa,” American Statistician 49, no. 1 (1995):
108–12, http://dx.doi.org/10.1080/00031305.1995.10476125.
14. Daniele Fanelli, “Positive” results increase down the hierarchy of the sciences,”
PLOS ONE 5, no. 4 (2010): e10068, http://dx.doi.org/10.1371/journal
.pone.0010068.
15. www.nature.com/nature/authors/gta/others.html.
16. https://www.elsevier.com/journals/cortex/0010-9452/guide-for-authors.
17. http://brain.oxfordjournals.org/for_authors/general.html.
18. http://www.oxfordjournals.org/our_journals/cercor/for_authors/general
.html.
19. Interview with outgoing editor of Psychological Science, Robert Kail: “Reflections
on five years as editor,” www.psychologicalscience.org/index.php/publi
cations/observer/2012/november-12/reflections-on-five-years-as-editor.html.
20. Parts of this section have been adapted from an article on my personal blog:
http://neurochambers.blogspot.co.uk/2012/03/you-cant-replicate-concept
.html.
21. John A. Bargh, Mark Chen, and Lara Burrows, “Automaticity of social behavior:
Direct effects of trait construct and stereotype activation on action,”
Journal of Personality and Social Psychology 71, no. 2 (1996): 230, http://dx.doi
.org/10.1037/0022–3514.71.2.230.
22. “Elderly-related words prime slow walking (#15),” http://www.psychfile
drawer.org/replication.php?attempt=MTU%3D.
23. “A primer for how not to respond when someone fails to replicate your work
with a discussion of why replication failures happen” by Dan Simons: http://
plus.google.com/+DanielSimons/posts/VJH8wXxxc3f.
24. Stéphane Doyen, Olivier Klein, Cora-Lise Pichon, and Axel Cleeremans,
“Behavioral priming: It’s all in the mind, but whose mind?,” PLOS ONE 7,
no. 1 (2012): e29081, http://dx.doi.org/10.1371/journal.pone.0029081.
25. John Bargh, “Nothing in their heads.” The original post by Bargh was removed
by Psychology Today but can still be read here: http://web.archive.org
/web/20120308225833/http://www.psychologytoday.com/blog/the-natural
-unconscious/201203/nothing-in-their-heads.
26. For Ed Yong’s excellent coverage of this debate, see “A failed replication
draws a scathing personal attack from a psychology professor,” http://blogs
.discovermagazine.com/notrocketscience/2012/03/10/failed-replication
-bargh-psychology-study-doyen/.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
222 | Notes to Chapter 1
27. Jay G. Hull, Laurie B. Slone, Karen B. Meteyer, and Amanda R. Matthews,
“The nonconsciousness of self-consciousness,” Journal of Personality and Social
Psychology 83, no. 2 (2002): 406, http://dx.doi.org/10.1037/0022–3514
.83.2.406.
28. Joseph Cesario, Jason E. Plaks, and E. Tory Higgins, “Automatic social behavior
as motivated preparation to interact,” Journal of Personality and Social Psychology
90, no. 6 (2006): 893, http://dx.doi.org/10.1037/0022–3514.90.6
.893.
29. “Priming effects replicate just fine, thanks” by John Bargh: http://www
.psychologytoday.com/blog/the-natural-unconscious/201205/priming-effects
-replicate-just-fine-thanks.
30. “How valid are our replication attempts” by Rolf Zwaan: http://rolfzwaan
.blogspot.be/2013/06/how-valid-are-our-replication-attempts.html.
31. For additional critique of conceptual replication, see Harold Pashler and
Christine R. Harris, “Is the replicability crisis overblown? Three arguments
examined,” Perspectives on Psychological Science 7, no. 6 (2012): 531–36,
http://dx.doi.org/10.1177/1745691612463401.
32. Norbert L. Kerr, “HARKing: Hypothesizing after the results are known,” Personality
and Social Psychology Review 2, no. 3 (1998): 196–217, http://dx.doi
.org/10.1207/s15327957pspr0203_4.
Although Kerr was the first to coin the term “HARKING,” the concept
had been proposed. Psychologist Adriaan de Groot (1956) was one of the
first to warn against the dangers of hindsight bias in in hypothesis testing.
See A. D. de Groot, “The meaning of ‘significance’ for different types of research
[translated and annotated by Eric-Jan Wagenmakers, Denny Borsboom,
Josine Verhagen, Rogier Kievit, Marjan Bakker, Angelique Cramer,
Dora Matzke, Don Mellenbergh, and Han L. J. van der Maas],” Acta Psychologica
148 (2014): 188–94, http://dx.doi.org/10.1016/j.actpsy.2014.02
.001.
33. Kerr, “HARKing,” http://dx.doi.org/10.1207/s15327957pspr0203_4.
34. Leslie K. John, George Loewenstein, and Drazen Prelec, “Measuring the
prevalence of questionable research practices with incentives for truth telling,”
Psychological Science (2012): 0956797611430953, http://dx.doi.org/10
.1177/0956797611430953.
35. D. J. Bem, “Writing the empirical journal article,” in Mark P. Zanna and John
M. Darley, eds., The Compleat Academic: A Practical Guide for the Beginning
Social Scientist (New York: Random House, 1987). An online version of this
article is available here: http://web.archive.org/web/20140701082131/http://
www.writingcenter.uconn.edu/pdf/Writing_the_Empirical_Journal_Article
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 2 | 223
_BEM.pdf. Similar sentiments were echoed in a 2012 online discussion by
Fritz Strack of the University of Würzburg: “Also, as I do still not believe in
ESP, I am not interested in the researchers’ beliefs and their a priori hypotheses.
I want to know what they learned and do not need to trace their many
errors along the way. I am aware that this may deviate from the basic tenets
of inferential statistics.” https://groups.google.com/d/msg/spsp
-discuss/JSUyofm1XDI/-uGnzzJTtHAJ.
Chapter 2. The Sin of Hidden Flexibility
1. For an excellent overview of NHST and common pitfalls, see the web resource
“Statistics done wrong” by Alex Reinhart: http://www.statistics
donewrong.com/index.html. See also Steven Goodman, “A dirty dozen:
Twelve p-value misconceptions,” Seminars in Hematology 45, no. 3 (2008):
135–40, http://dx.doi.org/10.1053/j.seminhematol.2008.04.003.
2. Ronald Fisher, “The arrangement of field experiments,” Journal of the Ministry
of Agriculture 33 (1926): 503–13. Available online at http://digital.library
.adelaide.edu.au/dspace/bitstream/2440/15191/1/48.pdf.
3. Valen E. Johnson, “Revised standards for statistical evidence,” Proceedings of
the National Academy of Sciences USA 110, no. 48 (2013): 19313–17, http://dx
.doi.org/10.1073/pnas.1313476110.
4. Joseph P. Simmons, Leif D. Nelson, and Uri Simonsohn, “False-positive psychology
undisclosed flexibility in data collection and analysis allows presenting
anything as significant,” Psychological Science (2011): 1359–66, http://dx
.doi.org/10.1177/0956797611417632.
5. Dorothy Bishop, “Interpreting unexpected significant results,” http://deevy
bee.blogspot.co.uk/2013/06/interpreting-unexpected-significant.html.
6. Will M. Gervais and Ara Norenzayan, “Analytic thinking promotes religious
disbelief,” Science 336, no. 6080 (2012): 493–96, http://dx.doi.org/10
.1126/science.1215647. A statistical critique can be found here: “Evidence of
publication bias in ‘Analytical thinking promotes religious disbelief,’” https://
web.archive.org/web/20130817140545/http://jt512.dyndns.org/blog/?p=130.
And see also “Do religious people lack analytical skills?” by Neuroautomaton,
https://web.archive.org/web/20130408053541/http://neuroautomaton
.com/religious-analytical-skills/.
7. E. J. Masicampo and Daniel R. Lalande, “A peculiar prevalence of p values
just below. 05,” Quarterly Journal of Experimental Psychology 65, no. 11 (2012):
2271–79, http://dx.doi.org/10.1080/17470218.2012.711335.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
224 | Notes to Chapter 2
8. Nathan C. Leggett, Nicole A. Thomas, Tobias Loetscher, and Michael E. R.
Nicholls, “The life of p: ‘Just significant’ results are on the rise,” Quarterly
Journal of Experimental Psychology 66, no. 12 (2013): 2303–9, http://dx.doi
.org/10.1080/17470218.2013.863371.
9. Daniël Lakens from the Eindhoven University of Technology has argued that
the spike in p values below .05, where it is shown to exist, could be due to
factors other than p-hacking—most notably publication bias. Lakens suggests
that rather than seeing a spike in p values immediately below .05 we
are actually seeing a dip in those immediately above .05. This could arise
owing to journals selectively rejecting studies that contain marginally nonsignificant
results (e.g., p = .06). For more details, see Daniël Lakens, “On
the challenges of drawing conclusions from p-values just below 0.05,” PeerJ 3
(2015): e1142, http://dx.doi.org/10.7717/peerj.1142.
10. Uri Simonsohn, Leif D. Nelson, and Joseph P. Simmons, “P-curve: A key to
the file-drawer,” Journal of Experimental Psychology: General 143, no. 2
(2014): 534, http://dx.doi.org/10.1037/a0033242. The full text is freely available
here: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2256237.
The authors also provide an online resource: http://p-curve.com/.
11. When discussed at a scientific conference in 2012, p-curve generated an
intense and heated debate, where (among other things) it was feared it
could be misused as a weapon against individual researchers. The argument
is summarized nicely in a blogpost by Sanjay Srivastava from the
University of Oregon: “Does your p-curve weigh as much as a duck?,”
http://hardsci.wordpress.com/2012/02/10/does-your-p-curve-weigh-as
-much-as-a-duck/.
12. Neuroskeptic, “False positive neuroscience,” http://blogs.discovermagazine
.com/neuroskeptic/2012/06/30/false-positive-neuroscience/.
13. For the study by Josh Carp, see Joshua Carp, “The secret lives of experiments:
Methods reporting in the fMRI literature,” Neuroimage 63, no. 1
(2012): 289–300, http://dx.doi.org/10.1016/j.neuroimage.2012.07.004.
For discussion of likely false positive rates in fMRI research, see Tor D.
Wager, Martin A. Lindquist, Thomas E. Nichols, Hedy Kober, and Jared X.
Van Snellenberg, “Evaluating the consistency and specificity of neuroimaging
data using meta-analysis,” Neuroimage 45, no. 1 (2009): S210–S221,
http://dx.doi.org/10.1016/j.neuroimage.2008.10.061.
For discussion of fMRI reliability, see Craig M. Bennett and Michael B.
Miller, “How reliable are the results from functional magnetic resonance
imaging?,” Annals of the New York Academy of Sciences 1191, no. 1 (2010):
133–55, http://dx.doi.org/10.1111/j.1749–6632.2010.05446.x, together with
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 2 | 225
a blog summary by Vaughan Bell: “How reliable are fMRI results?,” at http://
mindhacks.com/2010/03/04/how-reliable-are-fmri-results/.
In 2016, Russ Poldrack and colleagues published an authoritative overview
of the threat posed to fMRI by low reproducibility and transparency.
See Russell Poldrack, Chris I. Baker, Joke Durnez, Krzysztof Gorgolewski,
Paul M. Matthews, Marcus Munafò, Thomas Nichols, Jean-Baptiste Poline,
Edward Vul, and Tal Yarkoni, “Scanning the Horizon: Future challenges for
neuroimaging research,” bioRxiv (2016): 059188, http://biorxiv.org/content
/early/2016/08/01/059188.
14. Uri Simonsohn vs. Norbert Schwarz, https://web.archive.org/web/201605
16032001/http://opim.wharton.upenn.edu/~uws/SPSP/post.pdf; and see
https://groups.google.com/forum/#!searchin/spsp-discuss/simonsohn/spsp
-discuss/izOh5lrQDgQ/zLud5bZvPPUJ for a heated debate.
15. See “Replication, period. (A guest post by David Funder)” by David Funder,
http://hardsci.wordpress.com/2012/09/21/replication-period-a-guest-post-by
-david-funder/, and “Friday Fun: One researcher’s p-curve analysis” by Michael
Kraus, http://psych-your-mind.blogspot.co.uk/2012/02/friday-fun-one
-researchers-p-curve.html.
16. As Sanjay Srivastava from the University of Oregon notes, failure to replicate
a previous finding needn’t indicate that the nonreplication was itself statistically
different from the original finding. Showing this requires a direct test,
which was satisfied in both nonreplications of the original Bargh study.
http://hardsci.wordpress.com/2012/03/12/some-reflections-on-the-bargh
-doyen-elderly-walking-priming-brouhaha/.
17. “Priming effects replicate just fine, thanks” by John Bargh: http://www
.psychologytoday.com/blog/the-natural-unconscious/201205/priming-effects
-replicate-just-fine-thanks.
18. The “one-time” prevalence estimate refers to the percentage of psychologists
who have engaged in a particular practice on at least one occasion. It doesn’t
estimate wider prevalence.
19. Andrew Gelman and Eric Loken, “The garden of forking paths: Why multiple
comparisons can be a problem, even when there is no ‘fishing expedition’
or ‘p-hacking’ and the research hypothesis was posited ahead of time,” Department
of Statistics, Columbia University (2013), http://www.stat.columbia
.edu/~gelman/research/unpublished/p_hacking.pdf.
20. Mark Stokes, “Biased debugging,” http://the-brain-box.blogspot.co
.uk/2013/02/biased-debugging.html. See also an interesting example of how
such an error was caught by cognitive neuroscientist Russ Poldrack: “Anat-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
226 | Notes to Chapter 2
omy of a coding error,” http://www.russpoldrack.org/2013/02/anatomy-of
-coding-error.html.
21. I am grateful to Eoin Travers (http://www.eointravers.com/) for making me
aware of this term from studies of reasoning. For more information, see
https://books.google.co.uk/books?id=iFMhZ4dl1KcC&lpg=PR9&ots
=ZM1C6gn8p2&dq=evans%20newstead%20byrne&lr&pg=PA247#v
=onepage&q=selective%20scrutiny%20account&f=false.
22. “Are research psychologists more like detectives or lawyers?” by John A.
Johnson, http://www.psychologytoday.com/blog/cui-bono/201307/are
-research-psychologists-more-detectives-or-lawyers-0.
23. Joseph P. Simmons, Leif D. Nelson, and Uri Simonsohn, “A 21 word solution,”
available at SSRN 2160588 (2012), http://papers.ssrn.com/sol3/papers
.cfm?abstract_id=2160588.
24. See Michael J. Strube, “SNOOP: A program for demonstrating the consequences
of premature and repeated null hypothesis testing,” Behavior Research
Methods 38, no. 1 (2006): 24–27, http://dx.doi.org/10.3758/BF031
92746. Software is available here: http://web.archive.org/web/2016042222
0546/http://www.artsci.wustl.edu/~socpsy/Snoop.7z.
For a similar proposal, see, Daniël Lakens, “Performing high-powered
studies efficiently with sequential analyses,” European Journal of Social Psychology
44, no. 7 (2014): 701–10, http://dx.doi.org/10.1002/ejsp.2023. The
article can be downloaded freely from http://papers.ssrn.com/sol3/papers
.cfm?abstract_id=2333729.
25. For a lucid introduction to Bayesian hypothesis testing, see Zoltan Dienes,
“Bayesian versus orthodox statistics: Which side are you on?,” Perspectives on
Psychological Science 6, no. 3 (2011): 274–90, http://dx.doi.org/10.1177/17456
91611406920. This paper can be freely downloaded from: http://www.lifesci
.sussex.ac.uk/home/Zoltan_Dienes/Dienes%202011%20Bayes.pdf.
26. Eric-Jan Wagenmakers, Michael Lee, Tom Lodewyckx, and Geoffrey J. Iverson,
“Bayesian versus frequentist inference,” in Bayesian Evaluation of Informative
Hypotheses, 81–207 (New York: Springer, 2008), http://link.springer
.com/chapter/10.1007/978–0–387–09612–4_9#page-1.
27. A. W. Edwards, Likelihood: Expanded Edition (Baltimore: Johns Hopkins
University Press, 1992), 30. For a foundational and technically detailed text
on the likelihood principle, the reader is directed also to: James O. Berger,
Robert L. Wolpert, M. J. Bayarri, M. H. DeGroot, Bruce M. Hill, David A.
Lane, and Lucien LeCam, “The likelihood principle,” Lecture Notes—Monograph
Series 6 (1988), http://web.uvic.ca/~dgiles/blog/Berger_and_Wolpert
.pdf.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 3 | 227
28. Dave Nussbaum, “The Stapel continuum,” http://web.archive.org/web
/20140604081238/http://www.davenussbaum.com/the-stapel-continuum/.
Chapter 3. The Sin of Unreliability
1. Geoff Brumfiel, “Particles break light speed limit,” Nature News, 23 September
2011, http://www.nature.com/news/2011/110922/full/news.2011.554
.html.
2. Robert Evans, “‘Faster than light’ particles threaten Einstein,” Reuters, 23
September 2011, http://web.archive.org/web/20150321055654/http://www
.reuters.com/article/2011/09/23/us-science-light-idUSTRE78L4FH20110923.
3. Michael D. Lemonick, “Was Einstein wrong? A faster-than-light neutrino
could be saying yes,” Time, 23 September 2011, http://content.time
.com/time/health/article/0,8599,2094665,00.html.
