Scand. J. of Economics 110(2), 339–366, 2008
DOI: 10.1111/j.1467-9442.2008.00542.x
Stress that Doesn’t Pay:
The Commuting Paradox∗
Alois Stutzer
University of Basel, CH-4003 Basel, Switzerland
alois.stutzer@unibas.ch
Bruno S. Frey
University of Zurich, CH-8006 Zurich, Switzerland
bsfrey@iew.uzh.ch
Abstract
People spend a lot of time commuting and often find it a burden. According to standard
economics, the burden of commuting is chosen when compensated either on the labor or on
the housing market so that individuals’ utility is equalized. However, in a direct test of this
strong notion of equilibrium with panel data, we find that people with longer commuting
time report systematically lower subjective well-being. This result is robust with regard to a
number of alternative explanations. We mention several possibilities of an extended model
of human behavior able to explain this “commuting paradox”.
Keywords: Location theory; commuting; compensating variation; subjective well-being
JEL classification: D12; D61; R41
I. Introduction
Commuting is an important aspect of our lives that demands a lot of our
valuable time. There are conflicting ideas on the subject. For most people,
commuting is a mental and physical burden, giving cause for various complaints.
From an economic perspective, commuting is just one of numerous
decisions rational individuals make. If commuting has extra psychological
costs, then traveling longer distances to and from work is only chosen if
it is either compensated by an intrinsically or financially rewarding job
or by additional welfare gained from a pleasant living environment. Accordingly,
commuting is determined by an equilibrium state of the housing
and labor market, in which individuals’ well-being or utility is equalized
∗
We are grateful to Matthias Benz, Piet Bovy, Reiner Eichenberger, Reto Jegen, Gebhard
Kirchg¨assner, Gerrit Koester, Alan Krueger, Rafael Lalive, Stephan Meier, Uri Simonsohn,
J. D. Trout, Jos van Ommeren, two anonymous referees and various participants of the Assistants’
Conference in Berlin and the Labor Seminar at the Tinbergen Institute in Amsterdam
for helpful comments. Data for the German Socio-economic Panel were kindly provided by
the German Institute for Economic Research (DIW) in Berlin.
C The editors of the Scandinavian Journal of Economics 2008. Published by Blackwell Publishing, 9600 Garsington Road,
Oxford, OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA.
340 A. Stutzer and B. S. Frey
over all actual combinations of alternatives in these two markets. Thus,
any disagreement between the two perspectives is due to the strong belief
in economics that market forces lead to an equilibrium in which rents are
prevented.
The strong notion of equilibrium in urban and regional economic theory,
as well as in public economic theory, has only been partially tested so far.
Studies have not been carried out as to whether there are systematic rents:
rather, derived hypotheses within the equilibrium framework have been
analyzed. There is considerable evidence for capitalization of transportation
infrastructure in the price of land and for compensating wage differentials
due to commuting distance.1
However, these findings do not require an
equilibrium situation, but can also be explained by the law of marginal
substitution.
In order to assess the power of the equilibrium framework, a direct test is
necessary. Here we analyze data on subjective well-being as proxy measures
for people’s experienced utility in order to directly test the strong notion of
equilibrium in location theory. High quality data are available for Germany,
collected by the German Socio-economic Panel. In a data set spanning
19 years, we study whether commuters are indeed compensated for the
stress incurred, as suggested in economic models. If this is the case, we
should not find any systematic correlation between people’s commuting
time and their reported satisfaction with life.
Our main result indicates, however, that people with long journeys to and
from work are systematically worse off and report significantly lower subjective
well-being. For economists, this result on commuting is paradoxical.
The empirical finding is further analyzed in four ways. First, we study the
robustness of the empirical finding to different econometric specifications.
In particular, a large number of background variables and time-invariant
personality traits are taken into account in the estimation approach.
Second, biases in judgment due to the effects of the order in which questions
are asked, or differences in salience, might cover up actual compensation
in reported life satisfaction. Therefore, domain satisfaction is studied
in order to capture possible compensation on the labor and the housing
market at a disaggregate level, rather than in an overall measure.
Third, we discuss and empirically analyze two possible explanations
within the traditional economic framework that would account for the commuting
paradox: (i) While commuting might be a burden for those involved,
those people’s partners might benefit, so that, overall, the households’ wellbeing
is equalized. (ii) Transaction costs prevent people from adjusting to
economic shocks and the observed correlation might simply reflect equilibrium
in a “real” world with frictions. In fact, the general finding might
1
See the research cited in Section II below.
C The editors of the Scandinavian Journal of Economics 2008.
Stress that doesn’t pay: the commuting paradox 341
exemplify the importance of moving costs. People are trapped in their commuting
situation and experience lower subjective well-being when they had
bad luck and ended up with a long commute or did not foresee the costs
of commuting. Here, we study people who change either their job or their
place of residence, and thus have the possibility of re-optimizing their lives.
The question is asked whether they also suffer lower subjective well-being
with higher commuting time and we find that they do.
Finally, we suggest several possibilities of an extended model of human
behavior that may help us to better understand the “commuting paradox”.
The paper proceeds as follows. The costs and benefits of commuting
as discussed in economics and psychology are summarized in Section II.
The data set is described and the empirical analyses are conducted in Section
III. Several explanations of the commuting phenomenon are empirically
tested in Section IV. Section V briefly addresses the results in the light of
behavioral economics. Concluding remarks are offered in Section VI.
II. The Costs and Benefits of Commuting
The Physical and Mental Burden of Commuting
Commuting involves much more than just covering the distance between
home and work. Commuting not only takes time, but also generates outof-pocket
costs, causes stress and intervenes in the relationship between
work and family. In fact, it seems that commuting is the daily activity that
generates the lowest level of positive affect, as well as a relatively high
level of negative affect; see Kahneman, Krueger, Schkade, Schwarz and
Stone (2004). Moreover, commuting is salient in the everyday routines of
many people’s lives. Figure 1 gives a brief overview about commuting in
European countries and the United States. It clearly shows that commuting
is a widespread phenomenon. Workers in these countries commute between
29.2 minutes in Portugal and 51.2 minutes a day in Hungary. The average
daily commuting time in the former EU15 is 37.5 minutes. In the United
States, traveling to work takes, on average, 48.8 minutes.
