Attentional Capture by an Unannounced Color Singleton Depends on
Expectation Discrepancy
Gernot Horstmann
Bielefeld University
Eight experiments examined the conditions under which a color singleton that is presented for the 1st
time without prior announcement captures attention. The main hypothesis is that an unannounced
singleton captures attention to the extent that it deviates from expectations. This hypothesis was tested
within a visual-search paradigm in which set-size effects were used to infer attentional capture. The
results showed that attentional capture by an unannounced color singleton was due to a mismatch with
expectations concerning the color of the object and not due to its being a singleton. Thus, the results
imply that theories of attention have to consider expectation discrepancy as a determinant of attention
shifts.
Keywords: attention capture, involuntary orienting, visual search, expectation discrepancy, surprise
Expected and unexpected events differ fundamentally in their
nature. Expected events can, in principle, be attended to and
further processed according to their relevance for current goals or
intentions as they occur. The relevance of unexpected events,
however, is unknown before their appearance; yet it has to be
determined to safeguard adaptive behavior. This presupposes that
focused attention is directed to the unexpected event. The present
research is concerned with the conditions of attentional capture by
an unexpected color singleton (Gibson & Jiang, 1998; Horstmann,
2002, 2004, in press; Wilcocks, 1928). That is, it examines the
conditions under which an unexpected salient event quickly sum-
mons attention.
The main hypothesis pursued is that the unannounced--and
hence unexpected--presentation of a new color captures attention
to the extent that it is discrepant from expectations (e.g., Horst-
mann, 2002; Wilcocks, 1928). In the reported experiments, this
hypothesis is tested and refined via a visual-search paradigm (e.g.,
Jonides, 1981; Theeuwes, 1992; Treisman & Gelade, 1980; Yantis
& Egeth, 1999): Participants searched for target letters (e.g., U or
H) within an array of distractor letters and indicated the target
letter's identity with a key press. Reaction time (RT) was the
dependent variable. The number of distractor letters (set size) was
varied to test search efficiency: If RT was longer with 12 letters
than with 4 letters, search was assumed to have proceeded nonef-
ficiently, whereas RTs that were approximately the same with 12
and 4 letters revealed efficient search, indicating attentional cap-
ture to the target's location. To test whether the expectancy-
discrepant color captured attention, I presented it at the position of
the target only; that is, it was a color singleton (e.g., Nothdurft,
2000; Theeuwes, 1992, 1994; Yantis & Egeth, 1999).
Contingent Capture
Color singletons can be used to guide visual search quite effec-
tively (cf. Krummenacher, Mu¨ller, & Heller, 2002; Nothdurft,
2000; Wolfe, 1994; Yantis & Egeth, 1999): When observers know
that the singleton's position predicts the target's position, search
performance (as indicated by RT) is independent of set size,
revealing efficient search. Moreover, the phenomenal experience
is one of immediate awareness of the singleton, as if attention had
been attracted or captured by the color singleton. However, this
does not necessarily mean that attentional capture is completely
independent of intentions. On the contrary, if the singleton's
position is known to be nonpredictive of the target's position,
singletons can often be completely ignored (e.g., Yantis & Egeth,
1999; but see Theeuwes, 1992, 1994; Theeuwes & Burger, 1998;
Theeuwes & Godijn, 2001). This pattern of results indicates that
attentional capture critically depends on an attentional set: It
occurs in the presence of the intention to attend to the singleton but
not in the absence of such an intention and is thus contingent on
the attentional set (or attentional control settings; cf. Folk, Rem-
ington, & Johnston, 1992, 1993). According to the contingent-
capture hypothesis, the process underlying attentional capture is a
hybrid of controlled and automatic processes: The settings of an
attentional control system are determined "offline" (i.e., prior to
the presentation of the stimulus) depending on the respective
intention; thereafter, stimuli that match the attentional set capture
attention without the need of intervening intentional processes.
Several variants of attentional control settings have been de-
scribed. First, attentional control can be set to color, brightness,
size, or motion (e.g., Yantis & Egeth, 1999) if the target is defined
by one of these features. Initially, it was suggested that attentional
This research was partially supported by Deutsche Forschungsgemein-
schaft Grant HO 3248/1 and by University Grant 27102030. I am indebted
to Wilhelm Banmann, Christine Broermann, Sabine Dlugosch, Kathrin
Holtmann, and Lydia Wittenbreder for conducting the experiments; Lily-
Maria Silny for her assistance in manuscript preparation; and Ulrich
Ansorge, Stefanie Becker, Bradley Gibson, Wulf-Uwe Meyer, Ingrid
Scharlau, Daniel Simons, and Jan Theeuwes for their helpful comments on
earlier versions of this article.
Correspondence concerning this article should be addressed to Gernot
Horstmann, Department of Psychology, Bielefeld University, PO Box 100
131, Bielefeld D-33 501, Germany. E-mail: gernot.horstmann@
uni-bielefeld.de
Journal of Experimental Psychology: Copyright 2005 by the American Psychological Association
Human Perception and Performance
2005, Vol. 31, No. 5, 1039­1060
0096-1523/05/$12.00 DOI: 10.1037/0096-1523.31.5.1039
1039
control settings discriminate between static features, such as color
or brightness, and dynamic features, such as motion or flicker, but
not within the respective categories (Folk et al., 1992). Recent
evidence, however, suggests that the probability of capture de-
pends on physical similarity between stimulus and attentional set
even within dimensions (Ansorge & Heumann, 2003; Folk &
Remington, 1998). Second, apart from this set for target features
there may be a "set for displaywide features": Gibson and Kelsey
(1998) argued that attention should often be set to features that
discriminate the search display from other displays within a given
trial, and they provided evidence supporting this hypothesis. A
third set type was described by Bacon and Egeth (1994). If a
singleton discriminates between targets and distractors, performers
may adopt a singleton-search strategy, which entails searching for
any singleton present in the search display. In summary, atten-
tional control may be set to the target's features, to displaywide
features, or to any singleton present in the display.
Surprise Capture
That attentional capture by color singletons can be contingent on
the intentions of the observer does not imply that this is always the
case. Indeed, there is evidence that a color singleton can capture
attention in the complete absence of any intention to attend to the
singleton--that is, when the singleton is presented for the first time
without prior announcement and following a number of trials
without a singleton (e.g., Horstmann, 2002, 2004, in press).
Clearly, the presentation of a singleton under such circumstances
precludes any intentions toward the singleton (cf. Gibson & Jiang,
1998). Additionally, in Horstmann's (2002) experiments, color
was not a discriminative cue related to any aspect of the task to
prevent a set for color as a displaywide feature (Gibson & Kelsey,
1998): All stimuli preceding the critical trial (including the fixation
cross and the messages that informed the participants about the
beginning and the end of the practice trials) were presented in the
color of the stimuli in the homogeneous trials.
Two strategies have been used to test attentional capture by an
unannounced singleton. In the first paradigm, participants per-
formed a search task on very briefly presented letters, with accu-
racy as the dependent variable (Gibson & Jiang, 1998). For exam-
ple, in Horstmann (2002, Experiment 1), the positions of 12 letters
were preceded by color cues of the same color with a stimulus
onset asynchrony (SOA) of 500 ms, until, in the critical trial, the
cue at the position of the target letter had a new color. In this
critical trial, accuracy was significantly improved, indicating that
the color cue captured attention to its location, thereby facilitating
the identification of the target letter. These results were replicated
and extended (Horstmann, 2004, in press); they also indicated that
the time lag of surprise capture (relative to contingent capture) was
about 300­400 ms (Horstmann, in press) and that both benefits
from predictive singletons and costs from nonpredictive singletons
accrued (Horstmann, 2004).
The second approach was instantiated in Horstmann (2002,
Experiment 3), where the presentation time of 4 or 12 colored
letters was not restricted and RT was the dependent variable. In the
precritical trials, in which all letters were colored equally, RT was
strongly affected by the number of presented letters (set size),
indicating that finding the target required serial search; in contrast,
there was only a weak set-size effect in the critical trial, in which
the target letter appeared in a new color. This pattern of results
would be expected on the hypothesis that the new color captured
spatial attention.
These results are consistent with a hypothesized surprise­
attention interface (Darwin, 1872; Horstmann, 2002; Meyer, 1988;
Meyer, Niepel, Rudolph, & Schu¨tzwohl, 1991; Meyer, Reisenzein,
& Schu¨tzwohl, 1997; Niepel, Rudolph, Schu¨tzwohl, & Meyer,
1994; Prinz, 1983, 1990; Schu¨tzwohl, 1998; Selz, 1922; Wilcocks,
1928). In short, events that deviate from expectations attract at-
tention (besides other surprise-related changes, e.g., an interrup-
tion of current information processing and action). That is, the
unannounced color singleton captures attention because it is dis-
crepant from expectancies concerning the color of the stimuli in
the critical trial, with surprise capture being a direct consequence
of the color being expectancy discrepant. Such a shift of attention
to the surprise-eliciting stimulus is thought to be adaptive in that it
provides access to limited processing and response-production
capabilities.
Differences Between Contingent and Surprise Capture
Contingent and surprise capture differ in their conditions.
Whereas contingent capture depends on a stimulus's match with
intentions, surprise capture depends on a stimulus's mismatch with
expectations. However, they should not be seen as being in con-
flict. Rather, they can be conceived of as complementary mecha-
nisms, one concerned with expected events for which intentions
for action can be formed in advance, and the other dealing with
unexpected events whose relation to intentions and goals is ini-
tially unknown.
Contingent capture and surprise capture also differ in their
temporal characteristics. Several experiments presented the search
display only briefly (less than 100 ms) and assessed accuracy
(Horstmann, 2002, 2004, in press). With this procedure, it is
possible to compare the relative speed of contingent capture and
surprise capture. For example, Horstmann (in press, Experiment 1)
varied the SOA between the singleton and the search display and
found that with a 200-ms SOA, performance was not improved by
the unannounced first-time presentation of a singleton at the tar-
get's position after 48 homogeneous trials. In contrast, with
400-ms or 600-ms SOAs, performance was improved. Experiment
2 of the same study, in which a more extended range of SOAs was
realized, replicated this result with no significant benefits with
0-ms, 100-ms, or 200-ms SOAs but with significant benefits with
400-ms, 500-ms, and 600-ms SOAs. These results are consistent
with previous studies that did not find performance to have im-
proved with an SOA of 0 ms (Gibson & Jiang, 1998, Experiment
1; Horstmann, 2002, Experiment 2) but did find improvement with
an SOA of 500 ms (Horstmann, 2002, Experiment 1). Horstmann
(2004) obtained a similar time course for the performance costs
that incurred with a nonpredictive singleton (i.e., a singleton that
appears at a distance from the target's position at the position of a
distractor). In contrast, when observers know that the singleton is
presented in each trial at the same position as the target of the
search, performance is scarcely influenced by SOA; that is, per-
formance is nearly as good with simultaneous presentation (0-ms
SOA) as it is with positive SOAs (Horstmann, 2002, in press),
even when the singleton is presented unpredictably in only in 4%
of the trials (Horstmann & Ansorge, in press).
1040 HORSTMANN
Another discriminating temporal aspect is the time course of
surprise capture and contingent capture at repeated presentations.
Surprise capture can be elicited only once (or a very limited
number of times) with the same type of event, because surprise is
elicited by expectancy-discrepant events, and events are antici-
pated (expected) with repeated presentations (e.g., Horstmann &
Ansorge, in press; Meyer et al., 1997; Schu¨tzwohl, 1998). Note
that expectation is not equated with low probability in the present
context, because low-probability events are often expectation con-
gruent (Gibson & Peterson, 2001; Horstmann, in press; Horstmann
& Ansorge, in press). For example, there may be only a 10%
chance that a traffic light will change from green to red when the
driver approaches it--yet this is certainly an expected event. In
contrast, expectation is conceptually closely related to the concepts
of anticipation, hypothesis, and belief. Expectation discrepant thus
means contrary to anticipations, hypotheses, or beliefs (cf. Meyer
& Niepel, 1994). Whereas surprise capture can be elicited by the
same type of event only a limited number of times, contingent
capture can be elicited as often as the task requires the participant
to attend to a feature.
The differences in temporal characteristics suggest different
underlying mechanisms for contingent capture and surprise cap-
ture. In particular, contingent capture is assumed to result from a
match between the attentional control setting and characteristics of
the stimulus. In contrast, surprise capture is assumed to result from
a mismatch between aspects of a schematic expectation (e.g., that
the presented stimuli are red) and corresponding aspects of the
discrepant stimulus (e.g., that one stimulus is green). The expec-
tation is termed schematic for three reasons. First, the concept of
expectation (as used here) focuses on its content rather than on
judgments of probability. Second, the idea of a schema has been
used by Neisser (e.g., 1976, 1979) and others to emphasize the role
of anticipation and expectation in selective attention and conscious
perception, which is also assumed to be important to explain
surprise capture. In particular, it is assumed that expectations are
constant concomitants of cognitive activity and that they are gen-
erated continuously, without intention, without interference with
other cognitive processes, and without being necessarily con-
scious. That is, expectations are assumed to result from processes
that are automatic in the traditional notion (e.g., Posner & Snyder,
1975).
Third, schema theories (e.g., G. Mandler, 1985; J. M. Mandler,
1984; Rumelhart, 1984; Rumelhart & Ortony, 1977; Rumelhart,
Smolensky, McClelland, & Hinton, 1986; Schu¨tzwohl, 1998) em-
phasize the importance of knowledge in perception and cognition.
In particular, preceding and concurrent input suggest knowledge
structures as initial candidates for activated schemas, and once a
reasonable fit has been established between the schema's compo-
nents and the input, the schema is instantiated as a reasonable
hypothesis that accounts for the input. Highly conflicting informa-
tion on whether a given schema accounts for a given input con-
stitutes a schema discrepancy. The input to this process stems from
various sources, including basic features, objects, relations among
objects, concepts, and so forth. Only schema discrepancies that
stem from preattentive processes (i.e., simple features, e.g., color
or form) can produce surprise capture in a narrower sense, whereas
schema discrepancies that stem from attentive processes (e.g.,
recognized objects) only cause spatial attention to be located at the
surprising stimulus longer or more frequently.
Last, contingent and surprise capture differ in function. Contin-
gent capture allows for the quick selection of task-relevant features
and thereby allows one to ignore features that are known to be task
irrelevant or distracting. In contrast, the function of surprise cap-
ture is to make expectancy-discrepant--and thus highly informa-
tive--stimuli available for further processing.