4. Parts of this section have been adapted from: Chris Chambers, “Physics
envy: Do ‘hard’ sciences hold the solution to the replication crisis in psychology?,”
Guardian, 10 June 2014, http://www.theguardian.com/science/head
-quarters/2014/jun/10/physics-envy-do-hard-sciences-hold-the-solution-to
-the-replication-crisis-in-psychology.
5. For the full text of Feynman’s lecture, see http://web.archive.org/web/2016
0412070659/http://www.lhup.edu/~DSIMANEK/cargocul.htm.
6. Matthew C. Makel, Jonathan A. Plucker, and Boyd Hegarty, “Replications in
psychology research: How often do they really occur?,” Perspectives on Psychological
Science 7, no. 6 (2012): 537–42, http://dx.doi.org/10.1177/1745
691612460688.
7. Ioannidis reaches this estimation by taking into account the prior probability
of a studied effect in psychology being nonnull (i.e, H0 is false), low rates
of replication as a measure for ensuring self-correction, and the estimated
rate of false positive discoveries. See John P. A. Ioannidis, “Why science is
not necessarily self-correcting,” Perspectives on Psychological Science 7, no. 6
(2012): 645–54, http://dx.doi.org/10.1177/1745691612464056.
8. Christopher J. Ferguson and Moritz Heene, “A vast graveyard of undead theories:
Publication bias and psychological science’s aversion to the null,” Perspectives
on Psychological Science 7, no. 6 (2012): 555–61, http://dx.doi.org/10
.1177/1745691612459059.
9. The complete special issue of replication studies can be freely downloaded
from http://econtent.hogrefe.com/toc/zsp/45/3.
10. Simone Schnall’s blog post can be read at http://web.archive
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
228 | Notes to Chapter 3
.org/web/20140531143904/http://www.spspblog.org/simone-schnall-on-her
-experience-with-a-registered-replication-project/.
11. Dan Gilbert’s extraordinary comment can be read at http://web.archive
.org/web/20140531143904/http://www.spspblog.org/simone-schnall-on-her
-experience-with-a-registered-replication-project/#comment-17137.
12. See http://web.archive.org/web/20141214024408/http://www.wjh.harvard
.edu/~dtg/Bullies.pdf.
13. The language adopted by Gilbert and Schnall is reminiscent of immunologist
Jacques Benveniste’s reaction, in 1988, to the failed attempts to replicate
his experiments on so-called water memory as the basis of homeopathy.
Benveniste raised no objections to the replication methods before seeing the
results. However, upon discovering that the data provided no evidence to
support his “water memory” hypothesis, he condemned the entire process,
writing in Nature that “Salem witchhunts or McCarthy-like prosecutions will
kill science. Science flourishes only in freedom. We must not let, at any
price, fear, blackmail, anonymous accusation, libel and deceit nest in our
labs.” http://web.archive.org/web/20070225102815/http://br.geocities
.com/criticandokardec/benveniste02.pdf.
14. For the full text of Kahneman’s letter, see http://www.scribd.com/doc
/225285909/Kahneman-Commentary. In contrast to Kahneman’s concerns
about the reputational impact of nonreplications, a 2015 study by Adam Fetterman
and Kai Sassenberg found that scientists overestimate the negative
reputational impact of nonreplications. Moreover, they found that “admitting
wrongness about a nonreplicated finding is less harmful to one’s reputation
than not admitting.” Adam K. Fetterman and Kai Sassenberg, “The reputational
consequences of failed replications and wrongness admission
among scientists,” PLOS ONE 10, no. 12 (2015): e0143723, http://dx.doi.org
/10.1371/journal.pone.0143723.
15. Andrew Wilson, “Psychology’s real replication problem: Our Methods sections,”
http://psychsciencenotes.blogspot.co.uk/2014/05/psychologys-real
-replication-problem.html.
16. Jon Butterworth. Personal interview via e-mail. 2 June 2014.
17. Helen Czerski, personal interview via e-mail, 1 June 2014.
18. Katherine Mack. Personal interview via e-mail, 3 June 2014.
19. See http://journals.plos.org/plosone/s/criteria-for-publication#loc-2.
20. Jason Mitchell, “On the emptiness of failed replications,” http://web.archive
.org/web/20150604192510/http://wjh.harvard.edu/~jmitchel/writing
/failed_science.htm.
21. For a range of critical responses to Jason’s Mitchell essay, see Chris Said,
“Jason Mitchell’s essay,” http://web.archive.org/web/20140916000231/http://
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 3 | 229
filedrawer.wordpress.com/2014/07/07/jason-mitchells-essay/; Sean Mackinnon,
“Response to Jason Mitchell’s ‘On the emptiness of failed replications,’”
http://web.archive.org/web/20141013180548/http://osc.centerforopen
science.org/2014/07/09/response-to-jason-mitchell/; Neuroskeptic, “On ‘On
the emptiness of failed replications,’” http://blogs.discovermagazine.com
/neuroskeptic/2014/07/07/emptiness-failed-replications/; Bob Lewis, “In defense
of replication studies,” http://web.archive.org/web/20150904102838
/http://trustinbob.blogspot.co.uk/2014/07/in-defense-of-replication-studies
.html; Dorothy Bishop, “Replication and reputation: Whose career matters?,”
http://deevybee.blogspot.co.uk/2014/08/replication-and-reputation
-whose-career.html.
22. Wolfgang Stroebe and Fritz Strack, “The alleged crisis and the illusion of
exact replication,” Perspectives on Psychological Science 9, no. 1 (2014): 59–71,
http://dx.doi.org/10.1177/1745691613514450.
23. Daniel J. Simons, “The value of direct replication,” Perspectives on Psychological
Science 9, no. 1 (2014): 76–80, http://dx.doi.org/10
.1177/1745691613514755. See also Stefan Schmidt, “Shall we really do it
again? The powerful concept of replication is neglected in the social sciences,”
Review of General Psychology 13, no. 2 (2009): 90, http://dx.doi.org/10
.1037/a0015108.
24. Jacob Cohen, “The statistical power of abnormal-social psychological research:
A review,” Journal of Abnormal and Social Psychology 65, no. 3 (1962):
145, http://dx.doi.org/10.1037/h0045186.
25. Peter Sedlmeier and Gerd Gigerenzer, “Do studies of statistical power have
an effect on the power of studies?,” Psychological Bulletin 105, no. 2 (1989):
309, http://dx.doi.org/10.1037/0033–2909.105.2.309.
26. Scott Bezeau and Roger Graves, “Statistical power and effect sizes of clinical
neuropsychology research,” Journal of Clinical and Experimental Neuropsychology
23, no. 3 (2001): 399–406, http://dx.doi.org/10.1076/jcen.23.3.399.1181.
27. Katherine S. Button, John P. A. Ioannidis, Claire Mokrysz, Brian A. Nosek,
Jonathan Flint, Emma S. J. Robinson, and Marcus R. Munafò, “Power failure:
Why small sample size undermines the reliability of neuroscience,” Nature
Reviews Neuroscience 14, no. 5 (2013): 365–76, http://dx.doi.org/10
.1038/nrn3475.
28. Klaus Fiedler, Florian Kutzner, and Joachim I. Krueger, “The long way
from α-error control to validity proper problems with a short-sighted falsepositive
debate,” Perspectives on Psychological Science 7, no. 6 (2012): 661–69,
http://dx.doi.org/10.1177/1745691612462587.
29. In brief, this is because when p ≅.05, the left half of the normal distribution
(a symmetrical bell curve) that characterizes the observed effect will fall
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
230 | Notes to Chapter 3
within 1.96 standard deviations of the null distribution. Upon exact replication
with the same sample size, there is therefore a roughly 50 percent
chance of returning a nonsignificant test statistic (p>.05). Increasing the
power requires doubling or tripling the original sample size. For a full explanation,
see figure 1 in Button et al., “Power failure.”
30. Free software packages such as G* Power now make prospective power analysis
easy. http://www.gpower.hhu.de/en.html.
31. Scott E. Maxwell, “The persistence of underpowered studies in psychological
research: Causes, consequences, and remedies,” Psychological Methods 9,
no. 2 (2004): 147, http://dx.doi.org/10.1037/1082–989X.9.2.147. An additional
source of complexity in power analyses is deciding what effect size is
appropriate. Because of publication bias, published effect sizes overestimate
true effects sizes; therefore researchers should always power their studies
according to the lower bound estimate of the likely effect size. This computation
can be challenging and complex. For discussion, see Scott E. Maxwell,
Michael Y. Lau, and George S. Howard, “Is psychology suffering from a replication
crisis? What does ‘failure to replicate’ really mean?,” American Psychologist
70, no. 6 (2015): 487, http://dx.doi.org/10.1037/a0039400.
32. Rachel A. Smith, Timothy R. Levine, Kenneth A. Lachlan, and Thomas A.
Fediuk, “The high cost of complexity in experimental design and data analysis:
Type I and type II error rates in multiway ANOVA,” Human Communication
Research 28, no. 4 (2002): 515–30, http://dx.doi.org/10.1111/j.1468
–2958.2002.tb00821.x.
33. Etienne P. LeBel, Denny Borsboom, Roger Giner-Sorolla, Fred Hasselman,
Kurt R. Peters, Kate A. Ratliff, and Colin Tucker Smith, “Psychdisclosure.org
grassroots support for reforming reporting standards in psychology,” Perspectives
on Psychological Science 8, no. 4 (2013): 424–32, http://dx.doi.org/10
.1177/1745691613491437.
34. This raises the question of why nearly half of published studies included
measures that were apparently theoretically irrelevant. Given the high prevalence
in psychology of post hoc hypothesizing (a.k.a. HARKing), it is perhaps
more reasonable to assume that these measures were at least vaguely
important at the outset of many studies but that the researchers decided
after inspecting the results that they were no longer relevant to (or in agreement
with) the HARKed hypothesis. As we saw in chapters 1 and 2, this
form of cherry-picking violates the hypothetico-deductive model of the scientific
method.
35. Joseph P. Simmons, Leif D. Nelson, and Uri Simonsohn, “A 21 word solution,”
available at SSRN 2160588 (2012), http://papers.ssrn.com/sol3/papers
.cfm?abstract_id=2160588.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 3 | 231
36. For two recent surveys showing how psychologists frequently misunderstand
p values, see: Laura Badenes-Ribera, Dolores Frías-Navarro, Héctor
Monterde-i-Bort, and Marcos Pascual-Soler, “Interpretation of the p value:
A national survey study in academic psychologists from Spain,” Psicothema
27, no. 3 (2015): 290–95, http://dx.doi.org/10.7334/psicothema2014.283;
Laura Badenes-Ribera, Dolores Frias-Navarro, Bryan Iotti, Amparo Bonilla
Campos, and Claudio Longobardi, “Misconceptions of the p-value among
Chilean and Italian academic psychologists,” Frontiers in Psychology 7 (2016):
1247, http://dx.doi.org/10.3389/fpsyg.2016.01247.
37. For an excellent and thorough list of misconceptions about p values, the
reader is directed to see also Steven Goodman, “A dirty dozen: Twelve
p-value misconceptions,” Seminars in Hematology 45, no. 3 (2008): 135–40,
http://dx.doi.org/10.1053/j.seminhematol.2008.04.003.
38. Sander Nieuwenhuis, Birte U. Forstmann, and Eric-Jan Wagenmakers, “Erroneous
analyses of interactions in neuroscience: A problem of significance,”
Nature Neuroscience 14, no. 9 (2011): 1105–7, http://dx.doi.org/10.1038/nn
.2886; Andrew Gelman and Hal Stern, “The difference between ‘significant’
and ‘not significant’ is not itself statistically significant,” American Statistician
60, no. 4 (2006): 328–31, http://dx.doi.org/10.1198/000313006X152649.
This article can be downloaded freely from http://www.stat.columbia.edu
/~gelman/research/published/signif4.pdf.
39. For more information on the conditions that should lead to retraction of articles,
see the Retraction Guidelines issued by the Committee on Publication
Ethics: http://publicationethics.org/files/retraction%20guidelines.pdf.
40. Retraction due to failure to replicate is standard practice in sciences such as
physics. Adam Marcus, “Doing the right thing: Authors retract lubricant
paper whose findings they can’t reproduce,” http://retractionwatch.com
/2014/03/14/doing-the-right-thing-authors-retract-lubricant-paper-whose
-findings-they-cant-reproduce/. Similar examples can be found in neuroscience
and biology. Ed Yong, “Narcolepsy paper retracted,” http://phenomena
.nationalgeographic.com/2014/07/30/narcolepsy-paper-retracted/.
41. Minhua Zhang and Michael L. Grieneisen, “The impact of misconduct on
the published medical and non-medical literature, and the news media,” Scientometrics
96, no. 2 (2013): 573–87, http://dx.doi.org/10.1007/s11192-012
-0920-5.
42. Michael L. Grieneisen and Minghua Zhang, “A comprehensive survey of retracted
articles from the scholarly literature,” PLoS One 7, no. 10 (2012):
e44118, http://dx.doi.org/10.1371/journal.pone.0044118. The quoted figure
of 27 percent can be calculated from the data presented in figure S1 of this
article by comparing the overall rate of retractions calculated across the sum
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
232 | Notes to Chapter 3
of all psychology disciplines (32 retractions per 50,974 articles = 0.063 percent)
with the overall rate of retractions calculated across the sum of the remaining
233 disciplines (5,804 retractions per 2,501,816 articles = 0.23 percent).
The rate of retractions in psychology (0.063 percent) is therefore 27
percent of the rate of retractions of the remaining fields (0.23 percent).
43. Dan Simons, “Replication, retraction, and responsibility,” http://blog.dan
simons.com/2014/01/replication-retraction-and.html.
44. John A. Bargh and Idit Shalev, “The substitutability of physical and social
warmth in daily life,” Emotion 12, no. 1 (2012): 154–62, http://www.ncbi
.nlm.nih.gov/pmc/articles/PMC3406601/.
45. For a blog summary of the study, see http://traitstate.wordpress.com/2014
/01/24/warm-water-and-loneliness/. The peer-reviewed replication is M. B.
Donnellan, R. E. Lucas, and J. Cesario, “On the association between loneliness
and bathing habits: Nine replications of Bargh and Shalev (2012) Study
1,” Emotion 15, no. 1 (2015): 109, http://dx.doi.org/10.1037/a0036079. The
article can be downloaded freely from https://msu.edu/~cesario
/publications/Donnellan%20Lucas%20Cesario%20IN%20PRESS%20
EMOTION%20loneliness%20bathing.pdf.
46. Will Gervais, “More power!,” http://willgervais.com/blog/2014/3/5/more
-power.
47. Eric-Jan Wagenmakers and B. U. Forstmann, “Rewarding high-power replication
research,” Cortex 51, no. 10 (2014), http://dx.doi.org/10.1016/j.cortex
.2013.09.010.
48. Sanjay Srivastava, “A Pottery Barn rule for scientific journals,” http://hardsci
.wordpress.com/2012/09/27/a-pottery-barn-rule-for-scientific-journals/. We
are planning to implement Sanjay’s proposal for the first time in 2017 at the
journal Royal Society Open Science. For details see: http://neurochambers
.blogspot.co.uk/2016/11/an-accountable-replication-policy-at.html
49. Kimmo Eriksson and Brent Simpson, “Editorial decisions may perpetuate
belief in invalid research findings,” PLOS ONE 8, no. 9 (2013): e73364,
http://dx.doi.org/10.1371/journal.pone.0073364.
50. Keith R. Laws, “Negativland—a home for all findings in psychology,” BMC
Psychology 1, no. 1 (2013): 2, http://dx.doi.org/10.1186/2050–7283–1–2.
51. This example is based on a real case of a reported test for Alzheimer’s disease,
critiqued by David Colquhoun: http://www.dcscience.net/?p=6473.
See also this animation at the Daily Mirror: http://web.archive.org/web
/20160206134529/http://ampp3d.mirror.co.uk/2014/03/11/how-a-90
-accurate-alzheimers-test-can-be-wrong-92-of-the-time/. And for a technical
overview, see http://www.biomedcentral.com/1741–7015/9/20.
52. Psychologist Gerd Gigerenzer has shown that the base rate fallacy can be
largely overcome if problems are presented in terms of natural frequencies
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 3 | 233
(“1 in 100”) rather than probabilities (“1 percent”). Using a similar example
to the one described in this chapter, but doing so in terms of natural frequencies,
he found that correct answers among medical staff and students
was as high as 76 percent. Rephrasing the current problem in terms of natural
frequencies, as proposed by Gigerenzer, it would read like this:
10 out of 1000 people has Alzheimer’s disease. A test has been developed
to detect when a person has Alzheimer’s disease. For 8 of the 10 people
who have the disease, the test will come out positive. But sometimes the
test also comes out positive when it is given to a person who is completely
healthy. Specifically, out of every 1000 people who are perfectly
healthy, 50 of them test positive for Alzheimer’s disease. Imagine that we
have assembled a random sample of 1000 people. They were selected by
a lottery. Those who conducted the lottery had no information about the
health status of any of these people. How many people who test positive
for the disease will actually have the disease? ___ out of ___.
See Gerd Gigerenzer, “How to make cognitive illusions disappear: Beyond
‘heuristics and biases,’” European Review of Social Psychology 2, no. 1 (1991):
83–115, http://dx.doi.org/10.1080/14792779143000033. The article can be
downloaded freely at http://library.mpib-berlin.mpg.de/ft/gg/GG_How
_1991.pdf.