Engineers and social scientists have studied a wide range of the private
and social costs of commuting; for a review, see Koslowsky, Kluger and
Reich (1995). For example, it has been calculated for the United States
that a “typical household spends nearly 20 percent of its income on driving
costs—more than it spends on food”; see EPA (2001). Besides these private
costs for transportation (including commuting), there are the social costs
of commuting, due to congestion and pollution of the environment. The
calculation of the costs of congestion focuses on the value of time when
delays occur whilst traveling. In an extensive survey, Small (1992, p. 44)
C The editors of the Scandinavian Journal of Economics 2008.
342 A. Stutzer and B. S. Frey
Fig. 1. Average daily commuting time in Europe and the US
Data sources: Data for European countries are from the European Survey on Working Conditions,
conducted by the European Foundation for the Improvement of Living and Working
Conditions in 2000 for member countries and in 2001/02 for acceding and candidate countries.
Data for the US are from US Census Bureau, 2002 American Community Survey.
concludes that “a reasonable average value of time for the journey to work
is 50 percent of the gross wage rate, while recognizing that it varies among
different industrialized cities from perhaps 20 to 100 percent of the gross
wage rate, and among population subgroups by even more”.
Psychologists have focused on the non-pecuniary costs of commuting and
emphasize that it is an unpleasant experience that often has delayed effects
on health and family life; for surveys, see e.g. Novaco, Stokols and Milanesi
(1990) and Koslowsky et al. (1995). Commuting is associated with many
environmental stressors like noise, crowds, pollution and thermal conditions
C The editors of the Scandinavian Journal of Economics 2008.
Stress that doesn’t pay: the commuting paradox 343
that cause negative emotional and physical reactions. Reactions depend, of
course, not only on the time and distance involved in commuting, but also
on other factors that interact with the stressors mentioned above. Commuting
is more stressful when people are not in control of certain factors that
can crop up during the drive to work, e.g. due to traffic congestion or
when they are under considerable time pressure. The strain of commuting
is associated with raised blood pressure, musculoskeletal disorders, lowered
frustration tolerance and increased anxiety and hostility, being in a
bad mood when arriving at work in the morning and coming home in the
evening, increased lateness, absenteeism and turnover at work, as well as
adverse effects on cognitive performance; see Koslowsky et al. (1995).
The Benefits Associated with Commuting
People benefit from commuting when it allows them to get to an office or
a factory in order to supply their work, or when they can find either superior
or cheaper housing, albeit at a greater distance from work. Individuals
take these benefits, as well as the pecuniary and non-pecuniary commuting
costs mentioned above, into consideration when they make decisions on
where to live, where to work and how to commute. Accordingly, houses that
are further away from the location of work opportunities are less attractive
to people, and thus have a lower market value, ceteris paribus. Jobs that
involve a longer commute have to pay employees more in order to attract
them and keep them. If all the participants in a perfect housing and labor
market optimize, all the commuters are fully compensated for their traveling
costs from home to work, either by higher salaries or by lower rents. Individuals’
utility is then equalized over all possible locations within space.2
These insights have been established in classical urban location theory, as in
e.g. Alonso (1964), Muth (1969) and Huriot and Thisse (2000), and public
economics theory based on Tiebout’s (1956) model of fiscal competition
between jurisdictions; see e.g. Conley and Konishi (2002).3
They reflect
the strong belief in economics that market forces lead to an equilibrium in
which rents and discrimination are prevented.
2
This prediction is expected to hold in equilibrium. In the short run, people may not have
found their optimal portfolio. There are individuals who gain rents from commuting, while
others suffer from costs related to commuting that are not compensated. On average, however,
it is expected that people be compensated for costs incurred from commuting. It is thus
predicted that there is no systematic relationship between commuting time and people’s utility
level.
3
The efficient allocation of resources has been studied, based on the conviction set forth by
Wildasin (1987, pp. 1136ff.) that “migratory flows will arbitrage away any utility differentials
among jurisdictions. Therefore, it is appropriate to impose equal utilities as a constraint at
the outset, and to ask what allocation of resources will maximize the common level of utility
for all households”.
C The editors of the Scandinavian Journal of Economics 2008.
344 A. Stutzer and B. S. Frey
The strong notion of equilibrium in location theory has only been partially
tested so far. It has not been studied whether there are systematic
rents: rather, derived hypotheses within the equilibrium framework have
been analyzed. There is considerable evidence for capitalization of transportation
infrastructure in the price of land, and distance from job locations
and other amenities in housing prices, as in e.g. McMillen and Singell
(1992) and So, Orazem and Otto (2001), as well as for compensating wage
differentials due to commuting distance, as in e.g. van Ommeren, van den
Berg and Gorter (2000) and Timothy and Wheaton (2001).
However, these approaches do not allow us to assess whether the compensation
of commuters is complete and, if it is not, to calculate the amount
that would be needed. The extent of compensation would provide evidence
to judge the relevance of conclusions that are based on equilibrium theories.
In the next section, we propose a new approach of directly measuring
the degree to which commuters are compensated for the burden of com-
muting.
III. Empirical Analysis of the Effects of Commuting on
Subjective Well-being
Data and Descriptive Statistics
Individuals’ compensation for commuting has so far been studied in terms
of higher earnings and lower rents for housing. Here we apply a novel
approach and directly analyze commuters’ level of experienced utility.
Thereby, reported subjective well-being is used as a proxy measure for
utility.4
Although this is not (yet) standard practice in economics, indicators
of happiness or subjective well-being have increasingly been studied
and successfully applied; for surveys see e.g. Frey and Stutzer (2002a,b),
Layard (2005) and Di Tella and MacCulloch (2006).