Inattentional Blindness and Other Failures of Unexpected
Events to Capture Attention
A number of results from the literature might seem to contradict
the surprise-capture hypothesis. In the first experiment in which an
unexpected singleton was presented in a search task, Gibson and
Jiang (1998) did not find evidence for attentional capture for a
surprise singleton. The authors presented eight letters briefly, with
the participants' task being to search for H and U. All letters had
the same color in the precritical trials. In the critical trial and the
postcritical trials of Experiment 1, the target had a color different
from that of the remaining letters. Gibson and Jiang found perfor-
mance in the first unannounced presentation of the singleton target
in the critical trial to be no better than in the precritical trials but
significantly worse than in the postcritical trials. In Experiment 2,
the precritical trials corresponded to those of Experiment 1; in the
critical and in the postcritical trials, however, a distractor letter was
a singleton, whereas the target had the same color as the remaining
letters. If the singleton distractor captured attention, interference
with the identification of the target letter would be expected, given
the very limited presentation time of the letters. In contrast to this
hypothesis, performance turned out not to be significantly reduced
in the critical trial. Horstmann (2002, Experiment 2; see also
Horstmann, 2004, in press) replicated the absence of attentional
capture for a singleton target under conditions similar to those used
by Gibson and Jiang (1998). However, using a positive SOA
between the onset of the singleton and the onset of the letters,
Horstmann (2002) found clear evidence of attention capture in the
critical trial, a result that was subsequently replicated and extended
by a systematic variation of SOA. The failure to find evidence for
surprise capture with zero or small positive SOAs, which are
normally sufficient to measure attention to expected singletons, is
probably due to the fact that the latency of surprise capture lags
behind attention to an expected singleton by roughly 300 ms (see
also Horstmann, 2004, in press).
A further source of apparent evidence against the surprise-
capture hypothesis--inattentional blindness (IB)--is more diffi-
cult to review, not least because of the large number of experi-
ments that have been conducted, which in some cases were also
exploratory in nature (Mack & Rock, 1998). However, it is im-
portant to note at first that IB and surprise capture are not in
conflict as phenomena. In contrast, experiments on surprise often
reveal rates of IB similar to those found in experiments directly
concerned with IB. For example, Meyer et al. (1991; see also
Horstmann, 2004, in press) reported that about 20% of the partic-
ipants did not notice the unexpected stimulus change in the critical
trial, which is similar to the IB rates reported by Mack and Rock
(1998). That is, in both types of experiments, the majority of
participants were not blind to the stimulus change, whereas a
minority were.
Second, in some of the most striking demonstrations of IB, a
crucial factor appears to be the similarity of the unexpected stim-
1041CAPTURE AND EXPECTATION DISCREPANCY
ulus to the to-be-attended-to stimuli, on the one hand, and the
to-be-ignored stimuli, on the other hand. In particular, Most et al.
(2001) showed that IB was linearly related to the unexpected
event's similarity to the to-be-ignored stimuli and inversely related
to the similarity to the to-be-attended-to stimuli. For example, if
black items are to be attended to and white items are to be ignored,
a dark gray item is noticed with a higher probability than a light
gray item. This is consistent with a contingent capture account of
involuntary orienting, which states that attentional capture by an
item that is not searched for should be a function of its similarity
to the currently used attentional set (e.g., Folk et al., 1992).
Third, in some cases, the discrepancy of the stimulus from the
current set of schemas pertains to a semantic aspect: The notorious
gorilla among white- and black-dressed basketball players in Si-
mons and Chabris's (1999) experiment, which was not noticed by
many observers who attended to the white-dressed players, was
completely expectancy congruent with regard to its color (black),
way of movement (walking), and gross appearance (primate with
upright posture); it was expectancy incongruent, though, for its
identity. As stated above, it is quite probable that an expectancy
discrepancy can be preattentively detected only for rather simple
features (e.g., Treisman & Gelade, 1980) but not for conjunctions
of features or conceptual attributes implied by the stimuli.
Fourth, many of the experiments discussed in the book on IB by
Mack and Rock (1998) used display durations that were probably
too short for surprise capture. In particular, Mack and Rock's
standard procedure entailed that the stimuli were displayed for 200
ms and pattern masked afterward. As indicated before, surprise
capture appears to lag about 300 ms behind contingent capture, and
a display duration sufficient for intentional attention shifts may not
be sufficient for surprise capture.
Last, some of the displays might have induced rather nonspe-
cific expectations with regard to the features that happened to
characterize the target in the critical trial. For example, in Exper-
iment 3 of Most et al.'s (2001) study, many participants who
attended to white circles and ignored black circles did not notice a
red plus sign that was visible for 30 s. Although I do not want to
suggest this to be the sole factor responsible for the impressively
high rate of IB, the black and white circles had already instantiated
two colors, and prior research has shown that variability in the
critical feature during the precritical trials tends to reduce the
surprise response (Schu¨tzwohl, 1998).
Another Account of Surprise Capture: Singleton Capture
Some models of visual selection propose that salient stimuli can
capture attention in their own right. For example, Guided Search
2.0 (Wolfe, 1994) proposes that intention-driven and data-driven
processes independently contribute to the selection of stimuli,
implying that salient stimuli can capture attention even in the
absence of top-down activation. Similar positions have been
adopted by other authors (e.g., Itti & Koch, 2000; Kim & Cave,
1999; Koch & Ullman, 1985; Theeuwes, 1992, 1994; Theeuwes &
Godijn, 2001). For example, Theeuwes and Godijn (2001) as-
sumed that singletons capture attention in an involuntary and
unavoidable fashion. However, if observers know that the single-
ton is irrelevant to the task (i.e., nonpredictive of the target), they
are able to quickly reorient attention, and this reorientation of
attention is used to explain the ease with which singletons are
ignored under appropriate conditions (e.g., Yantis & Egeth, 1999).
Evidence concerning this position is heavily disputed. That
nominally irrelevant singletons modify performance in a way that
could be predicted by a singleton-capture hypothesis turned out to
be true in a number of studies (e.g., Bacon & Egeth, 1994; Kim &
Cave, 1999; Theeuwes, 1992, 1994; Theeuwes & Burger, 1998;
Todd & Kramer, 1994). What is debatable, however, is whether
such results reflect attentional capture and, if so, whether it is truly
noncontingent on intentions. For example, Kim and Cave (1999)
presented two singletons in each trial. Although only the form
singleton was task relevant and the color singleton was task
irrelevant, the color singleton interfered with task performance.
This result, however, may indicate not purely bottom-up-driven
capture but rather imperfect intentional control over the deploy-
ment of attention, because the task required the participants to
search for a singleton (cf. Bacon & Egeth, 1994). Furthermore,
Folk and Remington (1998) noted that the small costs (in term of
RT or accuracy) that incur with irrelevant singletons need not
necessarily indicate visuospatial shifts of attention but can also be
explained with nonspatial filtering costs that precede shifts of
attention. For example, with two singletons, one defining the target
(e.g., form) and the other being task irrelevant (e.g., color), the
target-defining singleton has to be discriminated from the task-
irrelevant singleton, producing small amounts of performance
costs.
The singleton-capture view is relevant in the current context
because attentional capture by the unannounced first presentation
of a singleton might indicate singleton capture rather than surprise
capture. For example, Theeuwes (e.g., Theeuwes & Godijn, 2001;
see also Kim & Cave, 1999) suggested that singletons generally
capture attention in a purely bottom-up fashion but that observers
may reorient attention very quickly if it is desirable to ignore the
singleton. Thus, it might be argued that the surprise singleton
captures attention precisely because the two conditions for single-
ton capture are met: (a) The singleton is salient, producing a large
amount of bottom-up activation, and (b) intentions to ignore the
singleton are absent. A first indication that attentional capture by
a surprise singleton may not be explained by a singleton-capture
account is that the attentional shift in the surprise trial is relatively
slow: As pointed out before, attentional capture appears to be
somewhat slower for unexpected than for expected singletons, and
the singleton-capture hypothesis cannot explain this difference
(Horstmann, 2004). Rather, proponents of the singleton-capture
hypothesis have proposed that singleton capture occurs within
50­150 ms after the singleton's presentation (e.g., Theeuwes &
Godijn, 2001; see also Kim & Cave, 1999). Notwithstanding this
point, the present experiments address the singleton-capture hy-
pothesis directly.
Objectives of the Present Research
According to the present hypothesis, an unexpected singleton
captures attention to the degree that it is discrepant from expecta-
tions. As in prior experiments, the contents of expectations are
manipulated via the precritical trials in the present experiments.
For example, if all items presented in the precritical trials are
green, it is assumed that the expectation concerning the critical
trial is that the items in that trial will also be green. If one or more
1042 HORSTMANN
red items are presented in the critical trial, these items will also be
expectancy discrepant and draw attention to themselves. In con-
trast, if some of the items in the precritical trials are green and
some are red, a red item in the critical trial will be expectation
congruent.
Second, as already proposed by Selz (1922), the detection of an
expectancy discrepancy is assumed to be an automatic process that
does not need an intention or conscious testing of the expectation.
There is plenty of evidence from previous experimental work on
surprise that this assumption holds, as participants were not re-
quired to test or monitor expectations but were distracted by a
stimulus change (e.g., Horstmann, 2001; Meyer et al., 1991; Nie-
pel et al., 1994; Schu¨tzwohl, 1998).
Third, whether a discrepancy can be detected preattentively
(which is a precondition for attentional capture) depends on
whether the discrepant stimulus aspect is preattentively available.
Of course, the detection of an expectancy discrepancy often de-
pends on attending to the stimulus first, in particular for stimuli
that are expectancy discrepant with respect to their specific com-
bination of basic perceptual features (e.g., Treisman & Gelade,
1980). This is because the combination of basic features as well as
their precise spatial relations are available only through attention.
In contrast, basic perceptual features, such as color, motion, or
orientation, are available prior to attention. Thus, an expectancy
discrepancy of those features can also be--at least in principle--
detected preattentively.1
Fourth, it is assumed that, following the discrepancy detection,
attention is involuntarily directed to the discrepant stimulus's
features and thereby to the discrepant stimulus's location. An
involuntary shift of attention is evident from the RT costs revealed
in the critical trials of many experiments in which the physical
appearance of a task-irrelevant experimental stimulus was differ-
ent from that in the precritical trials (e.g., Meyer et al., 1991;
Schu¨tzwohl, 1998). Of course, these experiments did not demon-
strate spatial shifts but showed selective attention in the sense of a
selection of the discrepant stimulus for further processing at the
cost of a selection of the target stimulus of the RT task.
I have previously described the most straightforward examples
of an expectancy-discrepant event (the first presentation of a new
color after consistent presentations of a different color) and of an
expectancy-congruent event (the presentation of a color in the
critical trial that has been already presented in the precritical
trials). As already reported, attentional capture by the first presen-
tation of a new color after consistent presentations of another color
has been tested and repeatedly found in several experiments
(Horstmann, 2002, 2004, in press). The aim of the present exper-
iments is to specify more precisely the conditions of surprise
capture. To this end, the present experiments vary the relation
between the expectancy and the unannounced singleton in the
critical trial. More precisely, the critical trial is essentially the same
in all experiments, that is, a color singleton in an otherwise
color-homogeneous display. To clarify the conditions of surprise
capture, I varied the stimuli displayed in the precritical trials
among experiments to induce different expectations and intentions.
The variations included presentations in the precritical trials of
differently colored nonpredictive singletons (Experiment 1), non-
predictive form singletons (Experiment 3), color-heterogeneous
displays with two colors alternating in adjacent positions (Exper-
iment 4), two differently colored homogeneous displays (Experi-
ment 5), a single color change with homogeneous displays (Ex-
periment 6), and form-heterogeneous displays with two forms
alternating in adjacent positions (Experiment 7).
All experiments used a modified version of the paradigm intro-
duced by Gibson and Jiang (1998, Experiment 1), in which in-
structions inform about the precritical trials but make no reference
to the display types in the critical and the postcritical (i.e., color-
heterogeneous) trials. The basic assumption underlying the present
experiments is that the precritical trials induce implicit expecta-
tions about what the following displays will look like. More
precisely, a schematic representation of the spatial and temporal
aspects of each trial is formed on the basis of the instructions and
the first trials, which represents the typical layouts and their
sequence in each trial as well as their relation to the task. Thus,
different presentation conditions are assumed to induce different
expectations, and the presentation conditions in the critical trial
may or may not match the expectations. If the singleton in the
critical trial is discrepant from expectations, surprise capture
should occur--that is, the position of the singleton should be
attended to.
I assessed attentional capture by observing the combined effects
of a set-size variation and the presentation of the singleton at the
target's position. Search demands were high with homogeneous
displays or heterogeneous displays with nonpredictive singletons
in the precritical trials, with RT depending on set size. In contrast,
if a feature at the target's position captures attention, RT should
not depend on set size. That is, set size and trial type should
interact when the singleton at the target's position captures atten-
tion. If, however, the singleton fails to capture attention, no Set
Size Trial Type interaction should occur.
For the interpretation of the critical trial RTs, it must be con-
sidered that RTs could be longer in the critical trial than in the
postcritical trials because of the surprise-induced RT delay (cf.
Gibson & Jiang, 1998; Horstmann, 2002; Meyer et al., 1991),
caused by a distraction through the surprising event. The RT delay
probably reflects decision-level processes concerning the surpris-
ing event (e.g., a determination of the meaning and the task
relevance of the surprising event) as well as automatic processes,
such as the change of the current sets of schemas (e.g., Schu¨tz-
wohl, 1998). The duration of these surprise-related processes
should thus depend on the type of the surprising change but should
be independent of set size. Further, because an attentional shift and
processes concerning the surprising event occur at different pro-
cessing stages, their effects must be additive (Sternberg, 1969). As
a concrete example, if one assumes that the attention shift to the
surprise singleton has a latency of 300 ms (independent of set size)
and the decision-level processes have a duration of 500 ms (also
independent of set size), the RT in the critical trial will be 800 ms,
independent of set size (ignoring the duration of response-related
1
Note that because the spatial position of a feature is preattentively not
precisely available (Treisman & Gelade, 1980), the discrepancy cannot
pertain to the "wrong" position of the feature but only to its presence or
absence in the entire scene. That is, if a red object is presented consistently
at a specific spatial position until its position is changed in the critical trial,
this will normally not be detected preattentively but only after the prior or
the actual position of the object has been attended to.
1043CAPTURE AND EXPECTATION DISCREPANCY
processes). If the mean RT with serial search (e.g., in the precrit-
ical trials) is 500 ms with a small set size and 1,200 ms with a large
set size, RTs in the critical trial will be longer than with serial
search with a small set size but shorter with a large set size. This
pattern of results was observed in Horstmann (2002, Experiment 3)
with the set sizes of 4 versus 12 stimuli. Thus, the critical result
indicative of attentional capture in the present experiments is a Set
Size Trial Type interaction but not a general decrease of RTs
with all set sizes. A general decrease of RTs with all set sizes is
only to be expected if the set sizes tested are sufficiently large (see,
e.g., the present Experiment 8).
The questions addressed in the experiments are as follows.