53. In limited situations Bayes factors can even be multiplied together—for instance,
suppose in two parallel experiments, values of B = 1.6 and B = 2.5
are uncovered in favor of H1 over H0. While neither value is sufficient to
generate substantial evidence (>3) for H1, collectively they lead to B = 4
(1.6 × 2.5). It should be noted that the multiplication of Bayes factors is
possible only when H1 is represented as a point hypothesis (i.e., a specific
value) that is unchanged by the addition of new data. This might also apply
in cases where two parallel (but independent) studies are run with identical
methods and thus identical priors. In most cases, however, where a replication
is conducted after an initial study, additional data will change the
precise parameters of H1; in this case, the second analysis would adjust H1
based on all available data, preventing simple multiplication of the priors.
54. For an accessible introduction to Bayesian hypothesis testing in psychology,
including worked-through examples, the reader is directed to two articles by
Zoltan Dienes from the University of Sussex: Zoltan Dienes, “Bayesian versus
orthodox statistics: Which side are you on?,” Perspectives on Psychological
Science 6, no. 3 (2011): 274–90, http://dx.doi.org/10.1177/1745691611406920,
which can be freely downloaded from http://www.lifesci.sussex.ac.uk/home
/Zoltan_Dienes/Dienes%202011%20Bayes.pdf; and Zoltan Dienes, “Using
Bayes to get the most out of non-significant results,” Frontiers in Psychology 5,
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
234 | Notes to Chapter 4
no. 781 (2014): 1–17, http://dx.doi.org/10.3389/fpsyg.2014.00781. Other excellent
resources include Alexander Etz’s series of free and highly accessible
blogposts, “Understanding Bayes”: http://alexanderetz.com/understanding
-bayes/; and Eric-Jan Wagenmakers, Richard D. Morey, and Michael D. Lee,
“Bayesian benefits for the pragmatic researcher,” https://osf.io/3tdh9/.
55. For an excellent example of adversarial collaboration, the reader is directed
to a recent study of eye movements and memory recall: Dora Matzke,
Sander Nieuwenhuis, Hedderik van Rijn, Heleen A. Slagter, Maurits W. van
der Molen, and Eric-Jan Wagenmakers, “The effect of horizontal eye movements
on free recall: A preregistered adversarial collaboration,” Journal of Experimental
Psychology: General 144, no. 1 (2015): e1, http://dx.doi.org/10
.1037/xge0000038.
56. Nature Neuroscience Editorial, “Raising standards,” http://www.nature.com
/neuro/journal/v16/n5/full/nn.3391.html.
57. For details on the initiative at Psychological Science, see https://www
.psychologicalscience.org/index.php/publications/journals/psychological
_science/ps-submissions#. See also http://www.theguardian.com/science
/head-quarters/2014/jul/09/case-report-forms-psychology-replication for a
call by psychologist Pete Etchells for psychologists to adopt more detailed
approach to reporting methodology called “case report forms.”
58. For a summary of the Center for Open Science badges initiative, see
https://osf.io/tvyxz/wiki/home/.
59. Adam Marcus and Ivan Oransky, “Time for a reproducibility index,”
http://www.labtimes.org/labtimes/ranking/dont/2013_04.lasso.
Chapter 4. The Sin of Data Hoarding
1. For the NIH data-sharing policy, see http://www.nlm.nih.gov/NIHbmic/nih
_data_sharing_policies.html.
2. For Science magazine’s requirements on data availability, see http://www
.sciencemag.org/authors/science-editorial-policies#data-deposition.
3. For Earl Miller’s tweet, see https://twitter.com/MillerLabMIT/status
/360368532774592512.
4. Uri Simonsohn, “Just post it: The lesson from two cases of fabricated data
detected by statistics alone,” Psychological Science 24, no. 10 (2013): 1875–
88, http://www.dx.doi.org/10.1177/0956797613480366.
5. http://www.nature.com/news/scientists-losing-data-at-a-rapid-rate-1
.14416#/b1.
6. Jelte M. Wicherts and Marjan Bakker, “Publish (your data) or (let the data)
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 4 | 235
perish! Why not publish your data too?,” Intelligence 40, no. 2 (2012): 73–76,
http://dx.doi.org/10.1016/j.intell.2012.01.004.
7. A fine example is the work of Mark Stokes at the University of Oxford, who
in 2013 reanalyzed a large archive of animal neurophysiology data to make a
key discovery about working memory: Mark G. Stokes, Makoto Kusunoki,
Natasha Sigala, Hamed Nili, David Gaffan, and John Duncan, “Dynamic
coding for cognitive control in prefrontal cortex,” Neuron 78, no. 2 (2013):
364–75, http://dx.doi.org/10.1016/j.neuron.2013.01.039.
Another nice example is a 2012 study by Gilles Dutilh and colleagues.
They took an existing data set that had been created to validate a new database
of Dutch word frequencies and used it to test theories of why people
are slower to make responses after committing errors in a decision-making
task: Gilles Dutilh, Joachim Vandekerckhove, Birte U. Forstmann, Emmanuel
Keuleers, Marc Brysbaert, and Eric-Jan Wagenmakers, “Testing theories
of post-error slowing,” Attention, Perception, and Psychophysics 74, no. 2
(2012): 454–65, http://dx.doi.org/10.3758/s13414–011–0243–2.
8. Heather A. Piwowar and Todd J. Vision, “Data reuse and the open data citation
advantage,” PeerJ 1 (2013): e175, http://dx.doi.org/10.7717/peerj.175.
9. Heather A. Piwowar, Roger S. Day, and Douglas B. Fridsma, “Sharing detailed
research data is associated with increased citation rate,” PLOS ONE 2,
no. 3 (2007): e308, http://dx.doi.org/10.1371/journal.pone.0000308.
10. Jelte M. Wicherts, Denny Borsboom, Judith Kats, and Dylan Molenaar, “The
poor availability of psychological research data for reanalysis,” American Psychologist
61, no. 7 (2006): 726, http://dx.doi.org/10.1037/0003–066X.61.7
.726. The article can be freely downloaded from https://web.archive.org
/web/20150420101309/http://wicherts.socsci.uva.nl/datasharing.pdf
11. http://memforms.apa.org/apa/cli/interest/ethics1.cfm#8_14.
12. Jelte Wicherts, personal interview via e-mail, 28 August 2014.
13. Jelte M. Wicherts, Marjan Bakker, and Dylan Molenaar, “Willingness to
share research data is related to the strength of the evidence and the quality
of reporting of statistical results,” PLOS ONE 6, no. 11 (2011): e26828, http://
dx.doi.org/10.1371/journal.pone.0026828.
14. Jelte Wicherts, personal interview via e-mail, 28 August 2014.
15. John A. Bargh and Idit Shalev, “The substitutability of physical and social
warmth in daily life,” Emotion 12, no. 1 (2012): 154–62, http://www.ncbi
.nlm.nih.gov/pmc/articles/PMC3406601/.
16. Dan Simons, “Replication, retraction, and responsibility,” http://blog.dan
simons.com/2014/01/replication-retraction-and.html.
17. American Psychological Association, 2010, Ethical principles of psychologists
and code of conduct, section 8.14, http://www.apa.org/ethics/code
/principles.pdf.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
236 | Notes to Chapter 4
18. Brent Donnellan, “What’s the first rule about John Bargh’s data?,” http://
traitstate.wordpress.com/2012/09/20/whats-the-first-rule-about-john-barghs
-data/.
19. Dan Simons, “The fog of data—secrecy and science,” http://blog.dansimons
.com/2012/09/the-fog-of-data-secrecy-and-science.html.
20. Ed Yong, “The data detective,” Nature News, 2 July 2012, http://www.nature
.com/news/the-data-detective-1.10937; and see also Ed Yong, “Uncertainty
shrouds psychologist’s resignation,” Nature News, 12 July 2012, http://www
.nature.com/news/uncertainty-shrouds-psychologist-s-resignation-1.10968.
21. J.B.S. Haldane, “The faking of genetical results,” Eureka 6 (1941), http://www
.archim.org.uk/eureka/27/faking.html.
22. For the PLOS data availability policy, see http://journals.plos.org/plosone/s
/data-availability.
23. DrugMonkey, “PLOS is letting the inmates run the asylum and this will
kill them,” https://drugmonkey.wordpress.com/2014/02/25/plos-is-letting
-the-inmates-run-the-asylum-and-this-will-kill-them/.
24. rxnm, “PLOS’s open data fever dream,” https://rxnm.wordpress.com/2014/02
/25/fan-fiction/
25. Erin C. McKiernan, “My concerns about PLOS’s new open data policy,”
https://emckiernan.wordpress.com/2014/02/26/my-concerns-about-ploss
-new-open-data-policy/.
26. Matthew D. MacManes, “Corner cases,” http://genomebio.org/corner
-cases/.
27. Some researchers question whether even standard anonymization is sufficient,
as multiple anonymized data sets could, collectively, contain enough
information to unblind participants.
28. At the time of writing, Figshare imposes a limit of 5GB per file, while Zenodo
and Dataverse allow up to 2GB (though may permit larger file sizes on
a case-by-case basis). Figshare offers individual user accounts for free but
charges for institutional accounts.
29. See https://dx.doi.org/10.6084/m9.figshare.3381565.v1 for the analysis of the
data.
30. Damian Pattinson, e-mail, 1 December 2015.
31. For guidelines at the journal Cognition, see http://www.elsevier
.com/journals/cognition/0010–0277/guide-for-authors. For Experimental Psychology,
see https://us.hogrefe.com/products/journals/exppsy. For Archives of
Scientific Psychology, see http://www.apa.org/pubs/journals/arc/.
32. Researchers are divided on the issue of whether providing data, alone, is a
significant enough contribution to justify coauthoship. In May 2016 I posed
this question to Twitter, https://twitter.com/chrisdc77/status/7287096084
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 5 | 237
71179268. Overall, most respondents felt that sharing data did not justify authorship.
Curiously, however, the community was more amenable to coauthorship
in exchange for data when the data was shared privately between
peers rather than publicly archived. Not sharing data was thus regarded as
the more rewarded practice, suggesting that the lack of data archiving in
psychology could be a form of rent seeking (“I’ll give you access to my data if
you make me an author on your paper”).
33. For details on the badges initiative, see http://www.psychologicalscience.org
/index.php/publications/journals/psychological_science/badges and https://
osf.io/tvyxz/wiki/home/.
34. For further information, see http://centerforopenscience.org/top/.
35. See https://centerforopenscience.org/top/#list.
36. See http://www.esrc.ac.uk/files/about-us/policies-and-standards/esrc
-research-data-policy/. One shortcoming of this policy is that data archiving
is required only at the end of the project; therefore in most cases there is no
link to the data in published peer-reviewed articles.
37. For more information, see https://opennessinitiative.org/.
38. Frederick Verbruggen, personal interview via e-mail, 22 September 2014.
39. D. V. M. Bishop, “Open research practices: Unintended consequences and
suggestions for averting them (commentary on the peer reviewers’ Openness
Initiative),” Royal Society Open Science 3, no. 4 (2016): 160109, http://dx
.doi.org/10.1098/rsos.160109. See also a provocative comment by Bishop
and Stephan Lewandowsky, http://www.nature.com/news/research-integrity
-don-t-let-transparency-damage-science-1.19219.
40. https://twitter.com/ImAlsoGreg/status/438179256284479488.
41. For a full outline of Candice and Richard’s proposed “data partner scheme,”
see their blogpost “Habits and open data: Helping students develop a theory
of scientific mind,” http://bayesfactor.blogspot.no/2015/11/habits-and-open
-data-helping-students.html.
42. Stephen Curry, personal interview via e-mail, 3 June 2014.
Chapter 5. The Sin of Corruptibility
1. In 2012, Stapel published a fascinating confessional of his rise and fall, entitled
Ontsporing (which translates from Dutch as “Derailed” or “Derailment”).
In 2014, psychologist Nick Brown translated the entire book into
English with the new title Faking Science: A True Story of Academic Fraud. It
can freely downloaded from http://web.archive
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
238 | Notes to Chapter 5
.org/web/20150325224546/https://errorstatistics.files.wordpress.com/2014
/12/fakingscience-20141214.pdf.
2. There is a question mark over the exact point in time that Stapel began committing
fraud. The Levelt Committee, which coordinated the formal investigation
of his fraud, examined three periods in his career: University of Amsterdam,
1993–99 (including his PhD studies), the University of Groningen,
2000–2006 (where he became full professor); and Tilburg University,
2006–11 (where he established a research center for behavioral economics
and became dean of research). Stapel claims that he used questionable practices
from early on, but committed outright fabrication only from the Groningen
period onward; however, the Drenth subcommittee, which was
charged with investigating his earlier time in Amsterdam, found evidence of
fraudulent practices as early as 1996. The veracity of Stapel’s confession
should therefore be viewed with a grain of salt. What seems clear is that Stapel
escalated the frequency and severity of his data manipulation and fabrication
as he progressed.
3. Stapel, Faking Science, trans. Brown, 100–101.
4. Ibid., 107.
5. Christopher J. Ferguson and Moritz Heene, “A vast graveyard of undead theories:
Publication bias and psychological science’s aversion to the null,” Perspectives
on Psychological Science 7, no. 6 (2012): 555–61, http://dx.doi.org/10
.1177/1745691612459059.
6. Stapel, Faking Science, trans. Brown, 109.
7. Ibid., 103.
8. Ibid., 101.
9. Ibid., 102.
10. Ibid., 103.
11. Ibid., 130.
12. Ibid., 128. There are two possible explanations for these illusory replications.
Stapel may well have guessed correctly on some occasions because his
hypotheses were often generated from a careful reading of the literature and
so may have been true. Less optimistically, as we saw in chapter 1, psychology
is plagued by confirmation bias in which researchers too easily see what
they want to see in the data. Intense pressure is applied to produce results
that agree with prior published effects, regardless of whether those effects
are real.
13. The complete and final report of his fraud, published by the University of
Tilburg, can be read here: https://www.tilburguniversity
.edu/upload/3ff904d7–547b-40ae-85fe-bea38e05a34a_Final%20report%
20Flawed%20Science.pdf.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 5 | 239
14. For the interim report of the Stapel case, see http://web.archive.org/web
/20160627142859/https://www.tilburguniversity.edu/upload/547aa461-6cd1
-48cd-801b-61c434a73f79_interim-report.pdf.
15. Kate Kelland, “Dutch psychologist admits he made up research data,” Reuters,
2 November 2011, http://www.reuters.com/article/us-dutch-scientist
-fraud-idUSTRE7A12PL20111102.
16. Number 1 in this unenviable ranking is anesthesiologist Yoshitaka Fujii, who
since 2012 has had 183 papers retracted for data fabrication. For the full list,
see http://retractionwatch.com/the-retraction-watch-leaderboard/.
17. http://retractionwatch.com/2014/10/03/curtain-up-on-second-act-for-dutch
-fraudster-stapel-college-teacher/#comment-31309.
18. http://retractionwatch.com/2014/10/03/curtain-up-on-second-act-for-dutch
-fraudster-stapel-college-teacher/#comment-31388.
19. The report rather damningly notes, “Suspicions about data provided by Mr
Stapel had also arisen among fellow full professors on two occasions in the
past year. These suspicions were not followed up. The Committee concludes
that the three young whistle-blowers showed more courage, vigilance and
inquisitiveness than incumbent full professors” (46), https://www.tilburg
university.edu/upload/3ff904d7–547b-40ae-85fe-bea38e05a34a_Final%
20report%20Flawed%20Science.pdf.
20. Stapel, Faking Science, trans. Brown, 88.
21. Daniele Fanelli, “How many scientists fabricate and falsify research? A systematic
review and meta-analysis of survey data,” PLOS ONE 4, no. 5 (2009):
e5738, http://dx.doi.org/10.1371/journal.pone.0005738.
22. Leslie K. John, George Loewenstein, and Drazen Prelec, “Measuring the
prevalence of questionable research practices with incentives for truth telling,”
Psychological Science (2012): 0956797611430953, http://dx.doi.org/10
.1177/0956797611430953.
23. Parts of this section appeared in a blog post at SciLogs: Chris Chambers,
“Tackling the F Word,” http://web.archive.org/web/20160316092653/http://
www.scilogs.com/sifting_the_evidence/tackling-the-f-word/.
24. For the initial report on the Smeesters case, see http://www.eur.nl/fileadmin
/ASSETS/press/2012/Juli/report_Committee_for_inquiry_prof._Smeesters
.publicversion.28_6_2012.pdf.
25. Rolf Zwaan, personal interview via e-mail, 28 August 2015.
26. The violation of the law of small numbers in Smeesters’s data uncovered by
Uri Simonsohn (see chapter 4) provides evidence that the data was generated
artificially, although this evidence fell short of providing a sufficient
standard to officially conclude fraud. Zwaan told me that the committee
couldn’t see any way either p-hacking or sloppy practices could produce such
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
240 | Notes to Chapter 5
a pattern. “We tried to think of innocent copy and paste errors in Excel
could have been the source of the problem. We couldn’t think of any way in
which this could happen but were also unable to completely rule it out.”
27. http://retractionwatch.com/2014/03/19/final-report-in-smeesters-case
-serves-up-seven-retractions/.
28. “Hoogleraar Erasmus ontkent fraude,” http://www.nu.nl/algemeen/2844
322/hoogleraar-erasmus-ontkent-fraude.html. Separately, Zwaan told me
that Smeesters was probably encouraged to resign by Erasmus University. “I
suspect he was pushed but this probably happened in his home department
and/or at higher levels.”
29. See http://dx.doi.org/10.6084/m9.figshare.3381574.v1. In the same interview,
Smeesters admits that he would do things differently next time. “I do know
that if I were to start over, I would deal with issues that are more relevant to
society and that contribute to the wellbeing of people. Then, it would be of
lesser importance whether or not something is innovative theoretically or
statistically significant. Journals almost exclusively publish studies with statistically
significant results, which can indeed lead to massaging your data.”