Measures of reported subjective well-being passed a series of validation
tests, revealing that people who report high subjective well-being smile
more often during social interactions and are less likely to commit suicide.
Changes in brain activity and heart rate account for substantial variance in
reported negative affects. Reliability studies found that reported subjective
well-being is fairly stable and sensitive to changing life circumstances; see
Frey and Stutzer (2002b) for references. However, in order to conduct welfare
comparisons on the basis of reported subjective well-being, a further
4
Subjective well-being is the scientific term in psychology for an individual’s evaluation of
his or her experienced positive and negative affect, happiness or satisfaction with life. With
the help of a single question or several questions on global self-reports, it is possible to get
indications of individuals’ evaluation of their life satisfaction or happiness; see Diener, Suh,
Lucas and Smith (1999).
C The editors of the Scandinavian Journal of Economics 2008.
Stress that doesn’t pay: the commuting paradox 345
condition has to be met. Well-being must be interpersonally comparable.
Economists are likely to be skeptical about this claim. However, evidence
has been gathered that it may be less of a problem on a practical level
than on a theoretical level. Happy people, for example, are rated as happy
by friends and family members, as reported by e.g. Lepper (1998), as well
as by spouses. Furthermore, ordinal and cardinal treatments of satisfaction
scores generate quantitatively very similar results in microeconometric happiness
functions; see e.g. Frey and Stutzer (2000) and Ferrer-i-Carbonell and
Frijters (2004). Therefore, throughout the paper, results from least-squares
estimations are reported. The existing state of research suggests that, for
many purposes, happiness or reported subjective well-being is a satisfactory
empirical approximation to experienced utility.
The current study is based on data on subjective well-being from the
German Socio-economic Panel Study (GSOEP). The GSOEP is one of the
most valuable data sets for studying individual well-being over time. It was
started in 1984 as a longitudinal survey of private households and persons
in the Federal Republic of Germany, and was extended to include residents
in the former German Democratic Republic in 1990. From this survey, we
primarily used the eight waves between 1985 and 2003 that contain information
about individual commuting time. Additional waves were taken into
account when studying commuting distance and when imputing information
on commuting time. All of our estimations are based on unbalanced
panels. People in the survey were asked a wide range of questions with
regard to their socio-economic status and their demographic characteristics.
Moreover, they reported their actual commuting time and their subjective
well-being. Commuting time is captured by the question, “How long does
it normally take you to go all the way from your home to your place of
work using the most direct route (one way only)?” Reported subjective
well-being is based on the question, “How satisfied are you with your life,
all things considered?” Responses range on a scale from 0 “completely
dissatisfied” to 10 “completely satisfied”. In order to study the effect of
commuting on individual well-being, we restricted the sample to those who
either commute on a regular basis to the same place or work at home and
who report being either employed or self-employed. Descriptive statistics
for the dependent variable life satisfaction, as well as all the covariates used
in the empirical analysis, are provided in Table A1 in the Appendix.
Figure 2 presents the distribution of reported commuting time in
Germany between 1985 and 2003. On average, people in the sample
commute 22 minutes one way (a total of 44 minutes a day) with a standard
deviation of 18 minutes. Median commuting time is 15 minutes. Commuters,
who report traveling to work taking an hour or more, comprise
6.8 percent of the sample.
C The editors of the Scandinavian Journal of Economics 2008.
346 A. Stutzer and B. S. Frey
Fig. 2. Distribution of average daily commuting time (one way)
Data source: GSOEP.
Commuting and Reported Satisfaction with Life
Testing Strategy. The concept of equilibrium in economics predicts that pecuniary,
as well as mental, costs of commuting are compensated for on
the labor and housing market. Thus, individuals’ utility level is equalized
over all actual combinations of alternatives in these two markets. This, of
course, only holds for homogeneous people. We start with this assumption
to introduce our empirical testing strategy. However, we also extend our
argument to include people with heterogeneous preferences. Empirical estimations
refer only to the latter case. In the underlying model, commuters’
utility is increasing in consumption c of goods, services and housing, and
decreasing in the disamenity D for commuting time, U = u(c, D).
Utility U is equal to ¯U for realized combinations of income yi, time
spent commuting Di and rent ri across individuals indexed by i:
Ui = u(yi , Di ,ri ) = ¯U for all i. (1)
Totally differentiating this equilibrium condition, we get
dU =
∂u
∂y
dy +
∂u
∂D
dD +
∂u
∂r
dr = 0. (2)
C The editors of the Scandinavian Journal of Economics 2008.
Stress that doesn’t pay: the commuting paradox 347
For variation in commuting time D, this implies that
dU
dD
=
∂u
∂y
dy
dD
+
∂u
∂D
+
∂u
∂r
dr
dD
= 0. (3)
The LHS of equation (3) states that the overall change in utility due
to a change in the disamenity commuting time is zero. A decomposition
of the total change is provided on the RHS of equation (3). There are
three effects of an increase in commuting time. There is a marginal gain
in utility due to a higher level of consumption that is reached because
jobs that require longer commutes offer a higher income. Moreover, longer
commuting time reduces rents for housing and thus leaves additional money
for consumption. Besides these two positive effects, there is a marginal
decrease in utility due to the burden of spending more time commuting.
Given that incomes and rents for housing exclusively reflect compensation
for commuting conditions, the three effects add up to zero.
The prediction in equation (3) can be tested directly. We take commuters’
reported satisfaction with life as a proxy measure for individual utility. The
idea for the empirical test is captured in the following regression equation:
ui = α+ βDi + εi . (4)
The coefficient β measures the total change in utility due to a change
in commuting time. Under the null hypothesis β = 0, commuting time is
entirely compensated by either higher salaries or lower rents for housing.
The alternative hypothesis β < 0 states that commuting time is not fully
compensated on the labor and housing market.