Experiments 1, 3, and 7 examine the specificity of expectations
within and between the static feature dimensions of color and
orientation. More precisely, I test whether the presentation of a
nonpredictive color versus form singleton in the precritical trials
eliminates attentional capture by a color singleton defined by a
new color. Experiments 4­6 further test whether the expectation
discrepancy instigating attentional capture concerns color or other
aspects of the display in the critical trial, such as heterogeneity. In
addition, Experiments 4­6 test whether the alternative hypothesis
that singletons always capture attention (e.g., Theeuwes, 1992,
1994; Theeuwes & Godijn, 2001) can account for attentional
capture by the unannounced first presentation of a singleton.
Finally, Experiment 8 seeks to replicate the original attentional
capture effect after color-homogeneous precritical trials with large
set sizes of 25 versus 49 stimuli.
Experiment 1A: Homogeneous Versus Heterogeneous
Displays in the Precritical Trials
Experiment 1A seeks to replicate and extend Horstmann's
(2002) Experiment 3, in which the first unannounced presentation
of a color singleton at the target's position resulted in a strong
reduction of the set-size effect on RTs. The replication part of the
experiment was instantiated in the homogeneous display condition,
which differs from Horstmann's Experiment 3 only in minor
changes in the stimulus material. As an extension of that experi-
ment, the heterogeneous display condition was created. In this
condition, a color singleton was presented in each trial; in the
precritical trials, however, the singleton did not predict the position
of the target and should therefore be ignored (e.g., Yantis & Egeth,
1999). In the critical and the postcritical trials, a different color
was used for the singleton; however, it was always the target.
I included the heterogeneous display condition to examine the
specificity of expectations concerning color. I reasoned that if
expectations are highly specific within the static feature dimension
of color, the new color in the critical trial should be expectancy
discrepant and thus capture attention. In contrast, if expectations
are broad and fuzzy and do not discriminate strongly within the
color dimension, the new color should be consistent with expec-
tation and thus fail to attract attention. Schu¨tzwohl's (1998; see
also Schu¨tzwohl, 1993) research suggests that variability within a
dimension in the precritical trials reduces the unexpectedness of a
new value on this dimension in the critical trial. For example,
Schu¨tzwohl (1993, Experiment 6) presented words in the precrit-
ical trials either in seven different typefaces or in a constant
typeface. A new typeface presented in the critical trial was con-
sidered as unexpected by fewer participants and was rated as less
surprising in the variable-typeface condition than in the constant-
typeface condition. Thus, there appears to be a degree of general-
ization over the specific stimuli presented, which suggests that the
new color in the critical trial of the present experiment will
probably not be highly expectancy discrepant. Moreover, because
the task is to ignore the color singleton, there is no additional
incentive to discriminate neatly between the presented colors.
Method
Participants. Fifty students (13 men and 35 women) participated in the
study for a small monetary exchange. Their mean age was 25.2 years
(SD 5.9).
Apparatus. A microcomputer, equipped with an Intel 80486/100MHz
central processing unit, a keyboard, and a 15-in (38.10-cm) cathode ray
tube monitor, was used for stimulus presentation and response registration.
Response keys were two adjacent keys in the lower row of the keyboard
(the arrow left and arrow down keys). ERTS (BeriSoft, Frankfurt, Ger-
many) was used for event scheduling and RT measurement.
Design. Four conditions resulted from orthogonally crossing the two
between-subjects variables, set size (4 vs. 12 letters) and display type, in
the precritical trials (homogeneous vs. heterogeneous).
Stimuli. Figure 1 illustrates the stimulus conditions. Stimuli were pre-
sented against a black background. Viewing distance was approximately 60
cm. The fixation cross subtended 0.2° of visual angle. The letters (all taken
from the same sans serif font type) were uppercase and subtended about
0.7° in height and 0.6°­0.9° in width. The target letters were L and R. For
the set-size 4 condition, the 3 distractor letters were P, T, and W. For the
set-size 12 condition, the 11 distractor letters were P, B, K, N, S, H, T, I,
E, J, and W. The colors of the letters were white, green, or red, depending
on condition. Colors were not matched for luminance; rather, it was
assumed that the subjectively large difference in hue would be the most
important difference among the colors (e.g., Folk et al., 1992). Distance
between letters (about 2°), rather than retinal eccentricity, was held con-
stant across the two set-size conditions. For the set-size 4 condition, the
letters were presented on the invisible circumference of a small (1.3°
radius) circle; for the set-size 12 condition, the letters were presented on the
circumference of a large (3.8° radius) circle. The reason for letting retinal
eccentricity rather than the distance between the letters (density) vary
between the two set sizes was that density is assumed to be an important
determinant of salience (cf. Yantis & Egeth, 1999; see also Nothdurft,
2000): Other things being equal, the higher the density of stimuli is, the
higher is the salience of a singleton. In contrast, eccentricity per se affects
RTs in visual search only slightly, given that targets and nontargets are
presented at equal eccentricities (Wolfe, O'Neill, & Bennett, 1998), as
secured by the use of circular displays in the present experiment.
Procedure. Each trial began with the presentation of the fixation cross
in the center of the screen for 1 s (see Figure 1). The fixation cross was
replaced by the letter array, which was presented until a response was
registered but for a maximum of 4 s. The imaginary circle on which the
letters were presented had its center at fixation. In the set-size 4 condition,
letters were presented on the 3, 6, 9, and 12 o'clock positions of the circle.
In the set-size 12 condition, letters were presented on the 12 full-hour
positions of the circle.
In each trial, 1 of the 2 target letters (L or R) plus 3 (in the set-size 4
condition) or 11 (in the set-size 12 condition) distractor letters were
presented. Which of the 2 target letters appeared, its position, and the
position of the distractor letters were determined at random, with a new
random sequence computed for each subject. The 2 target letters (L and R)
were equiprobable. The participants were instructed to press the left re-
sponse key if the search array contained an L and the right response key if
it contained an R. Response speed was strongly emphasized, but partici-
pants were instructed to be accurate as well; false responses were imme-
diately followed by error feedback, which consisted of a short tone.
1044 HORSTMANN
After 12 trials of familiarization with the task, there was one single
experimental block that comprised 96 trials. The first 48 trials constituted
the precritical trials. Trial 49 was the critical trial, and the following 47
trials constituted the postcritical trials. The transition between the last
precritical trial and the critical trial did not differ from the transition
between the other trials.
In the homogeneous-display condition, the distractor letters were all
green throughout the experiment. In this condition, the target letter was
likewise green in the precritical trials, and it was white in the critical and
in the postcritical trials. In contrast, in the heterogeneous-display condition,
one randomly determined letter was white in the precritical trials. In the
critical and postcritical trials, the target letter appeared as a red singleton
among green distractor letters.
Results
Errors and RTs longer than 3,000 ms were excluded from the
RT analysis, which resulted in a loss of 4.6% (3.6% due to errors;
see Table 1) of the experimental trials, two of them critical trials;
this reduced the number of participants to 48. (The design assumed
equal numbers of participants in each cell, but, because of an error,
there were 13 participants in the set-size 4 homogeneous-display
condition and 11 participants in the set-size 12 heterogeneous-
display condition.) Figure 2 shows the mean RTs for the
homogeneous-display condition (Figure 2, left panel) and the
heterogeneous-display condition (Figure 2, right panel). Error rates
are depicted in Table 1. In the precritical trials, the homogeneous
and the heterogeneous display groups both showed large set-size
effects (71 ms/letter and 91 ms/letter, respectively). In contrast, in
the postcritical trials, the set-size effect was very small (9 ms/letter
and 10 ms/letter, respectively). In the critical trial, however, the
results differed markedly: Whereas the set-size effect was only 11
ms/letter in the homogeneous display condition, it remained high
(86 ms/letter) in the heterogeneous display condition.
RT analysis paralleled that of Horstmann (2002, Experiment 3).
First, an overall analysis of variance (ANOVA) was conducted.
This 2 (set size: 4 vs. 12) 2 (precritical trials: homogeneous vs.
heterogeneous displays) 3 (trial type: precritical vs. critical vs.
postcritical trial) ANOVA revealed significant main effects of trial
type, F(2, 88) 43.1, p .01, and set size, F(1, 44) 33.1, p
.01, as well as a significant Trial Type Set Size interaction, F(2,
88) 12.3, p .01, and two marginally significant interaction
terms, Precritical Trials Set Size, F(1, 44) 4.0, p .05, and
Trial Type Precritical Trials Set Size, F(2, 88) 3.6, p .05
(here and henceforth, I computed the interaction terms involving
B L
T
I
J
EPK
S
W
N
H
B
T
I
J
EPK
S
W
N
H
TP
L
W
TP
W
eS t iS ze 12 eS t iS ze 4
am x
4, m000 s1, m000 s 1, m000 s
am x
4, m000 s
Figure 1. The trial structure in Experiment 1. Each trial consisted of two frames: a fixation cross presented for
1,000 ms, and the search display, which was presented until a response was made. Two display sizes--12 letters
and 4 letters--are depicted. There were two types of displays. In homogeneous displays (upper row), all letters
had the same color. In contrast, in heterogeneous displays (lower row), one letter had a different color
(symbolized by a stronger line). max maximum.
Table 1
Error rates (in percentages) for Experiments 1A and 1B
Condition Precritical Critical Postcritical
Experiment 1A
Homogenous display condition
Set size 4 3.4 0.0 3.1
Set size 12 5.0 0.0 3.4
Heterogeneous display condition
Set size 4 3.4 7.7 1.6
Set size 12 6.8 0.0 2.5
Experiment 1B
Set size 4 4.1 0.0 4.7
Set size 12 6.0 0.0 2.9
1045CAPTURE AND EXPECTATION DISCREPANCY
trial type using Huynh­Feldt corrected degrees of freedom to
adjust for violations of sphericity assumptions; the uncorrected
degrees of freedom are reported to maintain readability). A corre-
sponding error analysis (of error proportions) revealed no signif-
icant main effects or interactions (Fs 1.8, ps .10), indicating
that the interpretation of the RT results is not complicated by a
speed­accuracy trade-off.
Whether attentional capture occurred in the homogeneous and in
the heterogeneous display conditions was tested by two separate
Set Size (4 vs. 12) Trial Type (precritical vs. critical trial)
ANOVAs. Search was predicted to be serial in the precritical trials
in both conditions, implicating a large set-size effect in these trials.
In contrast, surprise capture by the singleton in the critical trial
should eliminate the set-size effects in the homogeneous-display
condition but not--or not as strongly--in the heterogeneous-
display condition. For this reason, I predicted a Set Size Trial
Type interaction for the homogeneous display condition (which
would indicate a reduced set-size effect in the critical trial) but not
for the heterogeneous display condition (which would indicate
similar set-size effects in the precritical and the critical trial). The
results were consistent with these predictions. The ANOVA re-
vealed a significant main effect of set size in both conditions--
homogeneous displays, F(1, 23) 6.2, p .05; heterogeneous
displays, F(1, 21) 41.6, p .01--that was modified by a
significant Trial Type Set Size interaction in the homogeneous
display condition, F(1, 23) 8.3, p .01, but not in the hetero-
geneous display condition (F 1; the trial-type main effects were
not significant, Fs 1). Follow-up t tests revealed that in the
homogeneous display condition, set size had an effect on RTs in
precritical trials, t(23) 6.7, p .01, but not in the critical trial,
t(21) 1, whereas in the heterogeneous display condition, set size
had an effect on RTs both in the precritical trials, t(21) 16.6, p
.01, and in the critical trial, t(21) 3.2, p .01.
I conducted the analogous analyses by comparing perfor-
mance in the critical trial with performance in the postcritical
trials. According to the present hypotheses, the RT in the
critical trial of the homogeneous-display condition reflects pri-
marily the additive components of (a) time for surprise capture
to occur, (b) time consumed by the decision-level processes
concerned with the surprising event, and (c) the processes that
translate the stimulus letter into the response. In contrast, in the
postcritical trials, RT should depend primarily on the additive
components of (a) the latency of contingent capture and (b)
stimulus­response (S-R) translation processes. Neither of the
mentioned processes in the critical and the postcritical trials
should be affected by set size. However, because the latency of
contingent capture was shorter than the latency of surprise
capture and because the processing of the surprising event was
absent from the postcritical trials, the RTs in the critical trials
should be longer than those in the postcritical trials. These
predictions were supported by the ANOVA for the homoge-
neous display condition, which revealed a significant main
effect of trial type only, F(1, 23) 25.8, p .01, whereas
neither effect involving set size was significant (Fs 1).
For the heterogeneous-display condition, RTs in the critical and
the postcritical trials should be different with respect to the effi-
ciency of the target search, being nonefficient in the critical trial
but efficient in the postcritical trials. Consistent with these predic-
tions, there were significant main effects of block, F(1, 21) 23.7,
p .01, and set size, F(1, 21) 12.3, p .01, as well as a
significant Block Set Size interaction, F(1, 21) 7.7, p .05.
This pattern of results reveals that the set-size effect was reduced
only after the critical trial.
When would observers in the heterogeneous-display condition
change their search strategy in response to the new stimulus
conditions? To answer this question, I examined the surprise trial
and the immediately following trials. There was a rapid decrease in
the set-size effect, beginning with the trial immediately following
the surprise trial (i.e., search rates of 86, 29, 24, and 17 ms/item in
the critical and the three following trials, Trials 49, 50, 51, and 52,
respectively), which indicates that the majority of participants
noticed the change in the surprise trial or in the following trial.
Figure 2. Results from Experiment 1. Mean reaction times (RTs) are presented for the three types of trials
(precritical, critical, and postcritical), separately for each set size (set-size 12: triangles; set-size 4: squares). The
left part of the figure shows the results of the homogeneous-display conditions, whereas the right part of the
figure shows the results of the heterogeneous-display condition. Exp Experiment; Homogen. homogeneous;
Heterogen. heterogeneous.
1046 HORSTMANN
Discussion
The findings for the homogeneous-display condition closely
replicate the results of Horstmann (2002, Experiment 3), which
also found that the set-size effect present in the precritical trials
was strongly reduced in the critical trial, which indicates that the
surprise singleton captured attention. Also, as in that prior exper-
iment, the set-size effect was similar to that in the postcritical
trials, which indicates that attention capture was similarly efficient
on its first presentation and in the postcritical trials, in which the
singleton could be used intentionally to guide attention.
RTs were elevated in the critical trial as compared with the post-
critical trials, reflecting distraction by the surprising event. This effect
is usually obtained when a surprising event is presented during a
reaction task (e.g., Horstmann & Schu¨tzwohl, 1998; Meyer et al.,
1991, 1997; Niepel et al., 1994; Schu¨tzwohl, 1998). Because the
decision-level processes reflected by the distraction and the attentional
shift toward the surprising event logically occur at different process-
ing stages, their effects should be additive (Sternberg, 1969). This
assumption is consistent with the additivity of the RT increase with set
size in the critical and the postcritical trials.