30. For the 2012 internal report written by the whistle-blower in the Förster
case, see https://retractionwatch.files.wordpress.com/2014/04/report
_foerster.pdf.
31. Bruce S. Weir, “The rarity of DNA profiles,” Annals of Applied Statistics 1, no.
2 (2007): 358, http://dx.doi.org/10.1214%2F07-AOAS128.
32. The whistle-blower told me: “[B]ecause all the data were gone, the committee
could not investigate whether this was actual fraud; hence there was no
verdict of any violation of scientific misconduct. . . . The highest administrators
of UVA [University of Amsterdam], one of the biggest universities in the
Netherlands, thought that because he’d lost all the data, he’s excused.”
33. For the English translation of the LOWI report, see https://retractionwatch
.files.wordpress.com/2014/05/translation_lowi.pdf. For reasons that remain
unclear, LOWI chose to scrutinize just one of the three publications highlighted
by the whistle-blower—the article by Förster and Denzler (2012).
The whistle-blower himself is unclear why this is, although it may have been
a limitation imposed by the university. “I have an email of the former chair
of the LOWI indicating that the University of Amsterdam only sent to the
LOWI the data from the last paper, which is the only one of the three papers
that had a co-author,” he said. Thus, although Förster claimed that the raw
data had been destroyed, summary data files were eventually provided for at
least this one paper. However, the whistle-blower questions the veracity of
this data, suggesting that during a period of sick leave, Förster had ample
time to read up on the Stapel fraud case and learn how to beat the system in
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 5 | 241
a way that the university was unable or unwilling to prevent. “The data files
were created about half a year after I filed the complaint. And created files,
not ‘saved as’ files. . . . It was during [Förster’s] 8 month sick leave. He had 7
months full time to make up the data. The data sets are completely useless.
. . . Don’t you think that if he had 7 or 8 months full time to save his career,
that he would have read the Stapel report? He did his homework. The
problem is the University of Amsterdam didn’t investigate this, they completely
refused it. It would have financial, reputational, consequences.” In
response to the allegation that he fabricated the data files after the fact,
Förster claims that the files did indeed contain the original data, but that
changing variable names from German to English led to the formation of a
new data file and thus a data file that postdated the study in question: http://
retractionwatch.com/2014/06/02/forster-on-defense-again-this-time
-weighing-in-on-timeline-controversy/. At the time of writing, however, he
has yet to publicly release this data or the original time-stamped files in German,
stating “I thought about [posting the data], but my current experience
with ‘the net’ prevents me from doing this. I will share the data with scientists
who want to have a look at it and who are willing to share their results
with me. But I will not leave it to an anonymous crowd that can post whatever
it wants, including incorrect conclusions and insults.” http://retraction
watch.com/2014/05/12/i-never-manipulated-data-forster-defends-actions-in
-open-letter/.
34. “Social psychologist Förster denies misconduct, calls charge ‘terrible misjudgment,’”
http://retractionwatch.com/2014/04/30/social-psychologist
-forster-denies-misconduct-calls-charge-terrible-misjudgment/.
35. “‘I never manipulated data’: Förster defends actions in open letter,” http://
retractionwatch.com/2014/05/12/i-never-manipulated-data-forster-defends
-actions-in-open-letter/. He elaborated that his lack of clear records prevented
his own investigation of who might be responsible. “During the time
of [the] investigation I tried to figure out who could have done something
inappropriate. However, I had to accept that there is no chance to trace this
back; after all, the studies were run more than 7 years ago and I am not even
entirely sure when, and I worked with too many people. I also do not want
to point to people just because they are for some reason more memorable
than others.”
36. http://www.socolab.de/main.php?id=66. For additional summaries of the
Förster affair, see http://osc.centerforopenscience.org/2014/05/29/forster
-case/; and the Data Colada investigation, http://datacolada.org/2014/05
/08/21-fake-data-colada/.
37. https://twitter.com/DegenRolf/status/643671007090339840. In November
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
242 | Notes to Chapter 5
2015, Förster agreed to retract two additional articles: http://retractionwatch
.com/2015/11/12/psychologist-jens-forster-settles-case-by-agreeing-to-2
-retractions/.
38. In 2013, the New York Times published an interview with Stapel in which the
headline described his fraud as “audacious”—a synonym for “brave” and
“daring.” The headline was later changed to remove the term. http://www
.nytimes.com/2013/04/28/magazine/diederik-stapels-audacious-academic
-fraud.html.
39. The full retraction notice can be read here: http://cercor.oxfordjournals.org
/content/23/8/2015.full.
40. “Fraud topples second neuroscience word processing paper,”
http://retractionwatch.com/2014/01/13/fraud-topples-second-neuroscience
-word-processing-paper/.
41. This was also confirmed by the University of Leuven Commission of Scientific
Integrity, which appended the original retraction notice at Cerebral Cortex
with a statement making clear that “the analysis of the data represented
in this paper was manipulated intentionally by the first author: Wouter
Braet.” See http://cercor.oxfordjournals.org/content/24/1/280.full.
42. Hans Op de Beeck, personal interviews via e-mail, August–December
2015.
43. http://www.standaard.be/cnt/dmf20130816_00694853. I thank Frederick
Verbruggen for supplying an English translation. In writing this book I contacted
Braet, but he declined the invitation to be interviewed. Op de Beeck
feels that this was an understandable reaction because, in his view, Braet’s
case was covered overly negatively by the Belgian media: “Even the so-called
high-quality newspapers behaved as tabloids.”
44. https://web.archive.org/web/20150907031741/https://ori.hhs.gov/content
/case-summary-savine-adam-c.
45. http://ccpweb.wustl.edu/ORIresponse.html#13.
46. Remarkably, Braver was made aware of the Office of Research Integrity’s
conclusion of academic fraud only after it was made public. Speaking at the
time to a journalist from the St. Louis Post-Dispatch, he said, “You learned of
the outcome of Adam’s case before I did, which I am pretty upset about—
given that I was the one to report him in the first place.” http://www.stltoday
.com/lifestyles/health-med-fit/washington-u-student-s-mentor-talks-about
-discredited-research/article_e2275d60–1ead-5906–851a-59c7a4daf6e5
.html. In 2015, Braver told me that ORI have since revised their policy to
keep principal investigators better informed.
47. http://ccpweb.wustl.edu//ORIresponse.html.
48. Todd Braver, personal interview via e-mail, 24 August 2015.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 6 | 243
49. The society’s full ethical policy can be found at https://www.sfn.org/member
-center/professional-conduct/sfn-ethics-policy.
50. When I asked the ethics section of the Society for Neuroscience to comment
on the ethical justification for their policy of collective punishment, my enquiry
was redirected to their Communications and Marketing division.
Their senior director of marketing stated that they “do not comment on individual
cases” and referred me to their online ethics policy, which—as I
made clear in my enquiry—makes no effort to justify the society’s policy of
collective punishment.
51. http://www.stltoday.com/lifestyles/health-med-fit/washington-u-student-s
-mentor-talks-about-discredited-research/article_e2275d60-1ead-5906-851a
-59c7a4daf6e5.html.
52. To ensure Kate’s status as a protected whistle-blower, it is impossible to corroborate
all of the content here with independent sources.
53. To protect Kate’s identity within a fairly small research field, I cannot describe
in detail the specific area in which she worked.
Chapter 6. The Sin of Internment
1. http://www.budapestopenaccessinitiative.org/read.
2. https://en.wikipedia.org/wiki/Creative_Commons_license. This ideal model
of OA is called libre OA, in which users are granted rights not only to read
the article but to reuse it.
3. Even traditional subscription publishing isn’t necessarily free for authors.
Many subscription journals, such as the Journal of Neurophysiology, levy page
charges against authors—even when readers have to pay again to view content—and
many journals in psychology and neuroscience include an additional
fee to publish figures in color as opposed to black and white. As the
saying goes, “there is always another fee.”
4. Like many new journals, Archives of Scientific Psychology offers a temporary
waiver on article processing charges, in this case until December 2016. At
the time of writing, BMC Psychology charges authors US$2,145 per article,
which is relatively high compared to other OA journals (e.g., PLOS ONE:
US$1,495) and much higher than others (e.g., PeerJ: US$1,095). Some journals
offer standing fee waivers or reductions for authors in low-income
countries or for authors who are unable to cover the charges.
5. An additional nuance here is that a number of publishers permit immediate
green OA on the author’s personal website but impose a 12-month embargo
for placing the same document in a public open access repository. In theory,
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
244 | Notes to Chapter 6
these forms of publishing are equally public; however, as the publishers are
no doubt aware, the public repository is more likely to be picked up by automated
search engines such as Google Scholar or a simple Internet search—
thus the embargo serves the publisher’s goal of temporarily limiting the visibility
of the work in order to stimulate the market for sales. For a detailed
and comprehensive database of publisher’s policies regarding green and gold
OA, the reader is directed to the excellent SHERPA/ROMEO website,
http://www.sherpa.ac.uk/romeo/.
6. For a full directory of all OA journals, the reader is pointed to https://doaj
.org/. It is notable that of more than 200 journals in this directory that are
linked to psychology or neuroscience (journal title search for “psych* OR
neur*” not one is regarded by the field as a prestigious and prominent outlet.
7. Parts of this section have appeared in the following Guardian blog: Chris
Chambers, “Those who publish research behind paywalls are victims not
perpetrators,” http://www.theguardian.com/science/blog/2013/jan/23/open
-access-publish-paywalls-victims-perpetrators.
8. ArXiv was launched in 1991 and now receives more than 8,000 submissions
per month, http://arxiv.org/. A similar initiative has since been launched in
biology (http://biorxiv.org/), and most recently in psychology (https://osf
.io/view/psyarxiv). The OA journal PeerJ also offers a preprint archiving
service.
9. Björn Brembs, Katherine Button, and Marcus Munafò, “Deep impact: Unintended
consequences of journal rank,” Frontiers in Human Neuroscience 7
(2013): 291, http://dx.doi.org/10.3389/fnhum.2013.00291.
10. For an impassioned argument on the morality of barrier-based publishing,
see this Guardian blogpost by Mike Taylor: http://www.theguardian.com
/science/blog/2013/jan/17/open-access-publishing-science-paywall-immoral.
11. For a damning analysis of this policy as launched at Queen Mary University
London, the reader is directed to David Colquhoun’s post, http://www
.dcscience.net/2012/06/29/is-queen-mary-university-of-london-trying-to
-commit-scientific-suicide/. Further examples of academic bean counting
will be covered in chapter 7.
12. http://www.nature.com/news/funders-punish-open-access-dodgers-1.15007.
The US National Institutes of Health have adopted a similar policy since
2008, which has experienced similar compliance problems despite
sanctions.
13. http://www.wellcome.ac.uk/Managing-a-grant/End-of-a-grant/wtx026513
.htm.
14. http://blog.wellcome.ac.uk/2015/03/03/the-reckoning-an-analysis-of
-wellcome-trust-open-access-spend-2013-14/.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 6 | 245
15. http://blog.wellcome.ac.uk/2015/10/22/10-years-of-open-access-at-the
-wellcome-trust-in-10-numbers/.
16. The Finch Report: Accessibility, Sustainability, Excellence: How to Expand Access
to Research Publications, 5. Available for free download at http://web
.archive.org/web/20160322075144/http://www.researchinfonet.org/wp
-content/uploads/2012/06/Finch-Group-report-FINAL-VERSION.pdf.
17. Ibid., 102.
18. Ibid., 7. One problem with gold OA is the question of who pays when neither
the institution nor the researcher have any funds available. This can be
a particular problem for early-career researchers based at less lucrative institutions.
As a colleague put it: “My institution has said that they won’t pay, so
I’ve been stuck with the PLOS ONE fee. I asked them to waive it, but all they
could do is reduce it. That’s still $400 I have to find from somewhere. I’m
still an early career researcher, so I don’t have any slush funds to pay for it.”
19. Ibid., 6.
20. Ibid., 36.
21. Harnad’s full comment can be read at https://web.archive.org/web/201604
18152013/http://openaccess.eprints.org/index.php?/archives/904-Finch
-Report,-a-Trojan-Horse,-Serves-Publishing-Industry-Interests-Instead-of-UK
-Research-Interests.html.
22. Stevan Harnad has referred to this double-dipping model as “fool’s gold OA”:
http://openaccess.eprints.org/index.php?/archives/1007-Pre-Green-OA-Fools
-Gold-vs.-Post-Green-OA-Fair-Gold.html.
23. http://www.rcuk.ac.uk/research/openaccess/policy/.
24. “From ‘as soon as possible’ to ‘immediate’ open access,” http://www.nwo.nl
/en/news-and-events/news/2015/from-as-soon-as-possible-to-immediate
-open-access.html
25. “Open access in the next Research Excellence Framework: policy adjustments
and qualifications,” http://www.hefce.ac.uk/media/HEFCE,2014
/Content/Pubs/2015/CL,202015/Print-friendly%20version.pdf.
26. Since 2012 the European Research Council has implemented an open access
requirement on all publications arising from its grants, including psychology.
It’s policy requires that researchers:
(a) as soon as possible and at the latest on publication, deposit a machine-readable
electronic copy of the published version or final peer-reviewed
manuscript accepted for publication in a repository for scientific
publications. Moreover, the beneficiary must aim to deposit at the same
time the research data needed to validate the results presented in the deposited
scientific publications.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
246 | Notes to Chapter 6
(b) ensure open access to the deposited publication—via the repository—at
the latest:
(i) on publication, if an electronic version is available for free via
the publisher, or
(ii) within six months of publication (twelve months for publications
in the social sciences and humanities) in any other case.
(c) ensure open access—via the repository—to the bibliographic
metadata that identify the deposited publication, which must include a
persistent identifier.
http://ec.europa.eu/research/participants/data/ref/h2020/grants_manual/
amga/h2020-amga_en.pdf.
27. Andrea Kuszewski’s tweet from 2011 that started the #icanhazpdf phenomenon:
https://twitter.com/AndreaKuszewski/status/28257118322688000.
28. For an insightful analysis of the #icanhazpdf phenomenon, see Carolyn Caffrey
Gardner and Gabriel J. Gardner, “Bypassing interlibrary loan via twitter:
An exploration of# icanhazpdf requests,” ACRL (Association of College and
Research Libraries) (2015), http://web.archive.org/web/20160702052610/
http://eprints.rclis.org/24847/2/gardner.pdf. During a three-month period in
2014, the authors found 824 #icanhazpdf requests in the Twitter archives,
which extrapolates to approximately 3,296 requests over 12 months. Most
requests (73 percent) were in the social and life sciences (including
psychology).
29. At the time of writing, Sci-Hub could be found under several domains, including
http://sci-hub.bz/ and http://sci-hub.cc/.
30. In particular, Elbakyan argues that publishers such as Elsevier stand in violation
of the following clause of Article 27: “(1) Everyone has the right freely
to participate in the cultural life of the community, to enjoy the arts and to
share in scientific advancement and its benefits.” http://www.un.org/en
/universal-declaration-human-rights/.
31. “Sci-Hub tears down academia’s ‘illegal’ copyright paywalls,” https://
torrentfreak.com/sci-hub-tears-down-academias-illegal-copyright-paywalls
-150627/.
32. In 2014, Elsevier—which is the largest publisher of academic journals—was
reported to return a 37 percent profit margin, equating to over $US1 billion,
http://web.archive.org/web/20160315064211/https://libraries.mit.edu/
scholarly/mit-open-access/open-access-at-mit/mit-open-access-policy/
publishers-and-the-mit-faculty-open-access-policy/elsevier-fact-sheet/.
33. The full complaint from Elsevier can be read here: https://torrentfreak.com
/images/elsevier-complaint.pdf.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 6 | 247
34. “Court orders shutdown of Libgen, Bookfi and Sci-Hub,” https://torrentfreak
.com/court-orders-shutdown-of-libgen-bookfi-and-sci-hub-151102/.
35. “Sci-Hub, BookFi and LibGen resurface after being shut down,”
https://torrentfreak.com/sci-hub-and-libgen-resurface-after-being-shut
-down-151121/.
36. For details of the indictment against Swartz, see http://www.wired.com
/images_blogs/threatlevel/2012/09/swartzsuperseding.pdf.
37. The poorest institutions and countries are somewhat protected from this
problem. Several major publishers participate in the Research4Life program,
which at the time of writing offers free journal access to 72 “Group A” countries
in the developing world, http://www.research4life.org/eligibility/.
38. In one of many such examples, Ross Mounce, a postdoctoral researcher at
the Natural History Museum in London, recounts his experience of discovering
that the University of Bath cut its subscription to the popular Royal Society
journal, Biology Letters, http://rossmounce.co.uk/2012/02/12/journal
-mega-bundles-thecostofknowledge/. In this case it appears the subscription
was axed not only to save money but also because it didn’t belong to a purchased
“bundle.” The largest for-profit publishers, such as Elsevier, often sell
journal subscriptions to universities in large bundles and do not allow the libraries
to selectively unsubscribe from specific journals. This means that
when the purse strings are tightened, more isolated subscriptions—such as
those to learned societies—can be more vulnerable to cuts (although note
that at the time of writing, the University of Bath has since restored its subscription
to Biology Letters). The details of library subscriptions, and what
they cost, are often treated as commercial secrets, although a 2014 study by
Theodore Bergstrom and colleagues was able to glean information from US
universities using Freedom of Information requests: Theodore C. Bergstrom,
Paul N. Courant, R. Preston McAfee, and Michael A. Williams, “Evaluating
big deal journal bundles,” Proceedings of the National Academy of Sciences USA
111, no. 26 (2014): 9425–30, http://dx.doi.org/10.1073/pnas.1403006111. In
2015, Mounce conducted a poll on social media to discover how many leading
universities could access a randomly selected article from 1949—an article
that was, in theory, available electronically. Of 70 institutions in the UK
and North America, just 10 reported that they could successfully download
the article: http://rossmounce.co.uk/2015/09/16/who-actually-has-access-to
-paywalled-research/.