Cross-section Evidence. Figure 3 provides a first visual test to see whether
there are indications of any kind of a correlation between commuting time
and people’s life satisfaction. Average life satisfaction is reported for the
four quartiles of commuting time. Contrary to the prediction of β = 0 in
equilibrium, results indicate that there is a sizable negative correlation between
commuting time and individuals’ well-being. For each subsequent
quartile of longer commuting time, we find, on average, a lower reported
satisfaction with life. While life satisfaction is 7.23 points, on average, for
people who commute 10 minutes or less (first quartile), average satisfaction
scores for the top fourth quartile (commuting time more than 30 minutes)
is 6.99 points, i.e., 0.24 points lower.
The raw correlation between commuting time and life satisfaction does
not take into consideration that we compare people with heterogeneous preferences
facing different restrictions. In other words, the optimal commuting
time is probably systematically different for different groups of people. Thus
the observed lower subjective well-being of people who spend more time
traveling from home to work might just reflect that these are people with
C The editors of the Scandinavian Journal of Economics 2008.
348 A. Stutzer and B. S. Frey
6.9
7.0
7.1
7.2
7.3
0 10 20 30 40 50 60
1st quartile
2nd quartile
3rd quartile
4th quartile
Fig. 3. Commuting time and average reported satisfaction with life, Germany, 1985–2003
Data source: GSOEP.
different socio-demographic and socio-economic characteristics. In order to
apply the test for compensation, groups of people who are very similar have
to be empirically constructed. Technically, a multiple regression approach
is applied to control for individual characteristics.
Equation (4) is extended in order to include a set of individual covariates
Xi:
ui = α+ βDi + γXi + εi . (5)
It is important to note that Xi does not include respondents’ labor income,
their household income or working hours. This is crucial, because income
(and to some extent also working hours) is one of the variables through
which people are compensated for the distance they cover to and from
work. Equation (5) only makes a sharp prediction of β = 0 if all channels
for compensation remain uncontrolled. If income is controlled, people who
spend more time commuting are, of course, worse off, ceteris paribus.
Heterogeneous preferences for commuting also imply sorting. It is the
quintessence of spatial economics that people reside where their preferences
are best met. It is this process of sorting and arbitrage that leads to the
prediction on compensation. How do heterogeneous tastes for commuting
and sorting affect any observed partial correlation between commuting time
and life satisfaction in a cross-section estimation?
C The editors of the Scandinavian Journal of Economics 2008.
Stress that doesn’t pay: the commuting paradox 349
Imagine that people have homogeneous tastes in all respects but commuting.
There are some people who strongly dislike commuting. Given
their possibilities on the labor market, they are worse off than people who
do not mind commuting. What commuting time do these people optimally
choose? They have a high willingness to pay for a short commute. Other
things equal, they thus live closer to where they work and are willing to pay
more for housing. From the two arguments, the following picture emerges:
people who dislike commuting have a disadvantage in our spatial economy.
While they choose a combination of job and housing that involves relatively
short commuting, they experience lower utility than people whose disutility
from commuting is small. Accordingly, all else equal, a positive correlation
between commuting time and a proxy measure for utility is expected. This
prediction runs counter to the correlation observed in our sample. With
regard to the specific sorting argument, we estimate a lower bound in the
following cross-section equation.5
In Table 1, equation (5) for the effect of commuting time on life satisfaction
was first estimated in a pooled least-squares regression, taking
a large number of individual characteristics into account, as well as year
dummies.6
The results in Table 1 show that people who spend more time
commuting report lower satisfaction with life, ceteris paribus. Based on
an F-test, the proposition that the two commuting variables together are
not correlated with reported life satisfaction is rejected on the 99 percent
level. An increase of an individual’s commuting time by one hour and an
initial commuting time of 0 refers, on average, to a 0.28 point (t = −9.20)
lower subjective well-being. For one standard deviation (i.e., 18 minutes)
the effect amounts to −0.09 (t = −5.92).
Evidence from Estimations with Individual-specific Fixed Effects. The
pooled estimation in Table 1 identifies the effect of commuting on reported
well-being, based on the variation in these two variables between people
and for each individual over time. It is assumed that any measurement
errors, as well as unobserved characteristics, are captured in the error term
of the estimation. Indeed, many mistakes in people’s answers are random
5
A similar argument holds for people with preferences for environmental qualities that are
positively correlated with commuting like gardens in residential areas. These people require
less compensation and are relatively better off. However, the observed correlations in a pooled
cross-section estimation overestimate the losses incurred (or they are in fact spurious) if people
have preferences for spatial characteristics that are positively correlated with commuting
but negatively with person-specific reporting behavior.
6
A discussion of the results for the socio-demographic and socio-economic factors in
Germany can be found in Stutzer and Frey (2004). Note that self-employment is taken
as a control variable even though some people may choose to be self-employed in order to
avoid the daily stress of commuting.
C The editors of the Scandinavian Journal of Economics 2008.
350 A. Stutzer and B. S. Frey
Table 1. Commuting and satisfaction with life, Germany 1985–2003 (dependent
variable: satisfaction with life)
(1) (2)
Coefficient t-Value Coefficient t-Value
Commuting time (in minutes) −0.0054 −5.04 −0.0054 −3.30
Commuting time squared 0.012e − 3 0.96 0.035e − 3 1.97
Individual characteristicsa Yes Yes
Individual fixed effects No Yes
Year fixed effects Yes Yes
F-test (Prob. > F) 0.000 0.000
Commuting time = 0 and
commuting time squared = 0
Effect of one hour of commuting −0.284 −9.20 −0.200 −3.99
No. of observations 39,141 39,141
No. of individuals 19,088 19,088
(3) (4)
Commuting time −0.0045 −9.88 −0.0025 −3.47
Individual characteristicsa Yes Yes
Individual fixed effects No Yes
Year fixed effects Yes Yes
Effect of one hour of commuting −0.270 −9.88 −0.151 −3.47
No. of observations 39,141 39,141
No. of individuals 19,088 19,088
Data source: GSOEP.