In contrast, the heterogeneous-display condition shows that not
all types of display changes involving singletons capture attention:
When observers were presented with a nonpredictive singleton in
each precritical trial, no immediate attentional response occurred
to an unannounced color change. This pattern of results contrasts
sharply with the homogeneous display condition. In the present
account, this result was due to expectations built up during the
precritical trials that were matched--or at least did not mis-
match--in the critical trial. That is, because of the heterogeneous
precritical trials, participants expected a color singleton in each
trial. An expectation of a color singleton could further be analyzed
as consisting of two confounded expectations: the expectation of a
certain color (or range of colors), and the expectation of a singleton
(or of heterogeneity of the display). One aim of the following
experiments is to clarify the contribution of these confounded
expectation components to the mismatch in the critical trial. How-
ever, the following Experiment 1B was conducted to first replicate
the heterogeneous display condition of Experiment 1 while elim-
inating a possible alternative account.
Experiment 1B: Heterogeneous Displays With
Digital-Clock Stimuli
Experiment 1B seeks to replicate the absence of attentional capture
in the heterogeneous display condition, with minor differences from
Experiment 1A in stimuli and procedure. The most important change
concerned the stimuli. In particular, digital-clock-like stimuli were
used to eliminate any differences in basic features between targets and
distractors (Gibson & Jiang, 1998; Horstmann, 2002, 2003). This was
to ensure that the participants would not enter a "feature-detection
mode" (Bacon & Egeth, 1994) on the basis of subtle differences
between targets and distractors on a basic feature dimension. Second,
the number of precritical trials was increased to test whether absence
of attentional capture in the heterogeneous display condition was due
to its insufficient exposure to the precritical stimuli. Third, the non-
predictive singleton in the precritical trials was a red letter among
green letters, whereas the predictive singleton in the critical and the
postcritical trials was a white letter. This change tests the possible
contention that the red letter in the critical trial of Experiment 1A did
not capture attention because it was less salient than the to-be-ignored
white letter in the precritical trials.
Method
Participants. Participants were 10 men and 15 women who partici-
pated for a small monetary exchange. Their mean age was 21.8 years
(SD 2.6). None of them had participated in the previous experiment.
(Naive participants were used in all experiments reported in this study.)
Design. Participants were randomly assigned to one of the two set-size
conditions (4 vs. 12 letters).
Apparatus, stimuli, and procedure. Hardware and software were of the
same type as in Experiment 1A. Stimuli were presented against a black
background. The letters were constructed from vertical and horizontal line
segments only. Each letter subtended 0.8° in height and 0.7° in width. The
target letters were H and U. For the set-size 4 condition, the distractor
letters were A, B, and D. For the set-size 12 condition, the distractor letters
were A, B, C, D, E, F, I, L, P, S, and T. The stimuli were presented on the
imagined circumference of a circle with radii of 0.9° and 2.3° in the set-size
4 and the set-size 12 conditions, respectively. The remaining stimuli were
the same as in Experiment 1A.
In the 24 practice trials and the following 72 precritical trials, the distractor
letters were green and the singleton letter was red. In the critical trial and the
following 23 postcritical trials, the target letter was always white. For com-
parison purposes, note that the present experiment is a clone of Horstmann's
(2002) Experiment 3, the only difference being in the precritical trials. That is,
it realizes stimulus conditions in the critical trial that have proven to cause
attentional capture after homogeneous precritical trials.
Results and Discussion
Errors and RTs longer than 3,000 ms were excluded from the
analysis, resulting in a loss of 5.3% of the experimental trials, includ-
ing one in the critical trial; this reduced the number of participants
contributing to the RT analysis to 24 (12 in each group).
Figure 2 (right panel) shows the mean RTs for the two groups in
the three trial types, and Table 1 shows the corresponding error
rates. RTs were first analyzed by means of a 2 (set size: 4 vs. 12)
3 (trial type: precritical vs. critical vs. postcritical trials) ANOVA.
Both main effects were significant--set size, F(1, 22) 49.8, p
.01; type of trial, F(2, 44) 39.8, p .01--as was the Set Size
Trial Type interaction, F(2, 44) 20.3, p .01.
A corresponding error analysis revealed a significant main ef-
fect for trial type only, F(2, 44) 15.4, p .01. The set-size main
effect (F 1) and the Trial Type Set Size interaction, F(2,
44) 2.0, p .10, were not significant. Fewer errors occurred in
the critical trial (0.0%) than in precritical trials (5.0%) and in
postcritical trials (3.8%), indicating that observers traded speed for
accuracy in the surprise trial. This trade-off, however, does not
complicate the interpretation of the RTs, because the main effect
for type of trial is not essential in the present analysis.
The main prediction concerned the interaction effect between
set size (4 vs. 12) and trial type (precritical vs. critical trial). A
corresponding ANOVA of the RTs revealed a set-size effect only,
F(1, 22) 62.0, p .01; the other main effect and the interaction
term were not significant (Fs 1). There was thus no indication
of attentional capture in the critical trial.
The critical and the first four postcritical trials were examined as
in Experiment 1A. Again, there was a rapid decrease in the set-size
effect, beginning in the trial immediately following the critical
1047CAPTURE AND EXPECTATION DISCREPANCY
trial. Similar to Experiment 1A, this decline was strongest imme-
diately after the critical trial, indicating that most of the partici-
pants recognized that the singleton was always the target letter and
changed their attentional control settings to the effect that they
rapidly shifted their attention to the singleton letter (calculated
search rates were 63, 20, 11, and 6 ms/item in the critical and the
three following trials, Trials 73, 74, 75, and 76, respectively).
One can draw the preliminary conclusion that two experiments,
with minor differences concerning stimulus shape, number of
precritical trials, and eccentricity, revealed nearly identical results;
with nonpredictive singletons in the precritical trials, an unan-
nounced color change of the singleton did not capture attention.
Experiment 2: Colored Squares
In some of the following Experiments 2­7, it was necessary to
dissociate the singleton feature from the letter stimulus. For this
reason, the singleton feature pertained to a color patch that was
presented as a background for the letters. This procedure had
already been used by Horstmann (2002, 2004, in press) to allow
for a temporal disconnection of the color singleton and the letters
relevant for the search task. However, because this procedure had
not been used before in the RT paradigm, I conducted Experiment
2 to first establish the surprise capture effect with these stimuli
before, in Experiments 3­7, testing further hypotheses.
Method
Participants. Participants were 6 men and 26 women with a mean age
of 24.4 years (SD 5.8). Compensation was 1.
Stimuli, procedure, and design. Black letters (0.7° 0.8°), all com-
posed of vertical and horizontal line segments (as in Experiment 1B), were
presented against colored squares (1.2° 1.2°). The background color was
black. In the set-size 12 condition, the compound letter/square stimuli were
presented in the 12 full-hour positions on the circumference of an imaginary
3.3°-radius circle, whereas in the set-size 4 condition, the stimuli were pre-
sented on the 3, 6, 9, and 12 o'clock positions of a 1.5°-radius circle. The
compound letter/square stimuli were presented until a response was made, but
a trial was aborted and counted as an error if there was no response within 4 s.
Error feedback was given if the wrong reaction key was pressed.
The precritical trials comprised 48 experimental and 12 practice trials.
The squares in the precritical trials all had the same color. In each trial,
each position had an equal probability of containing the target. The critical
trial and the following 23 postcritical trials differed from the precritical
trials only in that one square (the color singleton) was presented in a
different color. The color-singleton square always contained the target.
Participants were randomly assigned to one of four conditions that
resulted from orthogonally combining the variables set size (4 vs. 12
letters) and standard color­singleton color (red­green vs. green­red).
Results
Errors and RTs longer than 3,000 ms were excluded from the
RT analysis, resulting in a loss of 4.1% of the experimental trials.
One of the errors occurred in the critical trial, reducing the number
of participants in the RT analyses to 20, 10 per set size. Figure 3
(panel a) shows the main results. There was a large set-size effect
in the precritical trials (87 ms/letter) but only a small set-size effect
in the critical trial (4 ms/letter) and in the postcritical trials (15
ms/letter).
An overall ANOVA2
with the variables' set size (4 vs. 12) and
trial type (precritical vs. critical vs. postcritical) revealed signifi-
cant effects of set size, F(1, 18) 5.6, p .05, and of trial type,
F(2, 36) 23.9, p .01, and a significant Set Size Trial Type
interaction, F(2, 36) 13.6, p .01. A corresponding error
analysis revealed no significant effects whatsoever (Fs 1). Mean
error rates appear in Table 2.
The main test determined whether the effect of set size was
different for the precritical trials versus the critical trial. A 2 (set
size: 4 vs. 12) 2 (trial type: precritical vs. critical trial) ANOVA
revealed the predicted Set Size Trial Type interaction, F(1,
18) 16.6, p .01, and a significant set-size effect, F(1, 18)
6.1, p .05. The main effect of trial type was not significant, F(1,
18) 2.4, p .10. The significant interaction indicates that RTs
changed differently in the two set-size conditions: They were 497
ms longer in the set-size 4 condition, t(9) 3.4, p .01, but 225
ms shorter in the set-size 12 condition, t(9) 2.3, p .05
(two-tailed).
I further tested whether the set-size effect was different in the
critical and the postcritical trials. A 2 (set size: 4 vs. 12) 2 (trial
type: critical vs. postcritical trial) ANOVA revealed a main effect
of trial type only, F(1, 18) 32.1, p .01 (all other Fs 1),
indicating that the set-size effects did not differ but that RTs in the
critical trial were much longer than in the postcritical trials.
Discussion
The results replicate the results pattern of Experiment 1A,
indicating that it is largely irrelevant whether the singleton color is
a feature of the target or of a different object that occupies the
same spatial position as the target. Having established that a color
singleton captures attention with this experimental set-up, I return
in Experiment 3 to the question of why the color singleton did not
capture attention in the heterogeneous display conditions of Ex-
periments 1A and 1B.
Experiment 3: Changed Singleton Dimension
Experiment 1 showed that when the stimuli in the precritical
trials required the participants to treat color singletons as irrele-
vant, a new color singleton in the critical trial did not capture
attention. This result was predicted on the assumption that the
expectation concerning color is relatively broad within the dimen-
sion of color after the presentation of two rather different colors in
the precritical trials of the heterogeneous display conditions. It is,
however, unclear from this result whether it was indeed the color
variation or rather the presentation of a singleton per se in the
precritical trials that hindered attentional capture. To disentangle
color variation and singleton presence, I included in each precrit-
ical trial of the present experiment an orientation singleton that
was nonpredictive of the target's position. That is, the precritical
trials were orientation heterogeneous but color homogeneous. In
the critical trial, a color singleton was presented at the position of
the target. Thus, if an expectation of a singleton can cause the
absence of an attentional shift to the color singleton in the critical
trial, the set-size effect in the critical trial should be equal to that
2
The variable singleton/nonsingelton color, if included in the ANOVA,
revealed no significant main effect or interactions with the other variables,
with all Fs 1.
1048 HORSTMANN
in the precritical trials. If, conversely, color homogeneity in the
precritical trials is the more important condition for attention to a
new color in the critical trial, the set-size effect in that trial should
be equal to that in the postcritical trials.
Method
Participants. Participants were 6 men and 26 women with a mean age
of 24.4 years (SD 5.8). Compensation was 1.
Stimuli, procedure, and design. These were the same as in the preced-
ing experiment, except for the composition of the precritical trials. In the
precritical trials, all but one square (the orientation-singleton square) were
presented in upward orientation, whereas the orientation-singleton square
was rotated by 45° (see Figure 4). The position of the orientation singleton
was determined randomly, with all positions being used equally often. On
each trial, each position had an equal probability of containing the target,
such that the position of the orientation singleton did not predict the
position of the target. The critical trial and the following 23 postcritical
trials were the same as in Experiment 2, with a color-singleton square that
always contained the target.
Results
Errors and RTs longer than 3,000 ms were excluded from the
RT analysis, resulting in a loss of 4.6% of the experimental trials,
one in the critical trial; this reduced the number of participants to
31, 16 in the set-size 4 condition and 15 in the set-size 12
condition. Figure 3 (panel b) shows the main results. There was a
large set-size effect in the precritical trials (73 ms/letter) but only
a small set-size effect in the postcritical trials (6 ms/letter). The
set-size effect was reduced in the critical trial (38 ms/letter),
although it was clearly not zero.
Figure 3. Mean correct reaction times (RTs) for both set-size conditions (set-size 12: triangles; set-size 4:
squares) in the precritical, critical, and postcritical trials of Experiments 2­7.
Table 2
Error rates (in percentages) for Experiments 2­7
Experiment/Condition Precritical Critical Postcritical
Experiment 2
Set size 4 3.1 0.0 4.6
Set size 12 2.7 9.1 3.4
Experiment 3
Set size 4 2.6 0.0 3.8
Set size 12 3.9 6.3 3.8
Experiment 4
Set size 4 2.9 0.0 4.3
Set size 12 4.3 0.0 2.4
Experiment 5
Set size 4 2.4 7.7 1.3
Set size 12 2.7 0.0 3.2
Experiment 6
Set size 4 4.4 6.7 3.3
Set size 12 4.5 0.0 3.3
Experiment 7
Set size 4 3.8 0.0 2.5
Set size 12 4.3 0.0 3.9
1049CAPTURE AND EXPECTATION DISCREPANCY
An overall ANOVA with the variables set size (4 vs. 12) and
trial type (precritical vs. critical vs. postcritical) revealed signifi-
cant effects of set size, F(1, 29) 30.0, p .01, and of trial type,
F(2, 58) 62.7, p .01, and a significant Set Size Trial Type
interaction, F(2, 58) 22.2, p .01. A corresponding error
analysis revealed no significant effects whatsoever (Fs 1.4).
Mean error rates appear in Table 2.
The main test determined whether the effect of set size was
different for the precritical trials versus the critical trial. A 2 (set
size: 4 vs. 12) 2 (trial type: precritical vs. critical trial) ANOVA
revealed the predicted Set Size Trial Type interaction, F(1,
29) 9.0, p .01, and a significant set-size effect, F(1, 29)
36.1, p .01. The main effect of trial type was not significant
(F 1). The significant interaction indicates that RTs changed
differently in the two set-size conditions; they were 159 ms longer
in the set-size 4 condition, t(15) 3.2, p .01, but tended to be
123 ms shorter in the set-size 12 condition, t(14) 1.5, p .15
(two-tailed).
I further tested whether the set-size effect was different in the
critical and the postcritical trials. A 2 (set size: 4 vs. 12) 2 (trial
type: critical vs. postcritical trial) ANOVA revealed main effects
of set size, F(1, 29) 7.1, p .05, and trial type, F(1, 29) 75.4,
p .01, as well as a significant Set Size Trial Type interaction,
F(1, 29) 7.9, p .01. The Set Size Trial Type interaction
indicates that set size had some effect in the critical trial but
virtually no effect in the postcritical trials.