39. David Halpern, “Presidential column: Applying psychology to public policy,”
http://www.psychologicalscience.org/index.php/publications/observer/2014
/january-14/applying-psychology-to-public-policy.html.
40. http://www.behaviouralinsights.co.uk/.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
248 | Notes to Chapter 6
41. “Executive order—using behavioral science insights to better serve the
American people,” https://www.whitehouse.gov/the-press-office/2015/09/15
/executive-order-using-behavioral-science-insights-better-serve-american.
42. POST is an in-house group of civil servants that provides independent, indepth
scientific briefings called POSTNotes to members of Parliament,
http://www.parliament.uk/post.
43. Adriana De Palma, “Why all PhD students should do a policy placement,”
http://web.archive.org/web/20160601081827/https://therostrumblog.word
press.com/2015/01/12/why-all-phd-students-should-do-a-policy-placement/.
44. https://twitter.com/stevenhill/status/670645979167748100.
45. https://twitter.com/ersatzben/status/670939925106302977.
46. https://twitter.com/ersatzben/status/613782419011989504.
47. Ben Johnson, “Subscription publishers do not want my business,” http://web
.archive.org/web/20160409024908/https://ersatzben.com/2015/06/20
/subscription-publishers-do-not-want-my-business/. A similar case was documented
by former BMJ editor Richard Smith, http://web.archive.org/web
/20160414122448/http://blogs.bmj.com/bmj/2012/06/28/richard-smith-a
-bad-bad-week-for-access/.
48. Chris Hartgerink, “Why I content mine,” http://web.archive.org/web/201606
12071438/http://onsnetwork.org/chartgerink/2015/11/19/why-i-content
-mine/.
49. Chris Hartgerink, “Elsevier stopped me doing my research,” http://web
.archive.org/web/20160611032746/http://onsnetwork.org/chartgerink/2015
/11/16/elsevier-stopped-me-doing-my-research/. One of the most remarkable
aspects to this case is that Hartgerink’s activities appeared to be assumed,
without evidence, to be “stealing” of content. In an e-mail forwarded to
Hartgerink, a New York employee of Elsevier allegedly wrote: “I have noticed
what seems to be crawling at Tilburg recently. Recently we are seeing a
lot of content theft by people who steal legitimate IDs to gain access to university
resources. Will you please reach out to your contacts at Tilburg to ask
them to investigate.” Subsequently, the Dutch account manager contacted
Tilburg (in Dutch, translated by Hartgerink) and allegedly wrote: “Given
below is serious amounts of downloading which could indicate potential
abuse [of access]. Could I ask you to investigate as soon as possible what is
going on? If the downloads continue at this scale the access will be blocked
for security reasons.”
50. http://web.archive.org/web/20160611032746/http://onsnetwork.org/chart
gerink/2015/11/16/elsevier-stopped-me-doing-my-research/#comment-11.
51. Chris Shillum, “Elsevier updates text-mining policy to improve access for re-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 6 | 249
searchers,” http://www.elsevier.com/connect/elsevier-updates-text-mining
-policy-to-improve-access-for-researchers.
52. Chris Hartgerink, “Why Elsevier’s ‘solution’ is the problem,” http://web
.archive.org/web/20160611010128/http://onsnetwork.org/chartgerink
/2015/11/20/why-elseviers-solution-is-the-problem/.
53. The full memo is no longer available from Harvard University but can be
read at http://web.archive.org/web/20160317160330/http://gantercourses
.net/wp-content/uploads/2013/11/Faculty-Advisory-Council-Memorandum
-on-Journal-Pricing-%C2%A7-THE-HARVARD-LIBRARY.pdf.
54. https://osc.hul.harvard.edu/policies/.
55. “Dutch universities start their Elsevier boycott plan,” https://universonline
.nl/2015/07/02/dutch-universities-start-their-elsevier-boycott-plan.
56. Daniel Allington, “On open access, and why it’s not the answer,” http://web
.archive.org/web/20160402004400/http://www.danielallington.net/2013/10
/open-access-why-not-answer/.
57. http://en.wikipedia.org/wiki/Out-group_homogeneity.
58. For an excellent overview of the benefits of open access publishing to those
outside academia, see http://whoneedsaccess.org/.
59. John Bargh, “Nothing in their heads,” http://web.archive
.org/web/20120309203900/http://psychologytoday.com/blog/the-natural
-unconscious/201203/nothing-in-their-heads.
60. Stéphane Doyen, Olivier Klein, Cora-Lise Pichon, and Axel Cleeremans,
“Behavioral priming: It’s all in the mind, but whose mind?,” PLOS ONE 7, no.
1 (2012): e29081, http://dx.doi.org/10.1371/journal.pone.0029081.
61. http://journals.plos.org/plosone/s/criteria-for-publication.
62. Bohannon’s report can read freely at http://www.sciencemag.org/content
/342/6154/60.full.
63. https://doaj.org/.
64. Michael Eisen, “I confess, I wrote the arsenic DNA paper to expose flaws in
peer-review at subscription based journals,” http://web.archive.org/web
/20160508112019/http://www.michaeleisen.org/blog/?p=1439.
65. C. Hajjem, Y. Gingras, and S. Harnad, “Ten-year cross-disciplinary comparison
of the growth of open access and how it increases research citation impact,”
Bulletin of the IEEE Computer Society Technical Committee on Data Engineering
28 no. 4 (2005): 39–46, http://sites.computer.org/debull/A05dec
/hajjem.pdf.
66. For details, see http://sparceurope.org/oaca/.
67. For an insightful and authoritative review of the benefits of open access publishing,
the reader is directed to Jonathan P. Tennant, François Waldner,
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
250 | Notes to Chapter 7
Damien C. Jacques, Paola Masuzzo, Lauren B. Collister, and Chris H. J. Hartgerink,
“The academic, economic and societal impacts of Open Access: An
evidence-based review,” F1000Research 5 (2016), http://dx.doi.org/10.12688
%2Ff1000research.8460.2.
Chapter 7. The Sin of Bean Counting
1. Goodhart’s law is named after the economist Charles Goodhart, although
the formulation quoted here has been attributed to anthropologist Marilyn
Strathern. http://www.atm.damtp.cam.ac.uk/mcintyre/papers/LHCE
/goodhart.html.
2. Eugene Garfield, “Citation indexes for science,” Science 122 (1955): 108–11,
http://www.garfield.library.upenn.edu/papers/science_v122v3159p108y1955
.html.
3. Editorial: Scientometrics 1, no. 1 (1978): 3–8, http://dx.doi.org/10.1007
/BF02016836.
4. E. Garfield, “The history and meaning of the journal impact factor,” JAMA
295 no. 1 (2006): 90–93. The article can be freely accessed here:
http://garfield.library.upenn.edu/papers/jamajif2006.pdf.
5. Papers can differ widely in the number of citations they receive for reasons
unrelated to study quality. Articles can be highly cited to signpost a notable
flaw or because of the prominence of the journal in which the article appeared.
Studies can also be popular because they are novel (even if low quality)
or because of impact in the print and broadcast media. On top of this,
different scientific disciplines, and subdisciplines, have different baselines
for what constitute a typical number of citations, making it difficult to draw
comparisons. For an insightful discussion of these problems and more, see
Robert Adler, John Ewing, and Peter Taylor, “Joint committee on quantitative
assessment of research: Citation statistics,” Australian Mathematical Society
Gazette 35, no. 3 (2008): 166–88, http://www.austms.org.au/Gazette
/2008/Jul08/Gazette35%283%29Web.pdf#page=24.
6. See Per O. Seglen, “Why the impact factor of journals should not be used for
evaluating research,” BMJ: British Medical Journal 314, no. 7079 (1997): 498,
which showed no correlation between JIF and citation rate, http://www.ncbi
.nlm.nih.gov/pmc/articles/PMC2126010/pdf/9056804.pdf. See also George
A. Lozano, Vincent Larivière, and Yves Gingras, “The weakening relationship
between the impact factor and papers’ citations in the digital age,” Journal
of the American Society for Information Science and Technology 63, no. 11
(2012): 2140–45, which found a low correlation that is diminishing over
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 7 | 251
time, http://dx.doi.org/10.1002/asi.22731. This article can be freely downloaded
from http://www.ost.uqam.ca/Portals/0/docs/articles/2012
/JASISTno11.pdf.
7. As a personal example, one of my most highly cited articles, at the time of
writing, is published in the specialist Journal of Cognitive Neuroscience in
2006 (371 cites; JIF = 3.6). By comparison, a similar experimental paper
that I published two years earlier in the “high impact” journal Nature Neuroscience
has less than half the number of citations (176 cites; JIF = 16.7).
8. Ferric C. Fang, R. Grant Steen, and Arturo Casadevall, “Misconduct accounts
for the majority of retracted scientific publications,” Proceedings of the
National Academy of Sciences USA 109, no. 42 (2012): 17028–33, http://dx
.doi.org/10.1073/pnas.1212247109. See also a summary of this and other
problems with JIF on Björn Bremb’s blog: http://blogarchive.brembs.net
/news.php?item.766.11.
9. See PLOS Medicine editors, “The impact factor game,” PLOS Med 3, no. 6
(2006): e291, http://dx.doi.org/10.1371/journal.pmed.0030291.
10. See Björn Brembs, Katherine Button, and Marcus Munafò, “Deep impact:
Unintended consequences of journal rank,” Frontiers in Human Neuroscience
7 (2013): 291, http://dx.doi.org/10.3389/fnhum.2013.00291.
11. Eugene Garfield, “Interview with Eugene Garfield, chairman emeritus of the
Institute for Scientific Information (ISI),” Cortex 37, no. 4 (2001): 575–77.
The article can be freely accessed at http://www.garfield.library.upenn.edu
/papers/cortex37%284%292001.pdf.
12. At some Dutch universities there is an unofficial rule that students are expected
to publish a minimum number of articles in peer-reviewed journals
(usually at least two) before they can pass their PhD. On the positive side,
this ensures that the student gains experience in the practice of scientific
publishing; however, the much larger negative is that setting a minimum
quantity also encourages small, underpowered studies and the salami slicing
of work into minimal publishable units. Because of publication bias, it also
increases the pressure on PhD students to engage in biased research practices
(such as p-hacking) to ensure that their experiments produce positive,
publishable results.
13. http://www.ascb.org/dora/. The Association of Dutch Universities, of which
the University of Maastricht is a member, is organizational signatory 561.
14. See Brembs, Button, and Munafò, “Deep impact,” 291, http://dx.doi.org/10
.3389/fnhum.2013.00291.
15. Stephen Curry, “Sick of impact factors,” http://occamstypewriter
.org/scurry/2012/08/13/sick-of-impact-factors/.
16. Henry L. Roediger III, “Journal impact factors: How much should we care?,”
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
252 | Notes to Chapter 7
http://www.psychologicalscience.org/index.php/publications/observer/2013
/september-13/journal-impact-factors.html.
17. https://uk.sagepub.com/en-gb/eur/psychological-science-package
/journal201991.
18. https://twitter.com/CT_Bergstrom/status/657221745628200960.
19. http://www.ucl.ac.uk/hr/docs/proms/SnrProm_AnnexA.doc. In the event of
this link being taken down, an archived copy can be found at https://web
.archive.org/web/20160516082534/http://www.ucl.ac.uk/hr/docs/proms
/SnrProm_AnnexA.doc.
20. For a recent example of how the Times Higher Education reinforces the idea
that grants are prizes, see https://www.timeshighereducation.com/news
/grant-winners-3-december-2015.
21. Dorothy Bishop has delivered a blistering attacking on the absurdity of evaluating
scientists based on their grant income: http://deevybee.blogspot.co
.uk/2014/12/why-evaluating-scientists-by-grant.html.
22. In general there is very little assessment of whether grants awarded to psychologists
are well spent. One exception is the UK Economic and Social Research
Council’s rapporteur program, which asks reviewers to assess the
“scientific, economic and societal impact” of outputs stemming from
awarded grants, http://www.esrc.ac.uk/files/funding/guidance-for-peer
-reviewers/guidance-notes-on-the-rapporteur-quality-and-impact-comment
-form/. However, it is unclear how the feedback is used, and whether the
funder uses the information from the rapporteur reports to inform future
funding decisions.
23. In psychology it is possible to conduct entire research programs on the basis
of little-to-no grant funding, for instance by conducting research projects
with undergraduate research trainees or PhD students.
24. I put this scenario to a Twitter poll in December 2015, “Jane & Mary compete
for academic promotion. They’re matched in every way but Mary has
grants. Who should/would get promoted?” Of 98 respondents, 57 percent
believed that “Mary should and would” be promoted over Mary, while only
6 percent felt the “Jane should and would.” Interestingly, a substantial number
(33 percent) believed that “Mary would but Jane should,” perhaps reflecting
some recognition that inputs (grants) and outputs (publications, influence)
should be weighed against each other rather than summed. Very
few respondents (4 percent) felt that “Jane would but Mary should.” Results
of Twitter polls should be viewed with caution—this was not a properly sampled
survey and the background of the respondents is unknown, https://
twitter.com/chrisdc77/status/677829444103442432.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 7 | 253
25. UCL pharmacologist David Colquhoun has doggedly investigated and exposed
these cases. http://www.dcscience.net/2012/06/29/is-queen-mary
-university-of-london-trying-to-commit-scientific-suicide/.
26. David Colquhoun, “Bad financial management at Kings College London
means VC Rick Trainor is firing 120 scientists,” http://www.dcscience.net
/2014/06/07/bad-financial-management-at-kings-college-london-means-vc
-rick-trainor-is-firing-120-scientists/.
27. John Allen and Fanis Missirlis’s critique of Queen Mary’s retention policy
can be read freely at the Lancet: John F. Allen and Fanis Missirlis, “Queen
Mary: Nobody expects the Spanish Inquisition,” Lancet 379, no. 9828
(2012): 1785, http://www.thelancet.com/journals/lancet/article/PIIS0140–
6736%2812%2960697–7/fulltext.
28. The COPE guidelines can be freely read at http://publicationethics.org
/files/International%20standards_authors_for%20website_11_Nov_
2011_0.pdf.
29. Like all rules of thumb surrounding authorship, there are many exceptions.
For instance, in cases where two senior researchers work on a project together,
it is not uncommon for one researcher to take first authorship and
the other to take last authorship, despite both authors being of equal seniority.
Another caveat with last authorship is that its value is context dependent:
for researchers who are demonstrably independent, or on the cusp of
independence, last authorship is sign of academic leadership. But junior
scholars are generally better off settling for a second authorship compared
with a last authorship. If the community knows you are too junior to be a
principal investigator, a last authorship can be interpreted literally as contributing
the least to a study. The whims and fancies of such conventions
deliver another nail in the coffin to authorship rank as a marker of scientific
contribution.
30. For the American Psychological Association’s guidelines on authorship, see
http://www.apa.org/research/responsible/publication/.
31. For the ICMJE guidelines on authorship, see http://www.icmje
.org/recommendations/browse/roles-and-responsibilities/defining-the-role
-of-authors-and-contributors.html.
32. My article, entitled “Tough love II: 25 tips for early-career scientists”:
http://neurochambers.blogspot.co.uk/2012/05/tough-love-ii-25-point-guide
-for-early.html.
33. Dorothy Bishop’s well-aimed critique of my post can be read here: http://
neurochambers.blogspot.co.uk/2012/05/tough-love-ii-25-point-guide-for
-early.html?showComment=1338490130354#c1743257197227672347.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
254 | Notes to Chapter 8
34. My rather weak response to Dorothy: http://neurochambers.blogspot.co.uk
/2012/05/tough-love-ii-25-point-guide-for-early.html?showComment=1338
498718060#c5457911252285746290.
35. Uta Frith’s manifesto for slow science is pure wisdom. See http://frithmind
.org/socialminds/2015/10/11/slow-science/.
36. This isn’t always the case. It is not uncommon for papers to be cited inappropriately
based on superficial or erroneous reading of the work by the
authors.
37. Science is awash with metrics in addition to JIF. The Eigenfactor seeks to
quantify the citation impact of a journal by considering not just the number
of citations the articles receive but also the impact of the journal in which
the citing article occurs. The h-index (named after its creator, Jorge E.
Hirsch) attempts to reduce both the productivity and impact of individual
scientists into a single number. A scientist is said to have an h-index of 20 if
he or she has 20 articles that are cited at least 20 times each. In addition to
focus solely on citations, the h-index has been criticized for favoring senior
over junior scientists, failing to distinguish individual contributions to authorship,
and being vulnerable to manipulation through self-citation. Socalled
altmetrics are another burgeoning set of measures that seek to quantify
the impact of academic work in the print media, blogosphere, and social
media. All such metrics tell us something, but none are suitable for expert
judgments of quality and long-term consideration of how published research
impacts theory and practice.
Chapter 8. Redemption
1. It should also be pointed out that just because an area of research doesn’t
meet all these conditions for being “science” doesn’t mean that it is flawed,
useless, or not research. Qualitative psychological research, for instance, can
reveal rich insights into behavior and society even if such approaches do not
seek to test hypotheses or quantify phenomena.