Notes: Partial correlations are from least-squares estimations.
a
Individual control variables in specification (1) include age, age squared, sex, six categories for years of
education, two variables for the relationship to the head of household, nine variables for marital status, three
variables for the number of children in the household, the square root of the number of household members and
indicators for self-employment, residence in the New German Laender, foreigners with EU nationality, other
foreigners and first interview.
and thus do not bias the estimation results. This holds true, for example, for
the order of questions, the wording of questions, actual mood, etc. However,
non-sampling errors are not always uncorrelated with the variables
of interest. A measurement error perspective suggests that the inferences
can be clouded by unobserved personality traits that, in our case, influence
individuals’ commuting behavior, as well as how they respond to subjective
well-being questions. For instance, restless people who have difficulty settling
down may, on average, choose longer commutes and may also report
lower satisfaction with life. As a result, the observed correlation is biased.
A related concern involves heterogeneity in people’s income (generating
potential). If housing options close to workplaces are not feasible for some
people due to income constraints, they might be more likely to live in
suburbs and spend more time commuting. Long commuting time might
C The editors of the Scandinavian Journal of Economics 2008.
Stress that doesn’t pay: the commuting paradox 351
thus reflect low household income and the correlation in the cross-section
might be spurious.7
However, idiosyncratic effects that are time invariant
can be controlled for if the same individuals are re-surveyed over time.
This is the case for our longitudinal data set, in which it is possible to
consider a specific baseline well-being for each individual. The statistical
relationship between commuting and reported subjective well-being is then
identified by the variation in commuting time within observations for the
same person. In our sample, the mean standard deviation of individual
commuting experiences is 8.7 minutes.
The second estimation in Table 1 reports the result for an estimation
with individual fixed effects that excludes spurious correlation due to timeinvariant
unobserved characteristics of people. Partial correlations again
show a negative effect of commuting time on life satisfaction. The two
variables for commuting time are jointly statistically significantly different
from zero.8
People who spend one hour rather than 0 minutes commuting
(one way) report, on average, a −0.20 points (t = −3.99) lower level of subjective
well-being. For one standard deviation (i.e., 18 minutes), the effect
is −0.086 (t = −3.52). The size of the commuting effect for one standard
deviation is half the effect of finding or losing a partner for those who are
single (fixed-effects estimation). Compared to the effect of becoming unemployed
(= −0.671), as reported in Stutzer and Frey (2004, Table 4), an
increase in commuting time by one standard deviation (one hour) is about
one-eighth (one-fourth) as bad for life satisfaction.
Thus, the results of the raw correlation and the pooled estimation are
confirmed. All of them are at odds with the prediction of standard location
theory and the implicit assumption in many economics models that, on
average, people are compensated for commuting.
The two estimation approaches in Table 1 lead to somewhat different
results for the effect of commuting on subjective well-being. The partial
correlation is larger in the pooled regression, which also includes information
on variation between people. Potentially, this allows us to estimate
the correlation between commuting time and subjective well-being more
efficiently. In order to test whether the individual fixed effects are correlated
with the explanatory variables, a Hausman test was performed.
7
Contrary to the mentioned presumption, in an estimation of the covariates of commuting
time in Germany (not shown), household income is statistically significantly positively correlated
with commuting time. A doubling of household income is related to a slightly higher
commuting time of 0.63 minutes.
8
A quadratic specification of the effect of commuting time on life satisfaction is chosen
because we hypothesize that the marginal burden of commuting is falling. This is based on
the idea that monetary commuting costs increase in a less than proportional way to increases
in commuting time. In the fixed-effects estimation, this hypothesis is not rejected. However,
we also report results for linear specifications in the bottom half of Table 1.
C The editors of the Scandinavian Journal of Economics 2008.
352 A. Stutzer and B. S. Frey
The hypothesis that there are no systematic differences in the coefficients
between the fully efficient model in the first two columns and the less
efficient fixed-effects estimate in Table 1, however, is clearly rejected. The
negative effect of commuting in the fixed-effects model thus more accurately
measures the incompleteness in compensation.
The Role of Commuting Distance and the Mode of Transportation. In order
to broaden the view on the phenomenon, Table 2 takes commuting distance
and the mode of transportation into account.9
Commuting distance is an
alternative proxy for the burden of commuting. However, we judge it as
less accurate because distance as such is less closely related to the opportunity
cost of commuting than commuting time. We still find a statistically
significant small negative effect of commuting distance on reported life satisfaction.
The effect of a change in commuting time (e.g. an increase due
to worse congestion or a decrease due to a new road), when commuting
distance is kept constant, is estimated in specification (2). Not surprisingly,
a larger negative effect is estimated than in Table 1 as the variation in
unexpected changes in commuting time becomes more important in the estimation.
In contrast, Section IV below reports estimations for people who
either change their job and/or their residence. These estimations exploit
variation in commuting time which people are supposed to have known
about when they changed their job and/or residence.
The pleasures and pains of commuting depend on the mode of transportation.
We test whether there are also systematic differences in the negative
partial correlation between commuting time and subjective well-being
for people who commute either by car, by public transport or by some
other means. According to our interpretation, the question is whether there
are differences in the degree of incomplete compensation between users
of private and public transportation. In our sample, 63 percent of respondents
mainly commute by car, 14 percent mainly use public transport, and
24 percent use either other transportation modes (motorcycle, bike, on foot)
or a combination of different modes.
In order to test for systematic differences in the partial correlations,
interaction terms between commuting time and the mode of transportation
are included. Specification (3) restricts differences to slopes (i.e., there
are no transportation mode-specific intercepts). Specification (4) allows
9
Information about commuting distance is available for the following years: 1985, 1990 West
Germany, 1991 East Germany, 1992, 1993, 1995 and 1997–2005. As commuting distance
was included in the survey more frequently than commuting time, estimation (1) in Table 2
is based on 103,270 observations from 25,171 individuals.