Discussion
Experiments 1A and 1B showed that the new color in the critical
trial did not capture attention after repeated presentation of non-
predictive color singletons in the precritical trials. The present
experiment reveals that it was not the presence of a nonpredictive
singleton in the precritical trials per se that prevented attentional
capture from occurring. The set-size effect in the critical trial of the
present experiment was clearly reduced as compared with the
precritical trials. That is, the change from single-colored displays
in the precritical trials to the multicolored display in the critical
trial appears to be sufficient to draw attention. However, attention
capture was not as strong as in the present Experiment 2 (or as in
Horstmann, 2002, Experiment 3). This indicates that the combi-
nation of a new color and the first presentation of a singleton was
an especially potent attractor of attention.
In summary, the present experiment shows that the first presen-
tation of a new color, which need not necessarily be the first
presentation of a singleton, was sufficient to attract attention. In
the next experiment, I test whether the first presentation of a
singleton is sufficient to capture attention--that is, whether it
captures attention in the absence of a change from single-colored
to multicolored displays.
Experiment 4: Heterogeneous Nonsingleton Displays
The findings of the present and of prior experiments (Horst-
mann, 2002, 2004, in press) are consistent with the view that the
first and unannounced presentation of a color singleton captures
attention to the extent that the singleton color sufficiently deviates
from expectations and does not capture attention if the color is
largely consistent with expectations. However, the findings may be
seen as being consistent with a quite different view on attentional
capture: Singletons may always capture attention, but participants
may immediately reorient away from the singleton if they believe
that it is nonpredictive (e.g., Kim & Cave, 1999; Theeuwes &
Godijn, 2001). According to this reorientation account, an unan-
nounced singleton captures attention when preceded by homoge-
neous trials because, as participants do not expect a distracting
21eziSteS 4eziSteS
xam
sm000,4sm000,1 sm000,1
xam
sm000,4
Figure 4. Trial structure and display types in Experiment 3. All trials were either form heterogeneous but color
homogeneous (upper row) or color heterogeneous but form homogeneous (lower row). Display samples are
shown for both set sizes (left: 12 letters; right: 4 letters). Different colors are symbolized by different shades of
gray. max maximum.
1050 HORSTMANN
singleton, they have no control setting to reorient away from the
singleton. In contrast, an unannounced feature change of the sin-
gleton does not capture attention because participants know that
color singletons are nonpredictive and hence ignore them. Expec-
tation is perhaps a superfluous concept to explain the present
results.
The following three experiments were designed to deal with this
contention. In principle, the singleton-capture account can explain
all cases of attentional capture by an unexpected singleton. There-
fore, the experiments focus on conditions in which the singleton-
capture hypothesis, but not the surprise-capture hypothesis, pre-
dicts that the singleton will capture attention in the critical trial.
According to the singleton-capture account, a singleton in the
critical trial should reveal attentional capture unless the precritical
trials misguide the participants in ignoring singletons of any kind.
In Experiment 4, heterogeneous displays are presented in the
precritical trials that do not contain singletons; rather, each display
contains red and green squares that are arranged such that the
colors alternate. Because of the regular arrangement of the two
colors, salience (sensu Wolfe, 1994) should be equal at all posi-
tions. More precisely, because there is a feature contrast for each
of the squares with respect to the neighboring squares, no single
position should attract attention more strongly than any other. As
a consequence, there is no need to suppress an attentional response
to a singleton. If color singletons always capture attention, given
no intention to the contrary, a single red square among green
squares (or vice versa) should capture attention in the critical trial.
The conditions in the present Experiment 4 were optimized to
provide a clear test of a singleton-capture account for the present
experiments; they were less optimal in testing the expectation-
discrepancy account. As a result, the predictions are a bit more
complicated. An expectation-discrepancy account predicts no ini-
tial capture by the singleton, because the singleton's color is
consistent with expectations. However, this does not imply that the
observers will not become aware of the changed composition of
the critical trial display. However, detecting the difference in
composition amounts to comparing specific combinations of fea-
tures, and the detection of this discrepancy requires attention
according to feature integration theory (e.g., Treisman & Gelade,
1980). Thus, the discrepancy will probably be detected only during
serial search for the target letter. However, in this case, it is the
color homogeneity of most items that is expectancy discrepant, and
attending to this aspect of the display does not help in finding the
target, because the homogeneous items are all distractors. In sum,
I predict that the set-size effect in the critical trial will not be
reduced, although the RTs may be longer than in the precritical
trials, because of a detection of the discrepancy.
Method
Participants. Participants were 9 men and 15 women with a mean age
of 24.5 years (SD 5.5). Compensation was 1.
Apparatus, stimuli, procedure, and design. These were the same as in
Experiment 2, except for the composition of the precritical trials. In these
trials, heterogeneous displays were presented in which red and green
squares occupied alternating positions. The two possible presentation po-
sitions for the red and green squares were used equally often. The sequence
of the two positions was randomized. Target position was determined at
random, and color was nonpredictive of the position of the target. The
critical trial and the following 23 postcritical trials were identical to the
corresponding trials of Experiments 2 and 3. That is, there was a singleton
square (either red or green) among the nonsingleton squares at a random
position that always contained the target.
Results
Errors and RTs longer than 3,000 ms were excluded from the
analysis, resulting in a loss of 4.5% of the experimental trials.
Figure 3 (panel c) presents mean RTs; error rates are given in
Table 2. There was a large set-size effect in the precritical trials (81
ms/letter) and in the critical trial (66 ms/letter) but a small set-size
effect in the postcritical trials (16 ms/letter).
An overall ANOVA with the variables set size (4 vs. 12) and
trial type (precritical vs. critical vs. postcritical) revealed signifi-
cant effects of set size, F(1, 22) 59.1, p .01, and of trial type,
F(2, 44) 17.1, p .01, and a significant Set Size Trial Type
interaction, F(2, 44) 10.0, p .01. A corresponding error
analysis revealed a significant trial type effect, F(2, 44) 9.0, p
.01, indicating that participants traded speed for accuracy in the
critical trial, in which no errors occurred, relative to the precritical
and the postcritical trials (see Table 2). However, the main effect
for trial type is not of prime interest for the present examination.
The main test concerned whether the effect of set size was
different for precritical trials versus the critical trial. A 2 (set size:
4 vs. 12) 2 (trial type: precritical vs. critical trial) ANOVA
revealed main effects of set size, F(1, 22) 56.0, p .01, and trial
type, F(1, 22) 5.2, p .05. The Set Size Trial Type
interaction was not significant (F 1).
A corresponding ANOVA on the critical trial versus the post-
critical trials tested whether the set-size effect in the critical trial
was larger than in the postcritical trials. The ANOVA revealed
significant main effects of set size, F(1, 22) 21.4, p .01, and
trial type, F(1, 22) 7.9, p .01, and a significant Set Size
Trial Type interaction, F(1, 22) 8.8, p .01. That is, search in
the critical trial was not as efficient as in the postcritical trials.
As before, the trials immediately following the critical trial were
analyzed, and, similar to the prior experiments, the set-size effect
declined rapidly; it was 66 ms/item in the critical trial and 30, 19,
and 13 ms/item in the following three trials.
Discussion
The first presentation of a singleton in a visual-search task did
not capture attention if this presentation was preceded by hetero-
geneous displays without a singleton. Because of the construction
of the heterogeneous displays, there is no reason to assume that
participants were set to ignore singletons. The experiment thus
indicates that a singleton-capture account of attentional capture by
surprising color singletons is difficult to defend. Conversely, the
results are consistent with an expectation-discrepancy account of
surprise capture: If the first presentation of a singleton is preceded
by heterogeneous displays containing the very two colors pre-
sented in the critical trial, expectation deviation of the singleton's
color should be zero.
As outlined in the introduction, a central assumption of the
present model of surprise capture is that an expectancy discrep-
ancy can be detected either preattentively or postattentively, de-
pending on whether the expectation pertains to a simple feature or
a feature combination. Accordingly, the discrepancy of the critical
1051CAPTURE AND EXPECTATION DISCREPANCY
trial display cannot be determined preattentively in this experi-
ment. However, after attention had been deployed to some of the
stimuli in the display, the discrepancy should have become avail-
able. Consistent with this reasoning, most of the participants
apparently used the singleton to locate the target in the trials
immediately following the critical trial. It remains an open ques-
tion, however, as to at what point in time the discrepancy was
detected during the critical trial. Further experiments, possibly
using eye movements as a more direct indicator of movements of
attention during the critical trial, are required to answer this
question.
Experiment 5: Homogeneous Displays of Varying Colors
Experiment 4 indicates that the presentation of a singleton was
not sufficient to capture attention, even if the observers had no
intention of ignoring the singleton. This result is consistent with
the hypothesis that it is mainly the expectation of (task-irrelevant)
color variation that eliminates surprise capture by the new color.
This assumption suggests that color variation should eliminate
surprise capture by an unannounced singleton even if the critical
trial is preceded by homogeneous displays only, given that the
homogeneous displays presented in different trials vary in color.
Therefore, in Experiment 5 the precritical trials were composed of
red and green homogeneous displays that were presented in a
random sequence. The critical and the postcritical trials were the
same as in the previous experiment.
A singleton-capture account predicts that the singleton captures
attention because it is a singleton and the observers are not set to
reorient away from a singleton (because no singleton was pre-
sented in the precritical trials). In contrast, an expectancy-
discrepancy account predicts no capture of attention by the single-
ton, because the singleton's' color is consistent with expectations.
However, because the new combination of colors may be detected
after attention is focused to the singleton's position in the course of
the visual-search task, RTs in the critical trial may be longer than
in the precritical trials.
Method
Participants. Participants were 7 men and 19 women with a mean age
of 23.8 years (SD 5.6). Compensation was 1.
Apparatus, stimuli, procedure, and design. The only difference from
Experiment 4 concerned the composition of the precritical trials, which
comprised red and green homogeneous displays that appeared in a ran-
domly determined order.
Results
Errors and RTs longer than 3,000 ms were excluded from
analysis, which resulted in a loss of 3.5% of the experimental
trials, two in the critical trials, and reduced the number of partic-
ipants to 24, 12 per set-size condition. Figure 3 (panel d) shows the
mean RTs; Table 2 displays the corresponding error rates. There
was a large set-size effect in the precritical trials (80 ms/letter) and
in the critical trial (90 ms/letter) but only a small set-size effect in
the postcritical trials (8 ms/letter).
An overall ANOVA of the RTs with the variables' set size (4 vs.
12) and trial type (precritical vs. critical vs. postcritical) revealed
significant effects of set size, F(1, 22) 25.5, p .01, and of trial
type, F(2, 44) 30.5, p .01, and a significant Set Size Trial
Type interaction, F(2, 44) 15.9, p .01. A corresponding error
analysis revealed no significant main effects or interactions (Fs
1.2). Mean error rates appear in Table 2.
The main prediction concerned whether the effect of set size
was different for precritical trials versus the critical trial. A 2 (set
size: 4 vs. 12) 2 (trial type: precritical vs. critical trial) ANOVA
revealed a main effect of set size only, F(1, 22) 30.8, p .01.
The main effect of trial type, F(1, 22) 1.6, and the Set Size
Trial Type interaction (F 1) were not significant.
Conversely, the set-size effect was different for the critical trial
and the postcritical trials. A 2 (set size: 4 vs. 12) 2 (trial type:
critical vs. postcritical) ANOVA revealed significant main effect
of set size, F(1, 22) 24.1, p .01, and trial type, F(1, 22)
51.9, p .01, and a significant Set Size Trial Type interaction,
F(1, 22) 47.2, p .01.
As before, the trials immediately following the critical trial were
inspected. The set-size effect declined rapidly in these trials; it was
90 ms/item in the critical trial and 10, 13, and 10 ms/item in the
following three trials.
Discussion
The present experiment provides convergent evidence that pre-
senting a singleton in the absence of an intention to ignore single-
tons is not sufficient to elicit attentional capture. Only homoge-
neous displays were presented in precritical trials, such that there
was no reason whatsoever to adopt any set toward singletons.
Nevertheless, the singleton did not capture attention.
The present results are consistent with an expectation-
discrepancy account of surprise capture. Because more than one
color was used for the homogenous displays, the singleton's color
in the critical trial was consistent with expectations. Again, this
does not mean that the presentation of the singleton in the critical
trial following homogeneous displays was not at all expectation
discrepant. However, the present results indicate that this latter
kind of expectation discrepancy (i.e., concerning the composition
of the display by schema-congruent elements) either was very
weak or was detected only after attention had been allocated to the
singleton's location. The latter alternative is quite plausible, given
that the conjunction of features is achieved by focused attention
(e.g., Treisman & Gelade, 1980). This implies that unless attention
is focused on the location, each color is evaluated independently
with respect to its expectation consistency. Consistent with these
assumptions, the singleton was used by most or all participants
even in the trial that followed the critical trial, as indicated by the
very small set-size effect in this trial.
Experiment 6: Single Color Change
Experiments 4 and 5 yield evidence that a singleton that does
not deviate from expectations does not capture attention, even if
the participants are not set to ignore singletons. Experiment 6 is an
extension of Experiment 5 in that it changes (only) one aspect--
that is, the homogeneous displays in the precritical trials were
presented not in a random sequence but blockwise. For example,
Trials 1­24 were homogeneously green, Trials 25­48 were homo-
geneously red, and the critical trial presented a green singleton
among red nonsingleton stimuli. With this change, the degree of
1052 HORSTMANN
color variation necessary to induce an expectation of color varia-
tion is tested. When color changes only once, there is obviously a
very moderate manipulation of an expectation of color variation,
which thus tests a lower limit for the reported effect. For this
reason, Experiment 6 is more exploratory in nature than the pre-
ceding experiments: Given that relatively little is known about
expectations, it may be that the more recent trials (Trials 25­48)
have more impact than the earlier trials (Trials 1­24). Conversely,
there was little more than 1 min between Trial 24 and the critical
trial (Trial 49), which suggests that the earlier trials may still
influence the content of expectations.
Experiment 6 also provides an additional test of the current
model for the RTs in the critical trial. To recapitulate, I assume that
the RTs in the critical trial can be analyzed into three additive
components: (a) the time to find the target, (b) decision-level
processing of the surprising event, and (c) S-R translation pro-
cesses that eventually lead to the response. On the basis of the
present account of surprise capture and previous results, I assume
that the color change between Trials 1­24 and Trials 25­48 elicits
surprise, because the expectation primarily concerns color (e.g.,
present Experiments 3­5), and the change is preceded by homo-
geneous displays of only one color, which has proved to be a
condition in which surprise is reliably elicited (e.g., present Ex-
periment 1; Horstmann, 2002, 2004, in press). I therefore predict
that the decision-level processing of the surprising event will result
in a prolongation of the RTs that is independent of set size.
However, with a color change of the whole display, attending to
the changed color does not help in finding the target. Therefore,
the target can still only be found with serial search, as in the
preceding trials. Thus, the set-size effect present in Trials 1­24
should also be present in Trial 25. Because of the additivity of
these two components, I predict a general increase of RT for Trial
25, which does not interact with set size.
Method
Participants. Participants were 11 men and 18 women with a mean age
of 25.8 years (SD 9.2). Compensation was 1.