2. For a discussion of wider reproducibility problems in biomedicine, see the
2015 report issued by the UK Academy of Medical Sciences: http://www
.acmedsci.ac.uk/policy/policy-projects/reproducibility-and-reliability-of
-biomedical-research/.
3. Parts of this section are adapted from the following articles: Chris Chambers,
“Are we finally getting serious about fixing science?,” http://www
.theguardian.com/science/head-quarters/2015/oct/29/are-we-finally-getting
-serious-about-fixing-science; and Christopher D. Chambers, Eva Feredoes,
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 8 | 255
Suresh Daniel Muthukumaraswamy, and Peter Etchells, “Instead of ‘playing
the game’ it is time to change the rules: Registered reports at AIMS Neuroscience
and beyond,” AIMS Neuroscience 1, no. 1 (2014): 4–17. This article
can freely downloaded from http://www.aimspress.com/article/10.3934
/Neuroscience.2014.1.4/pdf.
4. Chris Chambers, “Why I will no longer review or publish for the journal
Neuropsychologia,” http://neurochambers.blogspot.co.uk/2012/09/why-i-will
-no-longer-review-or-publish.html.
5. Neuroskeptic, “Fixing science—systems and politics,” http://blogs.discover
magazine.com/neuroskeptic/2012/04/14/fixing-science-systems-and-politics/.
6. In my original proposal it was called a registration report, which I soon
changed because it sounded like some form of tedious government
bureaucracy.
7. See R. Rosenthal, Experimenter Effects in Behavioral Research (New York:
Appleton-Century-Croft, 1966). When I wrote to Rosenthal in 2015 to inform
him that we had finally put his plan in place after nearly 50 years, he
said he was pleased that we had implemented his “pipe dream” from the
mid-1960s.
8. G. William Walster and T. Anne Cleary, “A proposal for a new editorial policy
in the social sciences,” American Statistician 24, no. 2 (1970): 16–19, http://dx
.doi.org/10.1080/00031305.1970.10478884. Similar proposals were later
mooted by Robert Newcombe in 1987 and Erick Turner in 2013. See Robert
G. Newcombe, “Towards a reduction in publication bias,” BMJ 295, no. 6599
(1987): 656–59, http://dx.doi.org/10.1136/bmj.295.6599.656, freely available
at http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1257777/pdf
/bmjcred00037–0042.pdf; and Erick H. Turner, “Publication bias, with a
focus on psychiatry: Causes and solutions,” CNS Drugs 27, no. 6 (2013): 457–
68, http://dx.doi.org/10.1007/s40263–013–0067–9.
9. For my original open letter to Cortex, see http://neurochambers.blogspot.co
.uk/2012/10/changing-culture-of-scientific.html.
10. I considered the possibility of proposing the idea initially in private and then
going public only if it was rejected, but I was concerned that this could reek
of sour grapes. Also, this strategy wouldn’t have allowed the proposal to benefit
from wider peer review.
11. At the time of writing, the Registered Reports subcommittee at Cortex includes
me, Dr. Rob McIntosh from the University of Edinburgh, Dr. Pia Rotshtein
from the University of Birmingham, Professor Klaus Willmes from
RWTH Aachen University, and Zoltan Dienes from the University of Sussex.
12. This feature was inspired by a remark from psychologist Hal Pashler, who
in commenting in one of Neuroskeptic’s posts on study preregistration
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
256 | Notes to Chapter 8
(see Neuroskeptic, “Fixing science—systems and politics,” http://blogs
.discovermagazine.com/neuroskeptic/2012/04/14/fixing-science-systems
-and-politics/), said, “reviewers should get to specify some outcome-neutral
criteria for publishing the study, e.g., that you do not have a floor effect or
ceiling effect, that manipulation checks turn out OK, etc. If you don’t do
this, then you are asking journals to precommit to publishing studies that
fail to offer real tests of hypotheses.”
13. For an example of Registered Reports guidelines in full, the reader is directed
to the Cortex format: http://cdn.elsevier.com/promis_misc
/PROMIS%20pub_idt_CORTEX%20Guidelines_RR_29_04_2013.pdf.
14. Perspectives on Psychological Science offers a unique twist on Registered Reports,
focused on replications and calling for multisite collaborations attempts.
It also provides some funding to support these studies. The Perspectives
initiative was devised independently of Registered Reports by Dan
Simons, Alex Holcombe, and others.
15. Chris Chambers, Marcus Munafò, and 83 signatories: “Trust in science
would be improved by study pre-registration,” http://www.theguardian.com
/science/blog/2013/jun/05/trust-in-science-study-pre-registration.
16. The main opposition to Registered Reports is exemplified in this response
to our Guardian article by Sophie Scott, professor of cognitive neuroscience
at University College London: https://www.timeshighereducation.com
/comment/opinion/pre-registration-would-put-science-in-chains/2005954
.article. For a collection of mostly negative reactions assembled by Scott,
see also https://sites.google.com/site/speechskscott/SpeakingOut/willpre
-registrationofstudiesbegoodforpsychology.
17. When I make this point in seminars I sometimes get asked, “Why aren’t you
proposing an initiative to support exploratory science?” The answer is that
we are. An initiative called Exploratory Reports is currently in development
at Cortex, led by Rob McIntosh.
18. Since February 2014, the Open Scoop Challenge has remained standing
and unclaimed: http://software-carpentry.org/blog/2014/02/open-scoop
-challenge.html.
19. An additional disincentive for reviewers to scoop is that the “manuscript received”
date in the final published Registered Report refers to the initial
Stage 1 submission date and so will predate the “manuscript received” date
of any standard submission published by a competitor. Therefore, even if a
reviewer went ahead and stole the idea and published it first, the original authors
could prove they had the idea first.
20. By splitting the review process into two stages and preventing reviewers
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 8 | 257
from assessing the quality of papers according to the results of hypothesis
tests, we also prevent a form of bias by peer reviewers known as CARKing:
“critiquing after results are known” (a term coined by Brian Nosek and
Daniël Lakens: https://osf.io/vwfk2/). CARKing is a form of motivated reasoning
in which a reviewer, objecting to the results, raises spurious concerns
about the methods in order to prevent publication. For conventional (unregistered)
articles, CARKing is impossible to prove—when methods and results
are reviewed at the same time, there is no definitive way to distinguish
a true methodological objection from CARKing. For Registered Reports,
however, any CARKing is obvious and easy to prevent. We see evidence of
CARKing if a reviewer approves a submission at Stage 1 but then raises new
objections to the same methods after results are presented at Stage 2. While
reviewers are free to enter such comments into Stage 2 reviews, the validity
of the (already approved) method is not one of assessment criteria; therefore
CARKing is not only transparent but impossible.
21. Hint: they almost certainly wouldn’t. As a former colleague and eminent
neuroscientist once told me: “Never apply for a grant for something you
haven’t already done.”
22. See https://sites.google.com/site/speechskscott/SpeakingOut/willpre-registra
tionofstudiesbegoodforpsychology.
23. See http://neurochambers.blogspot.co.uk/2012/10/changing-culture-of
-scientific.html?showComment=1349772625668#c58363015480342
36209.
24. See https://sites.google.com/site/speechskscott/SpeakingOut/willpre-registra
tionofstudiesbegoodforpsychology.
25. This insult was an extraordinary misfire given that the signatories included,
among other luminaries, Dorothy Bishop, Morton Ann Gernsbacher, John
Hardy, John Ioannidis, Steven Luck, Barbara Spellman, and Jeremy Wolfe.
26. Some opponents of preregistration suggested that we should have preregistered
the Registered Reports initiative (e.g., https://nucambiguous
.wordpress.com/2013/07/25/preregistration-a-boring-ass-word-for-a-very
-important-proposal/#comment-540). Interestingly, what these critics
seemed not to consider is that by advocating preregistration as a way of presumably
enhancing the credibility of Registered Reports, they tacitly assume
that preregistration does something useful in the first place.
27. Loren K. Mell and Anthony L. Zietman, “Introducing prospective manuscript
review to address publication bias,” International Journal of Radiation
Oncology, Biology, Physics 90, no. 4 (2014): 729–732, http://dx.doi.org/10
.1016/j.ijrobp.2014.07.052.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
258 | Notes to Chapter 8
28. Robert M. Kaplan and Veronica L. Irvin, “Likelihood of null effects of large
NHLBI clinical trials has increased over time,” PLOS ONE 10, no. 8 (2015):
e0132382, http://dx.doi.org/10.1371/journal.pone.0132382.
29. Recent examples at Cortex include Jona Sassenhagen and Ina BornkesselSchlesewsky,
“The P600 as a correlate of ventral attention network reorientation,”
Cortex 66 (2015): A3–A20, http://dx.doi.org/10.1016/j.cortex.2014.12
.019; Tim Paris, Jeesun Kim, and Chris Davis, “Using EEG and stimulus context
to probe the modelling of auditory-visual speech,” Cortex (2015),
http://dx.doi.org/10.1016/j.cortex.2015.03.010. These and more are showcased
in a virtual special issue of Registered Reports at Cortex: http://www
.journals.elsevier.com/cortex/virtual-special-issues/virtual-special-issue
-registered-reports. See also the special issue on Registered Reports of replications
in Social Psychology: http://econtent.hogrefe.com/toc/zsp/45/3.
30. For a full list of currently participating journals and guidelines, see https://
cos.io/rr/.
31. http://www.acmedsci.ac.uk/policy/policy-projects/reproducibility-and
-reliability-of-biomedical-research/.
32. For more information on the TOP guidelines, see https://cos.io/top/.
33. I am the current handling editor for Registered Reports at Royal Society
Open Science, https://blogs.royalsociety.org/publishing/registered-reports/.
Even though the physical sciences suffer less from questionable research
practices and researcher bias, they are nevertheless subject to publication
bias and therefore stand to benefit from Registered Reports.
34. The freely available Open Science Framework (http://osf.io) provides a versatile
registry for many different kinds of research, including psychology.
Other options include Figshare, the American Economic Association’s registry
for randomized trials (https://www.socialscienceregistry.org/), and for
studies with health implications, researchers can also use the ISRCTN registry
(http://www.isrctn.com/) or http://clinicaltrials.gov.
35. Leif Nelson, “Preregistration: Not just for the empiro-zealots,”
http://datacolada.org/12.
36. Sylvain Mathieu, Isabelle Boutron, David Moher, Douglas G. Altman, and
Philippe Ravaud, “Comparison of registered and published primary outcomes
in randomized controlled trials,” JAMA 302, no. 9 (2009): 977–84,
http://dx.doi.org/10.1001/jama.2009.1242.
37. Sreeram Ramagopalan, Andrew P. Skingsley, Lahiru Handunnetthi, Michelle
Klingel, Daniel Magnus, Julia Pakpoor, and Ben Goldacre, “Prevalence of
primary outcome changes in clinical trials registered on ClinicalTrials.gov:
A cross-sectional study,” F1000Research 3 (2014), http://dx.doi.org/10
.12688%2Ff1000research.3784.1.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 8 | 259
38. Sreeram V. Ramagopalan, Andrew P. Skingsley, Lahiru Handunnetthi, Daniel
Magnus, Michelle Klingel, Julia Pakpoor, and Ben Goldacre, “Funding
source and primary outcome changes in clinical trials registered on
ClinicalTrials.gov are associated with the reporting of a statistically significant
primary outcome: A cross-sectional study,” F1000Research 4 (2015),
http://dx.doi.org/10.12688%2Ff1000research.6312.2
39. http://compare-trials.org/#.
40. Annie Franco, Neil Malhotra, and Gabor Simonovits, “Underreporting in
psychology experiments: Evidence from a study registry,” Social Psychological
and Personality Science 7, no. 1 (2016): 8–12, http://dx.doi.org/10.1177/1948
550615598377.
41. The use of standardized templates could help increase the precision of unreviewed
preregistration. A new tool developed by Uri Simonsohn and Joe
Simmons shows promise in this regard: https://aspredicted.org/.
42. Adam Marcus and Ivan Oransky, “Time for a reproducibility index,” http://
www.labtimes.org/labtimes/ranking/dont/2013_04.lasso.
43. A complementary idea for a “Replication Tracker” has been proposed by
Josh Hartshorne and Adena Schachner, http://dx.doi.org/10.3389/fncom
.2012.00008.
44. At least not without a sophisticated artificial intelligence that is beyond our
current capabilities.
45. For details, see http://www.iupsys.net/about/members/national-members
/index.html.
46. One avenue for producing a Reproducibility Index could be to develop it
conjunction with the Curate Science project led by Etienne LeBel,
http://curatescience.org/. This impressive new initiative tracks the replicability
of published results. For an introduction, see http://replicationnetwork
.com/2015/10/21/lebel-introducing-curatescience-org/
47. Leonard P. Freedman, Iain M. Cockburn, and Timothy S. Simcoe, “The economics
of reproducibility in preclinical research,” PLOS Biol 13, no. 6 (2015):
e1002165, http://dx.doi.org/10.1371/journal.pbio.1002165.
48. For the full study, see Open Science Collaboration, “Estimating the reproducibility
of psychological science,” Science 349, no. 6251 (2015): aac4716,
http://dx.doi.org/10.1126/science.aac4716. Alex Etz and Joachim Vandekerckhove
have provided a Bayesian analysis of these results; they find that
about 20 percent of the attempted replications provided strong evidence for
a replication failure and 25 percent provided strong evidence for replication
success, with the remainder occupying a middle ground of moderate or inconclusive
evidence either way: Alexander Etz and Joachim Vandekerckhove,
“A Bayesian perspective on the reproducibility project: Psychology,”
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
260 | Notes to Chapter 8
PLOS ONE 11, no. 2 (2016): e0149794, http://dx.doi.org/10.1371/journal
.pone.0149794. A lucid summary of these findings written for a more general
audience can be found on Alex Etz’s blog: http://alexanderetz.com/2015
/08/30/the-bayesian-reproducibility-project/.
49. The Registered Reports funding scheme was proposed in 2016 by Marcus
Munafò and myself and is currently under consideration by a number of
journals and funders.
50. Kevin M. Williams, Craig Nathanson, and Delroy L. Paulhus, “Identifying
and profiling scholastic cheaters: Their personality, cognitive ability, and
motivation,” Journal of Experimental Psychology: Applied 16, no. 3 (2010): 293,
http://dx.doi.org/10.1037/a0020773. The article can be freely read at https://
www.apa.org/pubs/journals/releases/xap163293.pdf.
51. Ibid., 305.
52. “Diederik Stapel settles with Dutch prosecutors, won’t face jail time,” http://
retractionwatch.com/2013/06/28/diederik-stapel-settles-with-dutch
-prosectors-wont-face-jail-time/.
53. “Former University of Queensland professor Bruce Murdoch charged over
alleged fake Parkinson’s research,” http://www.abc.net.au/news/2014–12–12
/university-of-queensland-professor-on-fraud-charges/5964476.
54. Dea Clark, “Former UQ academic Dr Caroline Barwood granted bail over
fraud charges,” http://www.abc.net.au/news/2014–11–06/former-uq
-academic-granted-bail-over-fraud-charges/5871436.
55. There will always be some disadvantages for those who are close to the suspected
fraudster, such as the retraction of coauthored papers. It would be
unrealistic to believe that we can ever make the act of whistle-blowing completely
harmless for the whistle-blower, but we can certainly do more to
minimize the damage.
56. For further discussion, see Chris Chambers and Petroc Sumner, “Replication
is the only solution to scientific fraud,” http://www.theguardian.com/
commentisfree/2012/sep/14/solution-scientific-fraud-replication.
57. For a list of RoMEO green journals, see http://www.sherpa.ac.uk/romeo/
browse.php?colour=green.
58. Björn Brembs, Katherine Button, and Marcus Munafò, “Deep impact: Unintended
consequences of journal rank,” Frontiers in Human Neuroscience 7
(2013): 291, http://dx.doi.org/10.3389/fnhum.2013.00291.
59. Sam Schwarzkopf, “Revolutionise the publication process,” http://
neuroneurotic.net/2015/07/17/revolutionise-the-publication-process/; Dorothy
Bishop, “Will traditional science journal disappear?,” http://www
.theguardian.com/science/head-quarters/2015/may/12/will-traditional
-science-journals-disappear. Another related idea is the Episciences project,
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Notes to Chapter 8 | 261
which preserves the journal model but seeks to conduct peer review and
publishing entirely with so-called overlay journals that operate alongside
open archives such as arXiv. See http://www.episciences.org/.
60. There are already gentle moves in this direction. Some journals now require
authors to stipulate their contribution to various aspects of a paper. For instance,
PNAS now requires authors to state their contribution to research design,
performing the research, contribution of new reagents/analytic tools,
data analysis, and writing of the paper.
61. Although an alphabetic model is preferable, in the transition between a contributorship
model and the traditional authorship model, the first author position
could be retained, with all subsequent authors listed alphabetically
and contributorship details noted for all authors.
For an insightful discussion of contributorship models of attribution, see
Amy Brand, Liz Allen, Micah Altman, Marjorie Hlava, and Jo Scott, “Beyond
authorship: Attribution, contribution, collaboration, and credit,” Learned
Publishing 28, no. 2 (2015): 151–55, which can be freely downloaded from
http://openscholar.mit.edu/sites/default/files/dept/files/lpub28–2_151–155
.pdf.
62. The danger of false precision could be addressed by replacing point value
percent contributions with either a percent range or a descriptive scale. Various
methods could be trialed.