C The editors of the Scandinavian Journal of Economics 2008.
Stress that doesn’t pay: the commuting paradox 353
Table 2. Commuting distance, transportation mode and satisfaction with life
(dependent variable: satisfaction with life)
(1) (2)
Coefficient t-Value Coefficient t-Value
Commuting time −0.0114 −7.70
Commuting time squared 0.048e − 3 3.13
Commuting distance (in km) −2.013e − 3 −2.25 9.308e − 3 5.66
Commuting distance squared 0.012e − 3 1.18 −0.065e − 3 −3.72
Individual characteristicsa Yes Yes
Individual fixed effects Yes Yes
Year fixed effects Yes Yes
No. of observations 103,270 38,818
No. of individuals 25,171 18,966
(3) (4)
Commuting time (CT) (car) −0.0111 −2.91 −0.0127 −2.88
Commuting time squared (car) 0.134e − 3 2.66 0.149e − 3 2.72
CT × public transport 5.948e − 3 1.28 2.368e − 3 0.23
CT2 × public transport −0.129e − 3 −1.82 −0.102e − 3 −0.96
CT × other transportation mode 5.311e − 3 1.11 0.0107 1.42
CT2 × other transportation mode −0.138e − 3 −1.85 −0.190e − 3 −2.04
Public transport 0.1014 0.44
Other transportation mode −0.0894 −0.91
Individual characteristicsa Yes Yes
Individual fixed effects Yes Yes
Year fixed effects Yes Yes
F-test (Prob. > F) 0.014 0.015
Commuting time = 0 and
commuting time squared = 0
F-test (Prob. > F) 0.140 0.215
CT × public transport = 0 and
CT2 × public transport = 0 (and
public transport = 0)
Effect of one hour of commuting −0.185 −1.71 −0.222 −1.84
by car
Effect of one hour of commuting −0.291 −2.31 −0.344 −2.56
by public transport
No. of observations 21,353 21,353
No. of individuals 16,288 16,288
Data source: GSOEP.
Notes: Partial correlations are from least-squares estimations.
a
The same control variables for individual characteristics as in Table 1 are included.
for transportation mode-specific intercepts. Both estimations offer similar
results. While the estimated negative effect of one hour of commuting
is larger for users of public transport than users of cars, the difference
is imprecisely measured and not statistically significantly different from
C The editors of the Scandinavian Journal of Economics 2008.
354 A. Stutzer and B. S. Frey
zero. For both equations, it is not rejected that the interaction terms for
commuting time and public transport (and a specific intercept for public
transport in equation (4)) are jointly equal to zero.
To our knowledge, the empirical analyses in Table 1 and specifications
(3) and (4) in Table 2 directly test, for the first time, the strong
notion of equilibrium in location theory. This is made possible by applying
individual reported subjective well-being as a proxy measure for utility.
Contrary to the common understanding in economics, there seems to be
a systematically incomplete compensation of people who spend more time
commuting between home and work.
Calculation of the Missing Compensation in Monetary Terms. How much
additional income would a commuter have to earn in order to be as well-off
as someone who does not commute? This calculation has to be taken with
a grain of salt as there are many unresolved issues in the assessment of
the marginal utility of income from data on subjective well-being; see the
discussion in Clark, Frijters and Shields (2008). We calculated the compensation
in three steps (exemplary for the mean commuting time of 22
minutes).
First, the life satisfaction differential was calculated that we attribute to
incomplete compensation. This calculation is based on the specification
and estimated coefficients in Table 1 (second estimation including fixed
effects):
U = u(D = 22) − u(D = 0)
= −5.425e − 3∗22 + 0.035e − 3∗222
− 0
= −0.1025. (6)
Second, the marginal utility of additional income was estimated based on
an extended microeconometric happiness function. In order to estimate a
coefficient for the gross marginal effect of additional income, a full specification
is necessary that keeps important determinants of income constant.
Here, commuting time and working hours are controlled for, in addition
to the covariates mentioned in Table 1. Income is measured in terms of
the real monthly net labor income (w) and the real monthly household income
(y) (consisting of the respondent’s labor income w as well as other
household members’ income v).10
U = α+ β1D + β2D2
+ γX + δ1w + δ2w2
+ δ3 y + δ4 y2
and y = w + v.
(7)
10
The results for this estimation can be obtained from the authors on request.
C The editors of the Scandinavian Journal of Economics 2008.
Stress that doesn’t pay: the commuting paradox 355
The marginal utility of additional labor income at the sample mean
( ¯w = 1,326 euros, ¯y = 2,800 euros) is
∂u
∂w
= δ1 + 2∗δ2∗w + δ3 + 2∗δ4∗y
= 0.157e − 3 + 2∗ − 7.80e − 09∗1,326 + 0.100e − 3
+ 2∗ − 3.24e − 09∗2,800
= 0.218e − 3. (8)
Third, the ratio between the loss in utility due to commuting and the
marginal utility of income was built to calculate the missing compensation
in monetary terms:
U
∂u/∂w
=
−0.1025
0.218e − 3
= −469.19. (9)
Full compensation for commuting 22 minutes (one way) compared with
no commuting at all, is estimated to require an additional monthly income
of approximately 470 euros or 35.4 percent of the average monthly labor
income. We do not want to insist on the specific number, but would like to
emphasize that the loss in well-being due to a suboptimal commuting situation
seems sizable whether put in perspective relative to other determinants
of subjective well-being or translated into monetary terms.
IV. Is There a Simple Explanation for the Commuting
Phenomenon?
The finding that people who spend more time commuting are systematically
worse off stands in sharp contrast to the equilibrium view in economics.
There are two completely different ways of reacting to this challenge: First,
the empirical finding may be misleading. In fact, equilibrium is maintained
when households are considered as units that are compensated, or when
utility from jobs and housing is studied directly. Second, equilibrium may
not be attained because of frictions. Transaction costs restrict residential
and job mobility and prevent commuters from being fully compensated.