Apparatus, stimuli, procedure, and design. The only difference from
Experiment 5 was that red and green homogeneous displays were presented
blocked, with one color used in Experimental Trials 1­24 and the other
color in Experimental Trials 25­48. The color in the 12 practice trials was
the same as in Trials 1­24. The nonsingleton color in the 24 heterogeneous
display trials (critical and postcritical trials) was the same as in the
immediately preceding trials (Trials 25­48).
Results
Errors and RTs longer than 3,000 ms were excluded from
analysis, resulting in a loss of 4.6% of the experimental trials, one
in the critical trial, and reducing the number of participants to 28,
14 per set-size condition. Figure 3 (panel e) shows the RT data for
the two set-size conditions in the precritical, critical, and postcriti-
cal trials. The corresponding error rates are listed in Table 2. There
was a large set-size effect in the precritical trials (74 ms/letter) and
in the critical trial (62 ms/letter) but only a small set-size effect in
the postcritical trials (6 ms/letter).
Color change within the precritical trials. A 2 (set size: 4 vs.
12) 2 (trial: Trial 1­24 vs. Trial 25 vs. Trial 26­48) ANOVA
was conducted to test whether the first presentation of a homoge-
neous display of a different color elicited surprise. (Because two
errors occurred in the 25th trial, the analysis was conducted with
only 26 participants.) The analysis revealed significant main ef-
fects of set size, F(1, 24) 40.9, p .01, and trial type, F(2,
48) 14.0, p .01, but no significant Set Size Trial Type
interaction (F 1). A corresponding error analysis yielded no
significant effects (Fs 2.7, ps .10). Planned comparisons
revealed that the mean RT--collapsed over the set-size variable--
was longer in Trial 25 (1,282 ms) than in Trials 1­24 (944 ms),
t(25) 4.1, p .01, and longer than in Trials 26­48 (976 ms),
t(25) 3.5, p .01.
Critical trial. An overall ANOVA with the variables set size
(4 vs. 12) and trial type (precritical vs. critical vs. postcritical)
revealed significant effects of set size, F(1, 26) 22.6, p .01,
and of trial type, F(2, 52) 25.4, p .01, and a significant Set
Size Trial Type interaction, F(2, 52) 6.3, p .05. A
corresponding error analysis revealed no significant main effects
or interactions (Fs 1.2). Mean error rates appear in Table 2.
The main test concerned whether the effect of set size was
different for precritical trials versus the critical trial. A 2 (set size:
4 vs. 12) 2 (trial type: precritical vs. critical trial) ANOVA
revealed main effects of set size, F(1, 26) 23.1, p .01, and trial
type, F(1, 26) 4.7, p .05, whereas the Set Size Trial Type
interaction was not significant (F 1). The effect of trial type
reflected the surprise-induced RT increase, with RTs being 214 ms
longer in the critical trial than in the precritical trials.
However, the set-size effect was different for the critical trial
and the postcritical trials. A 2 (set size: 4 vs. 12) 2 (trial type:
critical vs. postcritical) ANOVA revealed significant main effects
of set size, F(1, 26) 6.5, p .05, and trial type, F(1, 26) 35.5,
p .01, and a significant Set Size Trial Type interaction, F(1,
26) 5.4, p .05.
As before, the trials immediately following the critical trial were
inspected. The set-size effect declined rapidly in these trials; it was
62 ms/item in the critical trial and 3, 2, and 3 ms/item in the
following three trials.
Discussion
Experiment 6 reveals three interesting results. First and most
important, the results are at odds with a singleton-capture account
that predicts attentional capture by a salient singleton given that
observers do not intend to ignore it. In particular, although there
was no reason for participants to intentionally ignore a singleton
because they never saw a singleton in the precritical trials, no
attentional capture by the singleton in the critical trial was ob-
served. Results are thus more consistent with an expectation-
mismatch explanation of surprise capture. According to this ac-
count, the change of color of the homogeneous displays during the
precritical trials disconfirmed the observers' original expectation
concerning color and led them to broaden the range of expected
colors.
Second, it is remarkable that the acceptability of both presented
colors persisted for the 24 trials that lay between the color change
within the homogeneous displays and the critical trial. This result
supports an assumption underlying the present experiments: Sur-
prise can be induced only once via one type of surprising event
(e.g., a color change), because after the surprising event, expect-
ancies are changed (e.g., Schu¨tzwohl, 1998).
1053CAPTURE AND EXPECTATION DISCREPANCY
Third, the color change within the precritical trials resulted in
longer RTs as compared with the preceding and the following
trials, with the RT increase being additive to the set-size effect.
This supports the RT model assumed for the present paradigm, in
particular the assumption that the distraction responsible for the
RT increase is independent of set size and thus additive to the other
component processes of RT in the critical trial. In particular, the
new color in the 25th trial mismatched expectations, and this
incidence disrupted search and slowed RT. Because the expect-
ancy discrepancy concerned the display as a whole and did not
single out the position of the target, the set-size effect was not
affected. Rather, participants had to resume the search for the
target after the surprising color change had been processed.
Experiment 7: Form-Heterogeneous Nonsingleton
Precritical Trials
Experiment 7 was designed to control for the possibility that the
absence of attentional capture by the color singleton in Experi-
ments 4­6 was due to unspecific heterogeneity of the displays in
the precritical trials rather than to color heterogeneity. This pos-
sibility was already rendered implausible by Experiment 3. Exper-
iment 3 revealed that presenting a form singleton in the precritical
trials did not eliminate attentional capture by an unannounced
color singleton in the critical trial, which implies that the expec-
tation of variability does not strongly generalize over dimensions.
However, attentional capture was not as strong in that experiment
as in the experiments with entirely homogeneous displays in the
precritical trials, as indicated by a significant difference of the
set-size effects in the critical and the postcritical trials. To safe-
guard the conclusion that it is not heterogeneity per se that elim-
inates capture to an unannounced color singleton, I conducted the
present Experiment 7. It combines features of Experiments 3 and
4, in that each display in the precritical trials was orientation
heterogeneous, but unlike Experiment 3 the orientation-
heterogeneous displays did not contain singletons. Rather, analo-
gous to Experiment 4, the two orientations of the squares alter-
nated in adjacent positions in a checkerboard fashion. The critical
trial was the same as in Experiments 3­6--that is, all squares had
the same orientation, but one had a different color.
Method
Participants. Participants were 14 men and 15 women with a mean age
of 27.1 years (SD 8.1). Compensation was 1.
Apparatus, stimuli, procedure, and design. These were the same as in
the preceding experiment except for the composition of the precritical
trials. In these trials, upright-oriented and 45°-rotated squares alternated in
adjacent positions.
Results
Errors and RTs longer than 3,000 ms were excluded from RT
analysis, resulting in the loss of 4.7% of the trials, one of them a
critical trial. Mean RTs and error percentages are depicted in
Figure 3 (panel f) and in Table 2, respectively. The set-size effect
in the critical trial (28 ms/letter) was reduced relative to the
precritical trials (73 ms/letter) but was somewhat larger than in the
postcritical trials (3 ms/letter).
An overall ANOVA of the RTs yielded a significant main effect
of trial type, F(2, 52) 28.3, p .01, and set size, F(1, 26)
17.7, p .01, and a significant Trial Type Set Size interaction,
F(2, 52) 9.8, p .01. A corresponding error analysis yielded a
significant main effect for block only, F(2, 52) 11.3, p .01
(the other Fs 1). That is, participants traded accuracy for speed
in the critical trial (see Table 2). However, as in the previous
experiments, such a trade-off is of minor importance for the
present investigation, because the main prediction concerns the
Trial Type Set Size interaction.
An analysis including the precritical trials and the critical trial
yielded a significant main effect of set size, F(1, 26) 21.1, p
.01, and a significant Trial Type Set Size interaction, F(1, 26)
5.0, p .05, reflecting a smaller set-size effect in the critical trial
than in the precritical trials (224 ms vs. 585 ms). The main effect
for trial type was not significant, F(1, 26) 1.2.
When the analysis considered the critical trial and the postcriti-
cal trials, only a significant main effect for trial type was obtained,
F(1, 26) 38.7, p .01, whereas the main effect of set size and
the Set Size Trial Type interaction were not significant (Fs
1.9).
Discussion
Presenting form-heterogeneous nonsingleton displays in the pre-
critical trials gave results where the set-size effect reduction in the
critical trial was between completely homogeneous displays (Ex-
periment 1, homogeneous display condition) and form-
heterogeneous displays containing a singleton (Experiment 3).
More precisely, similar to Experiment 2, there was quite a strong
reduction of the set-size effect in the critical trial relative to the
precritical trials and no significant difference of the set-size effects
in the critical trial and the postcritical trials. This result suggests
that for a surprising color singleton to capture attention, color
homogeneity in the precritical trials is of prime importance,
whereas form heterogeneity is not. This result adds to the conclu-
sions drawn from Experiments 1 and 3 that expectations are
relatively specific.
However, as in Experiment 3, the set-size effect in the critical
trial was somewhat larger than in Experiments 1 and 2, suggesting
that form heterogeneity does have some influence (although not
significantly different from the postcritical trials in this case).
Thus, the data indicate that a feature-unspecific expectation of
homogeneity is a second--although apparently less important--
determinant of surprise capture, in addition to the feature-specific
expectation of certain colors.
Experiment 8: Large Set Sizes
Experiment 8 returns to the original manipulation to induce
surprise capture: the presentation of a color singleton after re-
peated presentations of color-homogeneous precritical trials
(present Experiments 1 and 2; Horstmann, 2002). One possible
objection to the previous experiments is that the pattern of results
assumed to be indicative of surprise capture is not as simple as one
might wish. In particular, although the predicted reduction of the
set-size effect was obtained in the critical trial, the RTs in the
critical trial were not always shorter than with serial search in the
precritical trials, a result that is usually obtained with attentional
1054 HORSTMANN
capture. In the current experiments, a reduction in RT was ob-
tained with set-size 12 but not with set-size 4, where RT increased.
This pattern of results was explained with reference to the RT
delay that is typically observed when a surprise stimulus is pre-
sented during an RT task.
If this explanation is valid, one should predict RT benefits in the
critical trial as compared with the serial-search task in the precrit-
ical trials with larger set sizes. That is, if the set sizes are suffi-
ciently large and RTs in the visual-search task are comparably
long, then RTs should be shorter in the critical trial despite the
"handicap" of additional processes concerned with the surprise
stimulus. Thus, with sufficiently large set sizes, it should be
possible to demonstrate an attentional shift to the surprise stimulus,
exhibiting both characteristics of attention capture: a reduction of
the set-size effect and an overall reduction of RTs in both the large
and the small set-size conditions.
Method
Participants. Participants were 25 women and 11 men with a mean age
of 25.3 years (SD 4.3). They were paid 1 for their services.
Apparatus, stimuli, procedure, and design. These were basically the
same as in the previous experiments, except that (a) a different monitor was
used (a 17-in [43.18-cm] Sony), (b) the two set sizes tested contained 25 or
49 stimuli, and (c) the display duration was changed to 10.000 ms to take
into account the increased difficulty of the serial search with the large set
sizes.
Stimuli. Targets and distractors were the same as in the set-size 12
conditions of the previous experiments. In contrast to the previous exper-
iments, however, the 11 distractors appeared several times in each display.
More precisely, the distractor positions were filled by randomly drawing
letters without replacement from sets of 11 distractors. The distractors were
arranged in a 5 5 matrix for the set-size 25 condition and in a 7 7
matrix for the set-size 49 condition. The center-to-center distance of the
letters was 1.5°. Similar to Experiments 1 and 2, the letters were them-
selves colored instead of being presented on colored patches, as in Exper-
iments 3­7. Two color pairs were used: red­green, as in the previous
experiments, and blue­yellow. Half of the participants had the letters
presented in red or green, and the other half in blue or yellow. Which color
of a pair served as the homogeneous display color and which served as the
singleton color were balanced over participants.
Results
Errors and RTs longer than 8,000 ms were excluded from
analysis, which concerned 7.9% of the experimental trials (al-
though none in the critical trial). The RT criterion was adjusted in
this experiment, because, as a result of the large set sizes, RTs
were much longer in the present experiments. The criterion of
8,000 ms was chosen because it excluded about 2% of the RTs,
which corresponds to the 3,000 ms criterion in the previous ex-
periments. Figure 5 shows the mean RTs; error rates are given in
Table 3. There was a large set-size effect in the precritical trials (42
ms/letter) but not in the critical trial (6 ms/letter) or in the post-
critical trials (2 ms/letter; for the RT results, see Figure 5).
An overall ANOVA of the RTs yielded a significant main effect
of trial type, F(2, 52) 91.8, p .01, and set size, F(1, 26) 6.3,
p .05, and a significant Trial Type Set Size interaction, F(2,
52) 9.98, p .01. (Including color pair [i.e., red­green vs.
yellow­blue] in the analysis did not change the results, although
this revealed shorter RTs with the yellow­blue color pair.) A
corresponding error analysis yielded a significant main effect for
block only, F(2, 52) 10.8, p .01 (the other Fs 1). That is,
participants traded accuracy for speed in the critical trial. How-
ever, as in the previous experiments, such a trade-off is of minor
importance for the present investigation, because the main predic-
tion concerns the Trial Type Set Size interaction.
The analysis considering the precritical trials and the critical
trial yielded a significant main effect of set size, F(1, 26) 7.2,
p .01, a significant main effect for trial type, F(1, 26) 37.1,
p .01, and a significant Trial Type Set Size interaction, F(1,
26) 11.0, p .01, reflecting a smaller set-size effect in the
critical trial than in the precritical trials (1,008 ms vs. 138 ms). t
tests revealed that RTs were shorter in the critical trial than in the
precritical trials in the set-size 25 condition, t(13) 2.0, p .05,
and in the set-size 49 condition, t(13) 6.5, p .01 (one-tailed).
When the analysis considered the critical trial and the postcriti-
cal trials, only a significant main effect for trial type was obtained,
F(1, 26) 61.1, p .01, whereas the main effect of set size and
the Set Size Trial Type interaction were not significant (Fs 1).
Discussion
In agreement with the predictions, shorter RTs in the critical
trial than in the precritical trials were obtained when large set sizes
were tested. The remaining pattern of results is consistent with the
previous experiments, revealing a large set-size effect in the pre-
critical trials that was strongly reduced in the critical trial and in
the postcritical trials. The present experiment thus shows that a
"simple" pattern of results indicative of attentional capture (with
Figure 5. Mean correct reaction times (RTs) for both set-size conditions
(set-size 25: diamonds; set-size 49: squares) in the precritical, critical, and
postcritical trials of Experiment 8.
Table 3
Error rates (in percentages) for Experiment 8
Condition Precritical Critical Postcritical
Set size 25 3.3 0.0 3.0
Set size 49 4.6 0.0 2.1
1055CAPTURE AND EXPECTATION DISCREPANCY
both a reduction of the set-size effect and shorter RTs than with
serial search) can be obtained for the first and unannounced
presentation of a color singleton, when the set sizes used are
sufficiently large.