63. https://twitter.com/david_colquhoun/status/664949963890278400.
64. Joseph P. Simmons, Leif D. Nelson, and Uri Simonsohn, “A 21 word solution,”
available at SSRN 2160588 (2012), http://papers.ssrn.com/sol3/papers
.cfm?abstract_id=2160588.
65. See http://opennessinitiative.org/.
66. See www.sherpa.ac.uk/romeo/.
67. In addition to the data partner scheme proposed by Richard and Candice
Morey in chapter 4 (see http://bayesfactor.blogspot.no/2015/11/habits-and
-open-data-helping-students.html), Kia Nobre’s lab at the University of Oxford
has recently proposed a similar “frenemy system,” https://sites.google
.com/site/todorovicana/musings/crisis_report.
68. http://www.ascb.org/dora/.
69. For details about the TOP guidelines, see https://cos.io/top/.
70. For details about Registered Reports, see https://cos.io/rr/.
71. For details about the Badges initiative, see https://osf.io/tvyxz/wiki/home/.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
INDEX
Page numbers followed by f indicate a figure.
α (significance threshold, false positive),
24, 55–56
Academy of Medical Sciences, 196
adversarial collaborations, 73, 261n67
Allen, John, 163
alternative hypothesis (H1), 9–10
altmetrics, 254n37
American Economic Association registry,
258n34
American Psychological Association
(APA): authorship guidelines of, 165;
data sharing requirements of, 78–80,
87–88; fee-based gold open access
model of, 129
analysis plan preregistration, 203
analytic methods transparency, 203
Archives of Scientific Psychology: article
processing charges (APCs) of, 128,
243n4; data sharing requirements of,
87
article processing charges (APCs), 128–
29, 133–36, 148, 243nn3–4
arXiv.org, 129–30, 244n8
Association for Psychological Science,
160, 215–17
Association of Universities in the Netherlands
(VSNU), 142
Attention, Perception, and Psychophysics,
184
authorship and attribution systems, 163–
69, 211–12, 253n29
availability heuristic, 7
β (probability of failing to reject a false
null hypothesis), 56–60
Bacon, Francis, 1, 4
Bakker, Marjan, 77–78
Bargh, John: approach to sharing data of,
80–81; on open access publishing,
144–45; priming task study of, 13–15,
34–35, 62; warm bath study of, 66
barrier-based publishing, 126–32, 213; citation
rates in, 147; HARKing practices
and, 130; hierarchy of journals
in, 130–32; journal subscription fees
in, 135, 142, 247nn37–38; judgments
of research quality in, 130, 144–46;
solutions for, 208–10. See also open
access (OA) publishing
Barwood, Caroline, 207
Bayesian hypothesis testing, 21, 44, 68–
73, 198, 213; advantages of, 72–73,
233n53; calculation of a Bayes factor
(B) in, 71–72; determining posterior
probability with, 64, 69–71, 232n52
Bayes theorem, 69–71
bean counting. See metric culture of
psychology
Beck, Mihály, 152
Behavioral Insights Team, 140
Bem, Daryl, 1–4, 19, 47, 195
Benveniste, Jacques, 227n13
Bergstrom, Carl, 161
Bergstrom, Theodore, et al., 247n38
Bezeau, Scott, 56
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
264 | Index
bias, 1–21; in Bem’s article on precognition,
1–4; conceptual replication and,
12–16, 20, 34, 49; in hindsight, 16–
21, 25, 35–38, 174, 222n32; neophilia
and positive results and, 8–12, 47–48,
51, 174–75; in selective debugging,
39–40, 226n21; solutions for, 20–21,
174–96; unreviewed preregistration
and, 197–98, 259n41. See also confirmation
bias; publication bias
bibliometrics, 152
bioRxiv, 244n8
Bishop, Dorothy: on data sharing and preregistration,
93–94; on grant funding,
162, 252n21; peer review and prepublication
proposal of, 209–10; on
p-hacking, 26; on publication rates,
166–67; Registered Reports and,
257n25
BMC Psychology: article processing
charges (APCs) of, 128, 134, 243n4;
replication policies of, 68
Bohannon, John, 145–46
Braet, Wouter, 112–16, 242n41, 242n43
Brain, 11
Braver, Todd, 114–17, 242n46
Brembs, Björn, 155, 156f, 159, 208–10
Brown, Derren, 23
Brown, Nick, 237n1
Budapest Open Access Initiative, 128–29,
134
Butterworth, Jon, 53
Button, Kate, 56, 155, 156f, 159, 208–10
Cameron, William Bruce, 149
careers in psychology research, ix–xi,
200; calculation of individual worth
in, 150–51, 156–57, 161–63, 168–70,
212, 214–15, 254n37; grant funding
and, 160–63, 169, 252nn21–24; publishing
and, 131–32, 138–39, 163–68,
251n12, 253n29; replication and, 74.
See also metric culture of psychology
cargo cult science, 46, 50, 173–74
CARKing (critiquing after results are
known), 256–57n20
Carp, Josh, 33
Caruso, Eugene, 51
CC-BY. See Creative Commons Attribution
(CC-BY) license
Center for Open Science, 74, 87, 115, 196,
201. See also Registered Reports
Cerebral Cortex, 11
Cesario, Joe, 81
“Citation Indexes for Science” (Garfield),
151–53
citation metrics, 147; JIFs and, 151–60,
168–69, 250–51nn5–7, 254n37;
Transparency and Openness Promotion
(TOP) guidelines and, 203, 214
Cleary, T. Anne, 178–79
Cognition, 87
Cognitive Psychology, 130–31
Cohen, Jacob, 56–57
Collabra, 148
Colquhoun, David, 212, 253n25
Committee on Publication Ethics, 163–
64
compare project, 197
conceptual replication, 12–16, 20, 34, 49,
174
confidentiality, 85, 236n27
confirmation bias, 4–21, 174, 213; availability
heuristic in, 7; conceptual replication
and, 12–16, 20, 34, 49; in
hindsight, 16–21, 25, 35–38, 174,
222n32; measurement of, 5–7; neophilia
and positive results and, 8–12,
47–48, 51, 174–75; peer-reviewed
study preregistration and, 174–96;
positive-test strategy in, 8; protections
against, 20–21. See also hidden
analytic flexibility
contrapositives, 6
contributorship model, 211–12,
261nn60–62
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Index | 265
corruptibility. See fraudulent research
practices
Cortex, 11; implementation of Registered
Reports by, 178–81, 182f, 255n11;
open access model of, 129
Creative Commons Attribution (CC-BY)
license, 128, 132, 135, 141, 209
Crockett, Molly, 192
Curate Science project, 259n46
Current Biology, 130–31, 155
Curry, Stephen, 95, 159
Czerski, Helen, 53
Dante, 4
data fishing, 85
data hoarding, 79–83, 90–91, 127; data
ownership defenses of, 76–77; hidden
misconduct and, 77, 81–83; imposed
secrecy and, 80–81; solutions for,
202–4
data partnerships, 94–95, 214, 261n67
data review, 209
data sharing, 43, 74, 76–95, 115, 202–4,
214; benefits of, 77–78, 235n7; coauthorship
debates and, 87, 236n32;
confidentiality considerations in, 85,
236n27; data partner plans and, 94–
95, 214, 261n67; free archive services
for, 85, 236n28; frenemy system
of, 261n67; funding challenges
in, 85; p-hacking and, 94; Peer Reviewers’
Openness (PRO) initiative
and, 88–89, 91, 202, 213, 214; public
archives for, 79, 214; publishing
requirements on, 84–88, 91,
237n36; Transparency and Openness
Promotion (TOP) guidelines on,
202–4, 214; by Verbruggen’s laboratory,
92–94
data transparency, 203
Dataverse, 85, 236n28
de Groot, Adriaan, 222n32
Dekker, Sander, 142
Della Sala, Sergio, 178–79
Denzler, Markus, 109f, 111–12, 240n33
de Palma, Adriana, 140
design and analysis transparency, 203
deterministic findings, 12, 16–19
Directory of Open Access Journals
(DOAJ), 145, 244n6
direct replication, 48–55, 67–68, 213,
227n7, 228nn13–14; in the physical
sciences, 50, 53–54, 66, 216; publishing
constraints on, 54–55, 177, 200–
201; rates of, 50; statistical power
and, 57–58, 229n29
disclosure statements, 42–43, 61–62
Donnellan, Brent, et al., 66, 80–81
DORA (San Francisco Declaration of Research
Assessment), 158–60,
214–15
Doyen, Stéphane, 14, 62
Dressler, Marc, 112
DrugMonkey, 84–85
Dutch National Board of Research Integrity
(LOWI), 110–12, 240n33
Dutilh, Gilles, 235n7
Easterbrook, Gregg, 22
Economic and Social Research Council
(ESRC), 88, 237n36, 252n22
Eigenfactor, 254n37
Eisen, Michael, 146, 249n64
Elbakyan, Alexandra, 137–38, 246n29
elderly priming effect, 13–15, 34–35, 62
Elsevier: access policies of, 129, 138, 141–
42, 246n30, 248n49; boycotts of, 142;
bundled journals of, 247n38; profit
margin of, 246n32
Episciences project, 260n59
Etz, Alex, 259n48
European Research Council, 136, 245n26
Europe Pubmed Central, 132–33
Experimental Psychology, 87
Exploratory Reports, 256n17
exploratory research, 185–86, 188
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
266 | Index
fabrication of results. See fraudulent research
practices
false negatives (β), 56–60
false positives (α): disconfirmation of,
56–57; nondisclosure of methodological
details and, 61–62; p-hacking and,
26, 33, 55–56; positive predictive
value (PPV) and, 58–60
falsified data. See fraudulent research
practices
Fanelli, Daniele, 11, 105
“Feeling the Future: Experimental Evidence
for Anomalous Retroactive Influences
on Cognition and Affect”
(Bem), 1–4
Ferguson, Chris, 51, 100
Fetterman, Adam, 228n14
Feynman, Richard, 46, 50, 173–74
Fiedler, Klaus, 56–57
Figshare, 85, 236n28, 258n34
file drawer effect. See publication bias
Finch, Janet, 133–36
first authors, 164, 166, 253n29
Fisher, Ronald, 24
fixes. See solutions
Förster, Jens, 108–12, 205, 240–41nn32–
35, 241n37
Forstmann, Birte, 65, 68
Franco, Annie, et al., 197–98
fraudulent research practices, 45, 96–125,
251n12; criminalization of, 206–7;
fabrication of data in, 77, 81–83; by
junior researchers, 112–16, 242n41,
242n43, 242n46; laboratory responses
to, 114–15; publication bias
and, 106, 108, 111–12, 251n12; rates
of, 105; replication and, 208; retraction
of papers for, 103, 153, 239n16,
260n55; slippery slope of, 105–12,
240n29, 240–41n33–35; solutions
for, 205–8; Stapel’s use of, 96, 99–
106, 207, 237–38nn1–2, 238n12,
239n19, 242n38; tips for success
with, 122–25; whistle-blowers on,
117–21, 207, 243nn52–53, 260n55
French, Chris, 3
frenemy system, 261n67
Frith, Uta, 168
Frontiers, 146
Fujii, Yoshitaka, 239n16
full open access (OA) model, 128. See also
open access (OA) publishing
Garfield, Eugene, 151–53, 155
Geek Manifesto (Henderson), 7
Gelman, Andrew, 35, 38
Gernsbacher, Morton Ann, 257n25
Gervais, Will, 67–68
gift authorship, 165
Gigerenzer, Gerd, 56, 232n52
Gilbert, Dan, 52, 227n13
Goldacre, Ben, 197
gold open access (OA) model, 128–29,
133–34, 148, 243n5, 245n18. See also
open access (OA) publishing
Goodhart’s law of economics, 150, 250n1
G* Power, 230n30
grant capture, 160–63, 169, 252nn21–24
Graves, Roger, 56
green open access (OA) model, 129, 133–
36, 208, 213, 243n5. See also open access
(OA) publishing
Grieneisen, Michael, 66
groupthink (herd behavior), 21
guerrilla open access, 136–38,
246nn28–30
Ha, Young-Won, 8
Hajje, Chawki, 147
Haldane, J. B. S., 83
Hale, Greg, 94
Hardy, John, 257n25
HARKing (Hypothesizing After Results
are Known), 17f, 18–21, 25, 174, 213,
222n32, 230n34; as exploratory, 183–
84; moral arguments on, 45; publish-
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Index | 267
ing access barriers and, 130; solutions
for, 41–44; unconscious introduction
of, 35–39; in unreviewed preregistered
studies, 198
Harnad, Stevan, 134–36, 147, 245n22
Hartgerink, Chris, 141–42, 248n49
Hartshorne, Josh, 259n43
Heene, Moritz, 51, 100
Hegarty, Boy, 50
Henderson, Mark, 7
hidden analytic flexibility, 15, 22–45, 174;
fallacy of incomplete evidence in, 23;
peculiar patterns in reporting of p values
in, 29–33, 224n9; peer-reviewed
study preregistration and, 174–96;
phantom replication and, 34–35,
224n11; researcher degrees of freedom
(p-hacking) in, 17f, 24–29, 44–
45, 55–56, 251n12; selective debugging
in, 39–40, 226n21; solutions for,
41–45; unconscious analytic tuning
in, 35–39. See also fraudulent research
practices; statistical methods
Higher Education Funding Council for
England (HEFCE), 136, 140–41
Hill, Steven, 140
h-index, 254n37
hindsight bias, 16–21, 25, 174, 213,
222n32; solutions for, 41–44; unconscious
introduction of, 35–39
hybrid open access (OA) model, 128–29,
132–36, 243n5, 245n26. See also open
access (OA) publishing
hypothetico-deductive (H-D) model of
scientific method, 16–19, 230n34; advancement
of theory with, 150–51,
168, 173, 254n37; direct replication
in, 48–55, 67–68, 213, 227n7,
228nn13–14; empirical evidence in,
48; generation of scientific laws in,
49f; Registered Reports and, 191–92;
study preregistration in, 20–21, 41–
42, 74, 174–96
#icanhazpdf, 137–38, 246n28
impact factors. See journal impact factors
(JIFs)
Institute for Scientific Information
(Thompson Reuters), 152, 155
International Committee of Medical Journal
Editors (ICMJE), 165
International Journal of Radiation Oncology,
Biology, Physics, 195
International Social Cognition Network
(ISCON), 89
internment. See barrier-based publishing
Ioannidis, John, 50, 227n7, 257n25
irreproducibility. See nonreplication
JIFs. See journal impact factors (JIFs)
John, Leslie, et al.: on falsification of data,
105; on nondisclosure of methodological
details, 61; on p-hacking and
HARKing rates, 19, 27, 35, 43–44,
107–8, 186
Johnson, Ben, 140–41
Johnson, John, 40
journal impact factors (JIFs), 151–60, 169,
214, 254n37; DORA renunciation of,
158–60, 214–15; purpose of, 152–53;
Registered Reports and, 184; Reproducibility
Index and, 199–200,
259n46; statistical vulnerability of,
153–55, 159, 250–51nn6–7
journalism practices, 215
Journal of Cognitive Neuroscience, 177
Journal of Experimental Psychology group,
130–31
Journal of Neuroscience, 128–31
Journal of Personality and Social Psychology,
1–4, 47
journals. See open access (OA) publishing;
publication bias; publishing
JSTOR, 138
Kahneman, Daniel, 52–54, 228n14
Kerr, Norbert, 18–19, 186, 222n32
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
268 | Index
Kiley, Robert, 133
Klayman, Joshua, 8
Krueger, Joachim, 3
Kuhn, Thomas, 186
Kuszewski, Andrea, 137
Lakens, Daniël, 44, 224n9, 256–57n20
Lalande, Daniel, 30–31
last authors, 164, 166, 253n29
Lauwereyns, Jan, 145
law of small numbers, 83
Laws, Keith, 68
lead authors, 164
LeBel, Etienne, et al., 61, 259n46
Leggett, Nathan, et al., 31
likelihood principle, 44
Loken, Eric, 35, 38
low statistical power. See statistical power
Luck, Steven, 257n25
Machiavelli, Niccolò, 193
Mack, Katie, 53–54
MacManes, Matthew, 84–85
Makel, Matthew, 50
malpractice, x–xi
Marcus, Adam, 74, 199
Masicampo, E. J., 30–31
Maxwell, Scott, 60
Mercier, Hugo, 8
Mertens, Myriam, 92
meta-analysis, 78, 235n7
methodological details, 48, 52–53, 61–62,
213
metric culture of psychology, 150–70,
194, 254n37; authorship and attribution
in, 163–69, 211–12, 253n29; calculating
individual worth in, 150–51,
156–57, 161–63, 168–70, 212, 214–
15, 251n12, 254n37; citations and JIFs
in, 151–60, 168–69, 250–51nn5–7;
DORA renunciation of, 158–60, 214–
15; vs. goal of advancement of theory,
150–51, 168, 254n37; grant capture
and, 160–63, 169, 252nn21–24; minimum
funding targets and, 163; solutions
for, 210–12. See also careers in
psychology research
middle authors, 164–65
Miller, Earl, 76
minimal publishable units, 167–68
Missirlis, Fanis, 163
Mitchell, Jason, 54
Morey, Candice, 94–95
Morey, Richard, 88–89, 94–95
Mounce, Ross, 247n38
multiple statistical hypotheses, 38
multisite collaborations, 256n14
Munafò, Marcus, 155, 156f, 159, 184,
208–10
Murdoch, Bruce, 207
National Institutes of Health: data sharing
rules of, 76; open access (OA) policies
of, 136, 244n12
National Science Foundation, 198
Nature, 11; access limitations of, 130–31;