Is Full Compensation Attained at the Household Level?
While commuting might be a burden for those involved, the members of
their family might benefit so that, overall, the households’ well-being is
equalized. The empirical finding can thus be explained by a too limited
selection of the decision-making unit. At a household level, the equilibrium
may still be attained.
This possible explanation of the commuting paradox is studied empirically
in Table 3. We analyze whether an individual’s subjective well-being
C The editors of the Scandinavian Journal of Economics 2008.
356 A. Stutzer and B. S. Frey
Table 3. Satisfaction with life and partners’ commuting time (dependent variable:
satisfaction with life)
(1) (2)
Coefficient t-Value Coefficient t-Value
Partner’s commuting time (in minutes) −0.0018 −0.71 −0.0012 −0.48
Partner’s commuting time squared 0.021e − 3 0.76 0.018e − 3 0.62
Commuting time −0.0063 −2.51
Commuting time squared 0.055e − 3 2.04
Irregular commuting −0.3573 −2.02
Commuting to different places −0.0237 −0.19
Not commuting −0.1759 −1.05
Individual characteristicsa Yes Yes
Individual fixed effects Yes Yes
Year fixed effects Yes Yes
No. of observations 19,054 19,054
No. of individuals 10,556 10,556
Data source: GSOEP.
Notes: Partial correlations are from least-squares estimations.
a
Individual control variables are the same as in Table 1. In addition, two control variables for work status
(unemployment and no paid work or other status) are included.
is increasing in relation to his or her partner’s commuting time. A positive
partial correlation could balance out the compensation that is missing
for those who actually commute. However, our results do not support this
alternative explanation. In a pooled least-squares estimation (not shown),
we find that the more time respondents’ partners spend commuting, the
less satisfied the respondents are. The negative effect is roughly a third of
the size of the effect that is measured for people’s own commuting (first
estimation in Table 1). This result indicates that commuting might even
result in negative externalities for other family members (consistent with
previous research on commuting and family tensions mentioned in Section
II). However, in the fixed-effects estimations shown in Table 3, the
negative effect of a partner’s commuting time is close to zero. In sum,
there is no evidence that people systematically benefit from the commuting
of other household members.
The issue of intra-household bargaining can be excluded if only singleperson
households are studied. However, the sample is then reduced substantially
to 3,622 observations and individuals are observed, on average,
only 1.5 times. In a fixed-effects estimation, we find a large negative
effect of commuting time on life satisfaction. The partial correlation for
the linear term is −0.0197 (t = −2.62) and the square term is 0.17e − 3
(t = 1.95). This amounts to a negative effect of a one-hour commute of
−0.578 (t = −2.68). However, the standard error of this estimation is
C The editors of the Scandinavian Journal of Economics 2008.
Stress that doesn’t pay: the commuting paradox 357
large and the 95 percent confidence interval includes the negative effect
estimated for the fixed-effects specification shown in Table 1. Still,
the finding strengthens the paradoxical finding from the previous section,
rather than any alternative explanation based on intra-household altruism or
bargaining.
Are There Indications for Compensation in Satisfaction with
Particular Life Domains?
There is a second reason why equilibrium could actually be attained, though
not be reflected accordingly in reported subjective well-being. When people
make a judgment about their well-being, particular life domains and experiences
might be more salient than others; see Schwarz and Strack (1999).
In our case, commuting might be over-represented in people’s evaluation
calculus at the time of the interview.
In order to detect possible compensation on the labor and the housing
market that might not be accurately measured in overall life satisfaction, we
additionally study domain satisfaction. The results are shown in Table 4.
Table 4. Commuting and domain satisfaction
Satisfaction with . . .
Health Job Dwelling Spare time Environment
Mean satisfaction 7.072 7.147 7.426 6.506 6.143
[std. dev.] [2.05] [2.02] [2.15] [2.29] [2.03]
Estimation coefficients
Commuting time −7.00e – 3 −8.69e – 3 −0.57e – 3 −0.014 −1.85e – 3
(−3.55) (−4.03) (−0.25) (−4.45) (−0.76)
Commuting time squared 0.05e – 3 0.07e – 3 0.00e – 3 0.05e – 3 0.00e – 3
(2.58) (3.35) (0.00) (1.42) (0.39)
Individual characteristicsa Yes Yes Yes Yes Yes
Individual fixed effects Yes Yes Yes Yes Yes
Year fixed effects Yes Yes Yes Yes Yes
F-test (Prob. > F ) 0.001 0.000 0.855 0.000 0.598
Commuting time = 0
and commuting time
squared = 0
Effect of one hour −0.223 −0.243 −0.034 −0.658 −0.075
of commuting (−3.70) (−3.67) (−0.49) (−6.98) (−1.00)
No. of observations 39,069 38,356 38,938 28,018 29,430
No. of individuals 19,063 18,756 19,014 17,901 16,068
Source: GSOEP.
Notes: Partial correlation coefficients are from least-squares estimations. t-Values are in parentheses.
a
The same control variables for individual characteristics as in Table 1 are included.
C The editors of the Scandinavian Journal of Economics 2008.
358 A. Stutzer and B. S. Frey
According to the initial notion of equilibrium, it is hypothesized that people
who spend more time commuting are compensated by a more attractive job
or home and, accordingly, report higher satisfaction with these two aspects.
However, results for domain satisfaction contradict these predictions. People
with a lengthy distance to and from work do not report increased satisfaction
with their dwelling and report even lower satisfaction with their job.
Employed and self-employed people who spend an hour commuting (one
way) report, on average, a 0.24 points (t = −3.67) lower satisfaction with
their job. Both findings are inconsistent with the idea of compensation
in location theory and sustain the commuting paradox. Results in Table
4 further indicate that commuting time is significantly negatively correlated
with health satisfaction and it has a large negative effect on people’s
satisfaction with their spare time.11
We thus find a negative partial correlation
for commuting time in one specific domain of satisfaction where we
would expect so (i.e., spare time) and a negative or no correlation for three
domains in which we would expect a positive one (i.e., job, dwelling and
environment).