General Discussion
The aim of the present study was to test the conditions of
attentional capture to the first and unannounced presentation of a
singleton during a visual-search task. The central hypothesis was
that attention can be captured by an unannounced stimulus if the
stimulus deviates from an expectation or schema and if the expec-
tation concerns a stimulus aspect that is preattentively available.
This hypothesis was tested with regard to expectations for the
dimension of color. The experiments demonstrated that a singleton
(e.g., a red patch among green patches) attracted attention on its
very first presentation if its color was new and if it followed
precritical trials without color variation (Experiments 1a, 2, 3, 7,
and 8). In contrast, the color singleton did not capture attention if
its color was not new but was repeatedly presented during the
precritical trials (Experiments 4, 5, and 6). This result indicates
that the expectation discrepancy is an important condition of
attentional capture by an unannounced singleton.
Several results indicate that the expectancy discrepancy that
triggers an attentional shift concerns the color of the singleton and
not the presence of the singleton per se. Recall that in most
experiments demonstrating surprise capture (e.g., present Experi-
ments 1, 2, and 8), the singleton in the critical trial is the first
object in the experiment with a given color, and it is the first
singleton presented in the experiment. That is, both aspects could
potentially be the instance that triggers attentional capture. Three
results indicate that the color dimension is more important. First, a
red or green singleton that followed red and green color-
homogeneous displays in the precritical trials (Experiment 5 and 6)
should have been discrepant from an expectation of a homoge-
neous display but not from expectations concerning color, yet the
color singleton in the critical trial did not capture attention. This
indicates that only when the first and unannounced presentation of
a singleton was accompanied by the occurrence of an expectation-
discrepant color did the singleton capture attention. Second, with
the critical trial following form-heterogeneous trials (Experiment 3
and 7), attention was attracted by the color singleton (although
attention capture was not as strong as in the most comparable
experiment, Experiment 2, in which the precritical trials were both
color and form homogeneous; this result is discussed shortly).
Third, the unannounced color change within the (homogeneous)
precritical trials in Experiment 6 induced an RT increase that was
additive to the set-size effect. Although this pattern of results does
not reveal a spatial shift of attention because the new color was an
aspect of the whole display and thus did not single out the target's
location, the RT increase indicates that the color change was
attended to. It might be noted that distraction by a nonpredictive
stimulus is an often used indicator of attentional capture (e.g.,
Theeuwes, 1992, 1994), although RT increases are susceptible to
alternative explanations (e.g., Folk & Remington, 1998). In sum,
evidence for attentional shifts covaried with the color change and
not with changes from homogeneous to heterogeneous conditions.
Apart from testing the conditions of surprise capture, a second
important aim of the present experiments was to test whether
purely bottom-up-driven "singleton capture" can account for the
capture effects in the present paradigm (e.g., Itti & Koch, 2000;
Kim & Cave, 1999; Koch & Ullman, 1985; Theeuwes & Godijn,
2001; Wolfe, 1994). According to this account, singletons always
capture attention, but participants quickly reorient away from the
singleton if they believe that the singleton is task irrelevant. The
results of Experiments 4­6 render this account unconvincing for
the present paradigm. The checkerboard-like heterogeneous pre-
critical trials in Experiment 4 did not contain singletons and thus
cannot have induced an intention to ignore singletons, yet no
attentional capture occurred with the first presentation of a single-
ton in the critical trial. Neither could the red and green homoge-
neous displays in the precritical trials of Experiments 5 and 6
induce an intention to ignore singletons, yet no attentional capture
occurred in the critical trial. In the face of the very consistent
results of Experiments 4­6, attentional capture by an unan-
nounced singleton cannot be accounted for easily by purely
bottom-up-driven singleton capture.
The present results allow some generalizations on the expecta-
tion mismatch that is assumed to enable or disable attentional
capture in the present paradigm. First, expectations could be spe-
cific with respect to feature dimensions (e.g., color or form), as
indicated by the fact that expecting form heterogeneity did not
eliminate attentional capture by a color singleton in Experiments 3
and 7. Second, expectations could be quite specific with respect to
feature values (e.g., that all stimuli are red), given that one feature
value was consistently presented in the precritical trials. In con-
trast, if several feature values (e.g., red and green) were presented
in the precritical trials, expectancies were broad and fuzzy within
the respective feature dimension (cf. Schu¨tzwohl, 1998). Conse-
quently, the introduction of a third color did not capture attention
(Experiment 1A, heterogeneous display condition, and Experiment
1B). This result is reminiscent of Schu¨tzwohl's (1998) finding that
an unannounced change in the physical appearance of a word was
less surprising (as indicated by the amount of RT interference) if
the word's font varied slightly in the precritical trials, as compared
with a constant font in the precritical trials. Third, following even
a single color change within homogeneous displays, the expected
range of colors remained sufficiently broad over quite a number of
trials that a singleton instantiating one of the previously presented
colors did not capture attention. Fourth, expectations that led to
surprise capture in the critical trial probably concerned basic
features but not combinations of basic features (cf. Treisman &
Gelade, 1980). This is not to say that combinations of features
cannot be expectation discrepant; rather, this discrepancy would
not be detected preattentively.
The present results support the notion that the detection of an
expectancy discrepancy (including attention to the surprising ob-
jects) can result from preattentively and from attentively processed
information. Because preattentive analysis provides separated but
not integrated information (Treisman & Gelade, 1980), surprise
capture can occur only if the presence of a feature is schema
discrepant with respect to any part of the schema. That is, after
repeated presentations of a green square and a green circle, a red
square is discrepant because none of the elements of the schema is
assumed to be red. In contrast, after repeated presentations of a
green square and a red circle, a green circle is not schema discrep-
ant on the basis of preattentively delivered feature information,
because the color green is consistent with elements of the schema,
1056 HORSTMANN
and no surprise capture should occur. However, after the green
circle has been attended to, the combination of form and color can
be surprising and promote more attentive processing.
Some Methodological Considerations
As repeatedly noted before, other research on attentional capture
with unannounced color singletons indicates that the attentional
shift has a latency of about 300 ms. That is, within the present task,
participants probably had already searched through one or two
item positions when their attention was drawn to the color single-
ton. This implies that, in the present task, the color singleton was
occasionally found during search and not by attentional capture.
Further, this occurred more often with set-size 4 than with set-size
12. In the following, I discuss what implications this possibility
has for the interpretation of the data, and I substantiate why this
does not alter the implications of the present experiments.
A situation in which two independently running processes com-
pete for the determination of some result is usually modeled as a
race (e.g., Horstmann, 2003; Logan, 1994; Logan & Cowan,
1984). For the present situation, search processes and discrepancy-
detection processes run in parallel and compete for the detection of
the singleton. The search processes proceed serially, and the prob-
ability of finding the singleton increases with time. In contrast, the
discrepancy-detection processes proceed in parallel and have a
fixed singleton-detection time of about 300 ms. The time for the
detection of the singleton in the critical trial is determined by the
winner of the race, that is, by the process that succeeds first in
detecting the singleton. Given that both--the surprise and the
search processes--detect the singleton in a portion of the critical
trials, the mean detection time is the weighted mean of the detec-
tion times of both processes. Because the probability that the
singleton will be detected during search is higher with set-size 4
than with set-size 12, the weight given to the finishing time of the
search processes is higher with set-size 4 than with set-size 12
(and, accordingly, the weight given to the finishing time of the
discrepancy-detection processes is lower).
As a consequence, the mean singleton-detection time, and there-
fore the mean RT in the critical trial, is more affected by trials on
which the singleton is found through search with set-size 4 than
with set-size 12. The important point is that the mean singleton-
detection time, and thus the mean RT in the critical trial, is shorter
the more probable a detection through search is. This is because
serial search beats discrepancy detection in the race only if it is
faster.
For this reason, the possibility that the singleton was found
during search, which is more probable in the set-size 4 than in the
set-size 12 condition, cannot explain the reduction of the set-size
effect in the critical trial. On the contrary, it can only explain the
presence of a set-size effect with empirically obtained RTs, where,
theoretically--on the basis of the assumption that mean RT is
determined solely by discrepancy-detection processes but not by
serial search in the critical trial--no set size effect is predicted.
Thus, this possibility works against the presently tested hypothesis.
Alternative and Complementary Accounts
It may appear that habituation/sensitization processes can ex-
plain the presence and absence of surprise capture. That is, habit-
uation of the response to the nonsingleton color could have oc-
curred because of its repeated presentations, whereas sensitization
of the response to the singleton color could have occurred because
of the long time during which it was not used. This hypothesis,
however, is difficult to reconcile with Experiment 6, because the
critical trial was preceded by 24 homogeneous trials with a con-
stant color, which should be sufficient to enable habituation/
sensitization effects. That is, although it is possible that habitua-
tion/sensitization processes add to surprise capture in some
instances, they are probably not the main mechanism.
The present investigation, in particular with respect to some of
the reported effects, is reminiscent of the novel pop-out hypothesis
that was discussed in the 1990s. According to this hypothesis,
attention is rapidly and automatically drawn to a single novel item
within a display of familiar items (e.g., Johnston, Hawley, Plewe,
Elliott, & DeWitt, 1990; for a critical review see Christie & Klein,
1996). It is clear that the novel color in the critical trial presented
along with familiar colors could be an example of novel pop-out,
according to this definition. However, there are a number of
differences from the research program of novel pop-out, both in
method and in theory.
First, Johnston et al. (1990) used letter strings in their experi-
ments that were either familiar in the experimental context (pre-
sented repeatedly during the experiment) or novel (presented for
the first time). Four items were presented on each trial that could
be either (a) all familiar, (b) all novel, or (c) mixed, with three
familiar and one novel item. This four-word display was replaced
by a probe display, in which one of the previously presented four
words was repeated at the four positions that were previously
occupied by the words in the four-word display (an intermediate
mask of "XXXXXX" intervened between the four-word display and
the probe display). The task was to indicate the position of the
probe word in the four-word display. Novel pop-out was inferred
from a localization advantage for the novel word ("the novel
pop-out effect") and a localization disadvantage for the familiar
words ("familiar sink-in"). A methodological and a corresponding
theoretical difference from the present research are immediately
obvious. In the novel pop-out paradigm, complex stimuli are used,
and it is claimed that the unfamiliarity of complex stimuli is
assessed preattentively, which is a controversial position. In con-
trast, the present experiments used simple stimulus features (e.g.,
color) for which preattentive processing is well documented in the
visual search literature (e.g., Yantis & Egeth, 1999). Furthermore,
although the present model assumes that attention is generally
attracted by expectancy-discrepant events, attention capture as a
spatial shift of attention following only a preattentive analysis is
assumed to be possible only for discrepant simple features, not for
complex configurations of features.
Second, the novel pop-out paradigm was designed specifically
to test novel pop-out and has not been used by other researchers
interested in attentional capture. In fact, as Christie and Klein
(1996) pointed out, the particularities of the procedure used to
assess novel pop-out render an unambiguous inference for fast
attentional capture difficult. In contrast, the present experiments
used one of the standard procedures to assess attentional capture.
It might be noted that Christie and Klein's suggestion to avoid
interpretational problems by using a probe task "that does not
entail the use of identity or other information stored in memory
about the probed items" (p. 207) was implemented in the present
1057CAPTURE AND EXPECTATION DISCREPANCY
experiments. That is, the probe task (H vs. U discrimination) did
not require the use of identity or other information about the
probed items (the color patches).
Third and finally, the most compelling evidence for novel pop-
out came from experiments that carried a confound: Because in
these experiments the novel words were more likely the targets
than the familiar words, attending to the novel items would be
advantageous, thus undermining the claim of automatic capture. In
contrast, the letter-search task was completely unrelated to the
color of the patches in the precritical trials of the present
experiments.
Surprise has been framed within a schema-theoretic frame-
work (cf. Neisser, 1976; Rumelhart, 1984; Rumelhart, Smolen-
sky, McClelland, & Hinton, 1986). According to this account,
perception is essentially hypothesis testing. For example, Neis-
ser's (1976) perceptual-cycle hypothesis assumes that observers
have schemas or expectations for what belongs to the scene
(i.e., which objects with what characteristics should be present),
and these expectations are used to guide attention. Attention
then picks up information from the scene, which, in turn,
fleshes out or modifies the schema. That is, perception is
viewed as a cyclic process of hypothesis building and testing.
Neisser proposed that the incorporation into the perceptual
cycle is essential for the information to be "seen" (see also Di
Lollo, Enns, & Rensink, 2000; Most & Simons, 2001). That is,
although some information outside the perceptual cycle can
give rise to an orienting response, this information cannot
acquire identity or meaning unless it is incorporated into the
cycle of anticipation and information pick-up.
This model implies a fundamental difference between expected
and unexpected events. Taking expected stimuli into consideration
first, the organism is prepared to process them so that the meaning
and the task relevance of each stimulus are readily available (see
also Meyer et al., 1997). In contrast, the observer is not prepared
to process unexpected stimuli. Rather, these stimuli have to be
incorporated in a new perceptual cycle. What, then, determines
whether they are incorporated? As I have said, one possibility is
that certain stimulus conditions kindle an orienting response that
triggers a new perceptual cycle. Among these conditions are prob-
ably high-intensity stimuli (loud noise, bright light, heavy touch)
and perhaps the appearance of new objects and highly salient
stimuli (cf. Most & Simons, 2001; remember, however, that the
present results are at odds with the view that salience alone is
sufficient to capture attention).
From the present theoretic viewpoint, experiments on sur-
prise capture suggest a complement to this model: that expec-
tation discrepancies are another condition to interrupt the on-
going perceptual cycle and instigate a new one. Note that
schema discrepancy is apparently sufficient to trigger a new
perceptual cycle: In Experiment 6, the color change between
homogeneous displays was quite a strong distractor from the
primary task activity (as indicated by the RT increase), although
the new color was not a salient event but solely expectation
discrepant. For the goal of extending the model, it suffices to
assume that the testing of expectancies is not necessarily con-
fined to attention but can proceed autonomously on both pre-
attentive and attentive information.
Should the Attentional Shift to the Unannounced Singleton
Be Called Attentional Capture?
Attentional capture and intentional orienting have inherited the
meaning components of their respective parent concepts, being
automatic versus voluntary processes (Jonides, 1981). That is,
attentional capture is characterized by its independence from in-
tentions, by its freedom from central capacity limitations, and by
its reliance on nonconscious processing (Posner & Snyder, 1975).