JIF of, 152
Nature Human Behaviour, 196
Nature Neuroscience: access limitations of,
130–31; methods checklist of, 73–74
Nelson, Leif: on methodological disclosure,
42–43, 61, 213; on p-hacking,
25–27, 31, 32f; on preregistration,
197
neophilia, 8–12, 47–48, 51, 174–75
Netherlands Organisation for scientific
Research (NWO), 136
Neuron, 130–31
Neuropsychologia, 177–78
Neuroskeptic, 31, 177–80
Neyman-Pearson hypothesis test, 10
NHST. See null hypothesis significance
testing
Nickerson, Raymond, 6–7
Nietzsche, Friedrich, 20
Nieuwenhuis, Sander, 65
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Index | 269
Nobre, Kia, 261n67
nonreplication, 198, 208; impact of,
228n14; rates of, 50; Repligate incident
of, 51–53; reproducibility of, 13–
14. See also unreliability
Nosek, Brian, 74, 87, 201, 216,
256–57n20
novelty. See neophilia
null hypothesis (HO), 9–10, 71. See also
Bayesian hypothesis testing
null hypothesis significance testing
(NHST), 10, 21, 24, 55–56, 68–69;
common misunderstandings of, 48,
63–65, 68–69; computing technology
for, 31; probability of failing to reject
a false null hypothesis (β) in, 56–
60; significance threshold (α) in, 24,
55–56; stopping rules in, 27–29, 43–
44, 72. See also p values
Nussbaum, Dave, 45
Ontsporing (Stapel), 237n1
Op de Beeck, Hans, 113–17, 242n43
open access (OA) publishing, xi, 128–48,
208, 213–15; article processing
charges (APCs) in, 128–29, 133–36,
148, 243nn3–4; citation rates in, 147;
counterarguments against, 138–47,
247nn37–38; directory of journals
using, 145, 244n6; Finch Report on,
133–36; full model of, 128; gold
model of, 128–29, 133–34, 148,
243n5, 245n18; green model of, 129,
133–36, 244n5; guerrilla movements
for, 136–38, 246nn28–30; hybrid
models of, 128–29, 132–36, 243n5,
245n26; incentives for, 132–36; journal
prestige and, 130–31; peer review
policies in, 144–46; public archives
for, 129–30, 209–10, 244n8; public
policy makers and, 139–41, 248n42
Open Biology JIF, 153, 154f
openness. See transparency
Open Science Framework, 115, 193, 213,
258n34
Open Scoop Challenge, 256n18
Oransky, Ivan, 74, 199
outgroup homogeneity, 143
p-curve analysis, 31–33, 42, 224n11
p-hacking (researcher degrees of freedom),
17f, 24–35, 55–56, 174,
251n12; addition of participants in,
26–29, 72; confirmatory vs. exploratory
decisions in, 25; data sharing
and, 94; detection of, 29–33, 42,
224n11; exclusion of statistical outliers
in, 25; moral arguments on, 45;
one-time prevalence estimate of, 35,
224n18; phantom replication and,
34–35, 224n11; rates of, 27–29, 35,
224n18; solutions for, 44–45; in unreviewed
preregistered studies,
197–98
p values, 24–26; Bayesian alternative to,
68–73, 198, 213, 232–33nn52–53;
common misunderstandings of, 63–
65, 68–69; misreporting of, 79;
p-curve analysis of, 31–33, 42,
224n11; peculiar patterns in reporting
of, 29–33, 224n9; researcher degrees
of freedom and, 25–26; significance
threshold (α) in, 24, 55
Parliamentary Office of Science and Technology
(POST), 140, 248n42
Pashler, Hal, 13, 255n12
paywalls. See barrier-based publishing
PeerJ, 128, 131, 148, 243n4
peer review, 8–10, 177; burdens of finding
reviewers for, 89; CARKing (critiquing
after results are known) in, 256–
57n20; data sharing and, 88–89, 91,
202; preregistration without, 196–98;
Registered Reports system of, 178–96,
214–17; Schwarzkopf and Bishop’s
proposals for, 209–10
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
270 | Index
peer-reviewed study preregistration. See
Registered Reports
Peer Reviewers’ Openness (PRO) initiative,
88–89, 91, 202, 213, 214
personality profiling, 205–6
Perspectives on Psychological Science, 68,
184, 256n14
phantom replication, 34–35, 224n11
PLOS journals: data-sharing enforcement
challenges of, 86–87; data sharing requirements
of, 84–86, 91; openaccess
policies of, 128, 131–34
PLOS Medicine, 155
PLOS ONE, 14, 54; article processing
charges (APCs) of, 243n4, 245n18;
data sharing compliance in, 86–87;
open access model of, 128, 131; peer
review standards of, 144–46
Plucker, Jonathan, 50
Popper, Karl, 186
positive predictive value (PPV), 58–60
positive-test strategy, 8
post-hoc hypothesizing. See HARKing
potential solutions. See solutions
Pottery Barn rule, 68, 200–201
precognition (psi), 1–4, 27, 195
preregistration, 20–21, 41–42, 74, 174–
96; Bishop’s peer review proposal
and, 209–10; online registries for,
197; Registered Reports initiative of,
174–96, 214–17; Rosenthal’s proposal
for, 178–79, 255n7; scooping and,
188–89, 256nn18–19; Transparency
and Openness Promotion (TOP)
guidelines for, 203, 214; unreviewed
options for, 196–98, 259n41
priming tasks, 1; Bargh’s study of, 13–15,
34–35, 62; replication studies of,
51–52
principal investigators, 164–65
probabilistic findings, 12
probability of failing to reject a false null
hypothesis (β), 56–60
Proceedings of the National Academy of Sciences
(PNAS), 128–31, 261n61
professional culture of psychology. See careers
in psychology research
profiles of researchers, 205–6
psyarxiv, 244n8
PsychFileDrawer, 13
Psychological Science, 11; access limitations
of, 130–31; data sharing badges
of, 87; on JIF, 160; methodological
disclosure statement of, 74
Psychonomic Society, 129
publication bias, 3–4, 11–14, 17f, 174,
220–21n13, 223n35; effect size and,
60, 230n31; fraudulent practices and,
106, 108, 111–12, 251n12; journal
prestige and, 131; neophilia and positive
results in, 8–12, 48, 51, 174–75;
preregistration and, 41–42, 178, 194–
95; vs. researcher bias, 30–31, 33, 99–
100, 224n9
publishing, 214–15; access restrictions
in, 126–32; author fees in, 135,
243n3, 245n22; authorship and attribution
system in, 163–69, 211–12,
253n29; citation rates in, 147, 151–
60, 168–69, 250–51nn5–7; contributorship
model in, 211–12, 261nn60–
62; data sharing requirements in, 76,
78–79, 84–88, 91, 237n36; of direct
replication studies, 54–55, 68, 74,
177, 200–201; disclosure of methodological
details in, 48, 52–53, 61–62,
73–74; impact on careers of, 131–32,
138–39; journal impact factors (JIFs)
in, 152–60, 163, 166, 169, 250–
51nn6–7; journal prestige and, 130–
32, 144–46, 150, 152; journal subscription
fees in, 135, 142,
247nn37–38; minimal publishable
units in, 167–68; obsolescence of
journals in, 208; in online journals,
62; open access models of, 128–48;
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Index | 271
peer review and prepublication proposals
for, 208–10; Peer Reviewers’
Openness (PRO) initiative and, 88–
89, 91, 202, 213, 214; peer-review
process of, 8–10, 89, 144–46; of preprints,
129–30; retraction of papers
in, 48, 65–67, 103, 153, 231n40,
231n42, 239n16; Transparency and
Openness Promotion (TOP) guidelines
in, 202–4, 214. See also preregistration;
Registered Reports
qualitative psychological research, 254n1
quality of research: advancement of theory
and, 150–51, 168, 173, 254n37; in
evaluating use of grant funds, 161–63,
252n22; journal prestige and, 130–31,
144–46; Registered Reports model
and, 178–96, 214–17
random data audits, 205
ranking system of author attribution,
165–66, 211–12
rating-only reviews, 209
reforms. See solutions
Registered Replication Reports, 68, 194–
95, 209
Registered Reports, 174–96, 213, 214–17;
analysis of preexisting data and, 194–
95; Cortex’s policies on, 178–81, 182f,
255n11; critical reaction to, 184–93,
256–57nn16–21, 257nn25–26; efficacy
of, 195; implementation of, 195–
96; in-principle acceptance (IPA) in,
183; publication bias and, 194–95;
rates of publishing and, 194; replication
initiative of, 68, 194–95, 198,
201–2, 209, 256n14, 259n49; shortterm
projects and, 193, 196–97; signatories
of, 184, 193, 257n25; Stage 1
criteria of, 180–83, 255n12; Stage 2
criteria of, 182f, 183–84; universal
goal of, 194
Registered Reports Committee, 196
reliability. See replication; unreliability
replication, x, 3–4, 47–48, 190, 198–202,
213–14, 259nn48–49; of Bem’s findings
on precognition, 3, 47; conceptual
approaches to, 12–16, 20, 34, 49,
174; definitions of, 14–15; detecting
fraud with, 208; direct methods of,
48–55, 67–68, 200–201, 213, 227n7,
228nn13–14; in discovery, 12;
hypothetico-deductive (H-D) model
and, 17f; JIF ratings and, 155; multicenter
initiatives in, 201; nondisclosure
of methodological details and,
61–62, 73–74, 230n34; of nonreplications,
13–14; phantom replication
and p-hacking in, 34–35, 55–56,
225n16; proposed reforms of, 67–68,
74, 198–202; rates of, 50; Registered
Reports and, 68, 194–95, 209,
256n14; retraction and, 65–67,
231n40, 231n42; Transparency and
Openness Promotion (TOP) guidelines
on, 204, 214. See also
unreliability
“The Replication Revolution: One Year
On” symposium, 215–16
Replication Tracker, 259n43
Repligate, 51–53, 201
Reproducibility Index, 199–200, 259n46
Research 4Life program, 247n37
Research Councils United Kingdom
(RCUK), 136
researcher degrees of freedom. See
p-hacking
researcher profiles, 205–6
Research Excellence Framework (REF),
x–xi
Research Integrity Office, 205
research materials transparency, 203
Responsible Research Publication: International
Standards for Authors, 163–
64
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
272 | Index
retractions, 48, 65–67, 231n40; for fraud,
103, 153, 239n16, 260n55; JIF rates
and, 153; rates of, 66, 231n42. See also
fraudulent research practices
reversed priming task, 1
Rights MEtadata for Open archiving
(RoMEO) project, 208, 213
Ritchie, Stuart, 3
Roediger, Henry L, III, 160
Rosenthal, Robert, 178–79, 220–21n13,
255n7
Royal Society Open Science, 196, 258n33
Rutherford, Ernest, 41
Sagan, Carl, 171
San Francisco Declaration on Research
Assessment (DORA), 158–60, 214–15
Sanna, Lawrence, 82–83
Sassenberg, Kai, 228n14
Savine, Adam, 114–16, 242n46
Schachner, Adena, 259n43
Schnall, Simone, 51–53, 227n13
Schwarzkopf, Sam, 209–10
Science: access limitations of, 130–31;
data sharing rules of, 76; Transparency
and Openness Promotion (TOP)
guidelines of, 87–88
Science Citation Index, 152
scientific method, ix–x, 171–73; advancement
of theory in, 150–51, 168, 173,
254n37; direct replication in, 49–55,
67–68, 227n7, 228nn13–14; generation
of scientific laws in, 49f;
hypothetico-deductive (H-D) model
of, 16–19, 230n34; Registered Reports
and, 191–92
Scientometrics, 152
Sci-Hub, 137–38, 246n29
Scott, Sophie, 256n16
second authors, 164, 253n29
secrecy. See data hoarding
Sedlmeier, Peter, 56
Selection Task, 5–6
selective debugging (scrutiny), 39–40,
226n21
senior authors, 164–65, 166, 253n29
Shalev, Idit, 66, 80–81
Simmons, Joe: on methodological disclosure,
42–43, 61, 213; on p-hacking,
25–27, 31, 32f; unreviewed preregistration
tool of, 259n41
Simons, Dan, 13; on failures of direct replication,
55; on failures to share data,
81; on retraction, 66
Simonsohn, Uri: fraud detection tool of,
77, 81–83, 107, 122, 205, 239n26; on
methodological disclosure, 42–43, 61,
213; on p-hacking, 25–27, 31, 32f, 33;
unreviewed preregistration tool of,
259n41
slow science, 167–68
Smeesters, Dirk, 83, 106–7, 112, 239n26,
240nn28–29
Smith, Rachel, 60
social priming. See priming tasks
Social Psychology, 51–53, 201
Social Science registry, 197–98
Society for Neuroscience, 116, 243n50
solutions, 171–217; for barrier-based publishing
(internment), 208–10; concrete
steps in, 213–15; for data hoarding,
202–4; for fraudulent research
practices, 114–15, 205–8; for hidden
analytic flexibility, 41–45; for the
metric culture of psychology (bean
counting), 210–12; for publication
bias, 41–42; Registered Reports as,
174–96, 214–17; for unreliability, 67–
74, 198–202; unreviewed preregistration
and, 196–98, 259n41. See also
data sharing; open access (OA) publishing;
transparency
SPARC Europe, 147
specialist statistical review, 209
Spellman, Barbara, 216, 257n25
Sperber, Dan, 8
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
Index | 273
Srivastava, Sanjay, 68, 200–201, 224n11,
225n16
standardized research practices, 44–45
Stapel, Diederik, 96, 99–106, 112, 117,
207, 237–38nn1–2, 238n12, 239n19,
242n38
statistical methods, xi, 12; Bayesian hypothesis
testing in, 21, 44, 68–73,
198, 213, 232–33nn52–53; common
misunderstandings of, 48, 63–65,
68–69; computing technology for, 31;
hidden analytic flexibility and, 22–45;
likelihood principle in, 44; probability
of failing to reject a false null hypothesis
(β), 56–60; significance threshold
(α) in, 24, 55–56; standardized
practices of, 44–45. See also null hypothesis
significance testing
statistical power, 48, 55–60, 198, 213;
calculation of (1–β), 56; complexity
of experimental design and, 60,
230n31; direct replication and, 57–
58, 67–68, 229n29; false negatives
and, 56–57; G* Power software for,
230n30; of the hypothetico-deductive
(H-D) model, 17f; JIFs and, 155, 156f;
positive predictive value (PPV) and,
58–60; sample size and, 57,
229–30nn29–31
Sterling, Theodore D., 220–21n13
Stokes, Mark, 39, 235n7
stopping rules, 27–29, 43–44, 72
Strack, Fritz, 54–55, 223n35
Stroebe, Wolfgang, 54–55
Strube, Michael, 44
subscription journals. See barrier-based
publishing
Swartz, Aaron, 138
The System (TV program), 23
Taylor, Mike, 126
test-retest reliability, 33. See also
unreliability
Thucydides, 4
Times Higher Education, 162
Tolstoy, Leo, 4
transparency, x, 27; Center for Open Science
initiatives on, 74, 87, 115, 196,
201; Creative Commons Attribution
(CC-BY) license and, 128, 132, 135,
141, 209; data sharing initiatives and,
76–95, 115; disclosure statements
and, 42–43; in laboratory data archives,
114–15; methodological disclosure
and, 48, 52–53, 61–62, 73–74,
213; open access (OA) publishing
and, xi, 128–48, 208, 213–15; Peer
Reviewers’ Openness (PRO) initiative
and, 88–89, 91, 202, 213, 214; public
funding obligations of, 76, 78, 85, 88,
127, 208, 237n36; Registered Reports
initiative and, 174–96, 214–17; Royal
Society initiative on, 196; study preregistration
and, 20–21, 41–42, 74,
177–78, 184, 195
Transparency and Openness Promotion
(TOP) guidelines, 196, 202–4, 214
21 word solution, 42–43, 61, 213
undergraduate research projects, 193,
196–97
unreliability, 33, 46–74, 259n48; adversarial
collaborations and, 73, 261n67;
low statistical power and, 48, 55–60,
67–68, 213; nondisclosure of methodological
details and, 48, 52–53,
61–62, 73–74, 230n34; opposition to
direct replication and, 3, 48–55, 67–
68, 213, 227n7, 228nn13–14; retraction
failures and, 48, 65–67, 231n40,
231n42; solutions for, 67–74, 198–
202; statistical fallacies in, 48, 63–
65
unreviewed preregistration, 196–98,
259n41
US Office of Research Integrity, 116, 205
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.
274 | Index
Vandekerckhove, Joachim, 259n48
variable stopping rule, 44
Verbruggen, Frederick, 92–94, 242n43
Wagenmakers, Eric-Jan, 4, 65, 68, 216
Walster, G. William, 178–79
Wason, Peter, 5–6
Wellcome Trust, 132–33
whistle-blowers, 117–21, 207, 243nn52–
53, 260n55
whoneedsaccess.org, 143–44
Wicherts, Jelte, 92, 100, 216
Wicherts, Jelte, et al., 77–80
Wilde, Oscar, 151
Williams, Kevin, et al., 205–6
Wilson, Andrew, 52–53
witchcraft trials, 6–7
Wolfe, Jeremy, 257n25
Yong, Ed, 14
Zenodo, 85, 236n28
Zhang, Minhua, 66
Zwann, Rolf, 15, 107–8, 239n26
Copyright©2017.PrincetonUniversityPress.Allrightsreserved.