Speed is also often considered an important characteristic of au-
tomatic processes. The attentional shift to the unannounced sin-
gleton is independent from intentions by definition, in the sense
that the capturing item is not related to the task performed by the
observer (e.g., Yantis, 1993). Although the present experiments do
not inform about the issue of conscious processing, the attention
shift does not appear to be strongly dependent on central capacity
mechanisms, as indicated by the rather weak influence of set size
on its latency. A problematic aspect is probably speed or, rather,
the lack of it. Experiments using the accuracy paradigm (Horst-
mann, 2002, 2004, in press) have characterized the effect as slower
than, for example, the fast attentional shift to a singleton expected
to be predictive of the target.3
If speed is considered the most
important criterion for attentional capture, then an attentional shift
to an unannounced singleton will not be regarded as an instance of
attentional capture. However, I argue that the most important
criterion for attentional capture is that it occurs independently of
the intentions in the sense that the capturing item is independent
from the task performed by the observer (e.g., Yantis, 1993), and
this clearly holds for the attention shift to the new color presented
without prior warning.
Final Remarks
Expectation-discrepant events in the real world are certainly not
always as simple as in the present experiments. Thus, surprise
capture reveals a boundary condition of the working of a more
general mechanism that orients attention to schema-discrepant
stimuli. Frequently, however, expectation discrepancies are de-
tected after attention has been directed to various objects in the
scene in the course of schema testing or task pursuance. More
precisely, schema discrepancies often are defined not by preatten-
tively available basic features but rather by the identity or
category-membership information that is available only after the
deployment of attention. Readers should bear this in mind when
generalizing the present results.
The present results also suggest a new look at the classical
dichotic-listening experiments of Cherry (1953). Cherry presented
participants with different auditory streams at each ear. The task
was to shadow one of the streams, which always contained speech,
while ignoring the other stream. When both streams contained
speech, participants had some knowledge of the shadowed stream
but very poor knowledge of the ignored stream. Cherry conducted
3
One may even suggest that the latency of surprise capture is in the
magnitude typical for endogenous shifts of attention. However, this hy-
pothesis is difficult to reconcile with the fact that an endogenous shift
presumes a reason, and because the new color is unrelated to the task, there
is no reason to attend to it. For a more thorough discussion of this point, see
Horstmann (in press).
1058 HORSTMANN
several experiments to explore what kinds of changes in the
ignored stream participants attended to. He found that physical
changes of the material, such as a change from a male to a female
voice or the change from speech to a pure 400-Hz tone were
recognized by nearly all of the participants. In contrast, reversing
the male speech (an operation that retains spectrum) or switching
to another language were rarely discerned. Because Cherry tested
naive participants who were not informed about possible changes,
there are fairly strong parallels between Cherry's and the present
experiments, in that stimuli that broadly shared physical properties
with the known properties of the to-be-ignored stream did not
receive attention (language change or speech reversal), whereas
stimuli that deviated strongly from the anticipated physical prop-
erties apparently captured attention (pitch change introduced by
the female speaker or pure tone). This suggests a common expla-
nation for these two phenomena from the visual and the auditive
domain: A mismatch between expectations and basic feature prop-
erties promotes an orientation of attention toward the unexpected
stimulus.
References
Ansorge, U., & Heumann, M. (2003). Top-down contingencies in periph-
eral cuing: The roles of color and location. Journal of Experimental
Psychology: Human Perception and Performance, 29, 937­948.
Bacon, W. F., & Egeth, H. E. (1994). Overriding stimulus-driven atten-
tional capture. Perception & Psychophysics, 55, 485­496.
Cherry, E. C. (1953). Some experiments on the recognition of speech, with
one and with two ears. Journal of the Acoustical Society of America, 25,
975­979.
Christie, J., & Klein, R. M. (1996). Assessing the evidence for novel
popout. Journal of Experimental Psychology: General, 125, 201­207.
Darwin, C. (1872). The expression of the emotions in man and animals.
London: Murray.
Di Lollo, V., Enns, J. T., & Rensink, R. A. (2000). Competition for
consciousness among visual events: The psychophysics of reentrant
visual processes. Journal of Experimental Psychology: General, 129,
481­507.
Folk, C. L., & Remington, R. (1998). Selectivity in distraction by irrelevant
featural singletons: Evidence for two forms of attentional capture. Jour-
nal of Experimental Psychology: Human Perception and Performance,
24, 847­858.
Folk, C. L., Remington, R. W., & Johnston, J. C. (1992). Involuntary
covert orienting is contingent on attentional control settings. Journal of
Experimental Psychology: Human Perception and Performance, 18,
1030­1044.
Folk, C. L., Remington, R. W., & Johnston, J. C. (1993). Contingent
attentional capture: A reply to Yantis (1993). Journal of Experimental
Psychology: Human Perception and Performance, 19, 682­685.
Gibson, B. S., & Jiang, Y. (1998). Surprise! An unexpected color singleton
does not capture attention in visual search. Psychological Science, 9,
176­182.
Gibson, B. S., & Kelsey, E. M. (1998). Stimulus-driven attentional capture
is contingent on attentional set for displaywide visual features. Journal
of Experimental Psychology: Human Perception and Performance, 24,
699­706.
Gibson, B. S., & Peterson, M. A. (2001). Inattentional blindness and
attentional capture: Evidence for attention-based theories of visual sa-
lience. In C. Folk & B. Gibson (Eds.), Attraction, distraction, and
action: Multiple perspectives on attentional capture (pp. 51­76). New
York: Elsevier Science.
Horstmann, G. (2001). Die Unterbrechungsfunktion der U¨ berraschung: Ein
neues experimentelles Paradigma und ein Test der Automatizita¨tshy-
pothese. [Unvero¨ffentlichte Dissertation, Universita¨t Bielefeld.] [The
interrupt-function of surprise: A new experimental paradigm and a test
of the automaticity hypothesis]. [Unpublished doctoral dissertation,
Bielefeld University]. Universita¨t Bielefeld.
Horstmann, G. (2002). Evidence for attentional capture by a surprising
color singleton in visual search. Psychological Science, 13, 499­505.
Horstmann, G. (2003). The psychological refractory period of stopping.
Journal of Experimental Psychology: Human Perception and Perfor-
mance, 29, 965­998.
Horstmann, G. (2004). Predictive and non-predictive surprising singletons.
Manuscript submitted for publication.
Horstmann, G. (in press). The time course of intended and unintended
allocation of attention. Psychologische Forschung.
Horstmann, G., & Ansorge, U. (in press). Attentional capture by rare
singletons. Visual Cognition.
Horstmann, G., & Schu¨tzwohl, A. (1998). Zum Einfluss der Verknu¨p-
fungssta¨rke von Schemaelementen auf die Sta¨rke der U¨ berraschung-
sreaktion [Strength of interconnections between schema elements and
intensity of the emotion of surprise]. Zeitschrift fu¨r Experimentelle
Psychologie, 45, 203­217.
Itti, L., & Koch, C. (2000). A saliency-based search mechanism for overt
and covert shifts of visual attention. Vision Research, 40, 1489­1506.
Johnston, W. A., Hawley, K. J., Plewe, S. H., Elliott, J. M. G., & DeWitt,
M. J. (1990). Attention capture by novel stimuli. Journal of Experimen-
tal Psychology: General, 119, 397­411.
Jonides, J. (1981). Voluntary versus automatic control over the mind's
eye's movement. In J. Long & A. D. Baddeley (Eds.), Attention &
performance IX (pp. 187­203). Hillsdale, NJ: Erlbaum.
Kim, M. S., & Cave, K. R. (1999). Grouping effects on spatial attention in
visual search. Journal of General Psychology, 126, 326­352.
Koch, C., & Ullman, S. (1985). Shifts in selective visual attention: To-
wards the underlying neural circuitry. Human Neurobiology, 4, 219­
227.
Krummenacher, J., Mu¨ller, H. J., & Heller, D. (2002). Visual search for
dimensionally redundant pop-out targets: Parallel-coactive processing of
dimensions is location specific. Journal of Experimental Psychology:
Human Perception and Performance, 28, 1303­1322.
Logan, G. D. (1994). On the ability to inhibit thought and action. A users'
guide to the stop signal paradigm. In D. Dagenbach & T. H. Carr (Eds.),
Inhibitory processes in attention, memory, and language (pp. 189­239).
San Diego, CA: Academic Press.
Logan, G. D., & Cowan, W. B. (1984). On the ability to inhibit thought and
action: A theory of an act of control. Psychological Review, 91, 295­
327.
Mack, A., & Rock, I. (1998). Inattentional blindness. Cambridge, MA:
MIT Press.
Mandler, G. (1985). Cognitive psychology: An essay in cognitive science.
Hillsdale, NJ: Erlbaum.
Mandler, J. M. (1984). Stories, scripts, and scenes: Aspects of schema
theory. Hillsdale, NJ: Erlbaum.
Meyer, W.-U. (1988). Die Rolle von U¨ berraschung im Attributionsproze
[The role of surprise in the attribution process]. Psychologische Rund-
schau, 39, 136­147.
Meyer, W.-U., & Niepel, M. (1994). Surprise. In V. S. Ramachandran
(Ed.), Encyclopedia of human behavior (Vol. 4, pp. 353­358). Orlando,
FL: Academic Press.
Meyer, W.-U., Niepel, M., Rudolph, U., & Schu¨tzwohl, A. (1991). An
experimental analysis of surprise. Cognition and Emotion, 5, 295­311.
Meyer, W.-U., Reisenzein, R., & Schu¨tzwohl, A. (1997). Toward a process
analysis of emotions: The case of surprise. Motivation and Emotion, 21,
251­274.
Most, S. B., & Simons, D. J. (2001). Attention capture, orienting, and
awareness. In C. Folk & B. Gibson (Eds.), Attraction, distraction, and
1059CAPTURE AND EXPECTATION DISCREPANCY
action: Multiple perspectives on attentional capture (pp. 151­173).
Amsterdam: Elsevier Science.
Most, S. B., Simons, D. J., Scholl, B. J., Jimenez, R., Clifford, E., &
Chabris, C. F. (2001). How not to be seen: The contribution of similarity
and selective ignoring to sustained IB. Psychological Science, 12, 9­17.
Neisser, U. (1976). Cognition and reality: Principles and implications of
cognitive psychology. San Francisco: Freeman.
Neisser, U. (1979). The control of information pickup in selective looking.
In A. D. Pick (Ed.), Perception and its development: A tribute to Eleanor
J. Gibson (pp. 201­219). Hillsdale, NJ: Erlbaum.
Niepel, M., Rudolph, U., Schu¨tzwohl, A., & Meyer, W.-U. (1994). Tem-
poral characteristics of the surprise reaction induced by schema-
discrepant visual and auditory events. Cognition and Emotion, 8, 433­
452.
Nothdurft, H.-C. (2000). Salience from feature contrast: Variations with
texture density. Vision Research, 40, 3181­3200.
Posner, M. I., & Snyder, C. R. R. (1975). Attention and cognitive control.
In R. L. Solso (Ed.), Information processing and cognition: The Loyola
Symposium (pp. 55­85). Hillsdale, NJ: Erlbaum.
Prinz, W. (1983). Wahrnehmung und Ta¨tigkeitssteuerung. Berlin:
Springer.
Prinz, W. (1990). Unwillku¨rliche Aufmerksamkeit. In C. Meinecke & L.
Kehrer (Eds.), Bielefelder Beitra¨ge zur Kognitionspsychologie (pp. 49­
57). Go¨ttingen: Hogrefe.
Rumelhart, D. E. (1984). Schemata and the cognitive system. In R. S. Wyer
& T. K. Srull (Eds.), Handbook of social cognition (Vol. 1, pp. 161­
188). Hillsdale, NJ: Erlbaum.
Rumelhart, D. E., & Ortony, A. (1977). The representation of knowledge
in memory. In R. C. Anderson, R. J. Spiro, & W. E. Montague (Eds.),
Schooling and the acquisition of knowledge (pp. 99­135). Hillsdale, NJ:
Erlbaum.
Rumelhart, D. E., Smolensky, P., McClelland, J. L., & Hinton, G. E.
(1986). Schemata and sequential thought processes in PDP models. In
D. E. Rumelhart, J. L. McClelland, & PDP Research Group (Eds.),
Parallel distributed processing: Explorations in the microstructure of
cognition (Vol. 2, pp. 7­57). Cambridge, MA: MIT Press.
Schu¨tzwohl, A. (1993). Schemata und U¨ berraschung: Untersuchungen zum
Zusammenwirken von Kognition und Emotion. Unvero¨ffentlichte Dis-
sertation, Universita¨t Bielefeld [Schema and surprise: Investigations
concerning the interaction of cognition and emotion. Unpublished doc-
toral dissertation, Bielefeld, University]. Universita¨t Bielefeld.
Schu¨tzwohl, A. (1998). Surprise and schema strength. Journal of Experi-
mental Psychology: Learning, Memory, and Cognition, 24, 1182­1199.
Selz, O. (1922). Zur Psychologie des produktiven Denkens und des Irr-
tums: eine experimentelle Untersuchung. [The psychology of productive
thinking and of error: An experimental study.] Bonn, Germany: Cohen.
Simons, D. J., & Chabris, C. F. (1999). Gorillas in our midst: Sustained
inattentional blindness for dynamic events. Perception, 28, 1059­1074.
Sternberg, S. (1969). The discovery of processing stages: Extensions of
Donders' method. Acta Psychologica, 30, 276­315.
Theeuwes, J. (1992). Perceptual selectivity for color and form. Perception
& Psychophysics, 51, 599­606.
Theeuwes, J. (1994). Stimulus-driven capture and attentional set: Selective
search for color and visual abrupt onsets. Journal of Experimental
Psychology: Human Perception and Performance, 20, 799­806.
Theeuwes, J., & Burger, R. (1998). Attentional control during visual
search: The effect of irrelevant singletons. Journal of Experimental
Psychology: Human Perception and Performance, 24, 1342­1353.
Theeuwes, J., & Godijn, R. (2001). Attentional and oculomotor capture. In
C. Folk & B. Gibson (Eds.), Attraction, distraction, and action: Multiple
perspectives on attentional capture (pp. 121­149). Amsterdam: Elsevier
Science.
Todd, S., & Kramer, A. F. (1994). Attentional misguidance in visual
search. Perception & Psychophysics, 56, 198­210.
Treisman, A. M., & Gelade, G. (1980). A feature-integration theory of
attention. Cognitive Psychology, 12, 97­136.
Wilcocks, R.-W. (1928). The effect of an unexpected heterogeneity on
attention. Journal of General Psychology, 1, 286­319.
Wolfe, J. M. (1994). Guided Search 2.0: A revised model of visual search.
Psychonomic Bulletin & Review, 1, 202­238.
Wolfe, J. M., O'Neill, P., & Bennett, S. C. (1998). Why are there eccen-
tricity effects in visual search? Visual and attentional hypotheses. Per-
ception & Psychophysics, 60, 140­156.
Yantis, S. (1993). Stimulus-driven attentional capture and attentional con-
trol settings. Journal of Experimental Psychology: Human Perception
and Performance, 19, 676­681.
Yantis, S., & Egeth, H. E. (1999). On the distinction between visual
salience and stimulus-driven attentional capture. Journal of Experimen-
tal Psychology: Human Perception and Performance, 25, 661­676.
Received December 12, 2003
Revision received October 29, 2004
Accepted March 29, 2005
1060 HORSTMANN