Is Equilibrium Not Attained as a Result of Frictions?
The reasoning so far might be countered by arguing that there are disequilibrium
models (or search models) in urban and regional economics that
complement Alonso–Muth-type residential location models; for surveys see
e.g. Clark and van Lierop (1986) and Crampton (1999). These models
take transaction costs explicitly into account, as in e.g. Weinberg, Friedman
and Mayo (1981) and van Ommeren, Rietveld and Nijkamp (1997). While
they generate similar predictions for individual behavior on the urban labor
and housing market to the former ones, they predict lower utility for those
in a disadvantaged situation with long commuting times; see e.g. van
Ommeren (2000). Transaction costs prevent people from adjusting
to economic shocks. In particular, transaction costs might hinder people
from experiencing a longer or more disturbing commuting time ex post than
expected ex ante from re-optimizing. Therefore, people might be locked into
a disadvantaged commuting situation. It is very difficult to reject an explanation
based on transaction costs (in particular as transaction costs might
also be systematically involved in behavioral explanations).
A related reasoning links the opportunities for optimization to economic
status. It might be hypothesized that poor people have less chance of
11
The finding that there are significant differences in the negative effect of commuting on
domain satisfaction indicates that the results cannot simply be interpreted as response biases,
whereby less happy people paint an overall gloomier picture in every dimension, i.e., they
overstate commuting time, report lower domain satisfaction and so on.
C The editors of the Scandinavian Journal of Economics 2008.
Stress that doesn’t pay: the commuting paradox 359
optimizing, due to powerful agents on the housing and labor markets,
so that they end up spending more time commuting that is not compensated.
Contrary to this presumption, in our sample, people from low-income
households do not commute more on average (see footnote 7). However,
they seem to experience reduced compensation of the burden of commuting
compared to people from high-income households. In the subsample
of people with a low household income (below median), a fixed-effects
estimation specified as in Table 1, panel (2) shows an effect of one hour
of commuting on life satisfaction of −0.251 (t = −3.27), while the effect
is −0.134 (t = −1.94) for people from high-income households (median or
above). However, it cannot be rejected that the coefficients for commuting
time and commuting time squared are the same in the two subsamples
(Prob. > F = 0.203). Due to the limited longitudinal variation, statistically
significant differences in commuting effects between subsamples are difficult
to establish.
In the remainder of this subsection, we focus on persons who changed
their job or their place of residence between those survey waves for which
we have information on commuting time. These people have the possibility
of re-optimizing. Thus, it is not expected that any changes in commuting
time will be systematically linked to reported life satisfaction. In contrast,
if individuals for some reason accept commuting, despite not being compensated
(the paradoxical case), a negative effect of increased commuting
time on utility would again be observed.
If uncompensated commuting is a reflection of the cost of re-optimizing,
individuals who change either their residence or their job might overcome
the inferior situation and choose optimal commuting time. Of course,
it might still be that movers who increase their commuting time do so
because of bad luck and have no better alternative (e.g. because they were
fired). The degree of compensation associated with moving was tested based
on observations for which consecutive information on commuting time is
available, i.e., for the years 1985/90, 1990/93 for the Old German Laender;
1992/93 for the New German Laender; and 1993/95, 1995/98 and
1998/2003 for both. A new panel was generated, restricted to people who
either changed their job and/or their place of residence anytime between
the respective years. Missing information for the commuting time between
years is imputed following a simple rule. As long as respondents stay in
their job and in their residence, commuting time is carried forward to the
following year with missing information. Accordingly, for years in the past
when respondents stayed in the same job and residence, commuting time
is imputed backwards. If someone only moves once between years with
reported information on commuting time, commuting time can be imputed
throughout. The new panel thus restricts the variation in commuting time
to changes due to moving. It consists of episodes between two (old German
C The editors of the Scandinavian Journal of Economics 2008.
360 A. Stutzer and B. S. Frey
Laender 1992/93) and six (1985/90 and 1998/2003) annual observations.
Reports of life satisfaction right before and after somebody moves can be
taken into account and a temporary effect of having a new job or residence
can be captured empirically. The same specifications as in Table 1 were
estimated. Results are shown in Table 5.
Table 5. Compensation of people who relocate or change jobs (dependent variable:
satisfaction with life)
All All Change of Change of
movers movers residence job
(1) (2) (3) (4)
Commuting time −0.0041 −0.0019 −0.0047 −0.0027
(−3.10) (−1.08) (−1.39) (−1.00)
Commuting time squared 4.70e – 6 −0.852e – 6 0.026e – 3 0.011e – 3
(0.31) (−0.05) (0.67) (0.39)
Change of residence 0.1661 0.1379 0.1104 0.3158
(5.65) (5.61) (3.73) (2.38)
Change of job −0.1057 0.0179 0.1096 0.0457
(−4.11) (0.80) (0.96) (1.60)
Individual characteristicsa Yes Yes Yes Yes
Individual fixed effects No Yes Yes Yes
Year fixed effects Yes Yes Yes Yes
F-test (Prob. > F) 0.000 0.035 0.169 0.306
Commuting time = 0 and
commuting time squared = 0
Effect of one hour of −0.230 −0.115 −0.192 −0.122
commuting (−6.09) (−2.20) (−1.88) (−1.45)
No. of observations 25,712 25,712 9,818 11,052
No. of individuals 5,560 5,560 2,316 3,031
(5) (6) (7) (8)
Commuting time −0.0037 −0.0019 −0.0027 −0.0017
(−6.79) (−2.59) (−1.76) (−1.49)
Change of residence 0.1660 0.1379 0.1100 0.3157
(5.65) (−5.61) (3.72) (2.38)
Change of job −0.1057 0.0179 0.1085 0.0454
(−4.11) (0.80) (0.95) (1.59)
Individual characteristicsa Yes Yes Yes Yes
Individual fixed effects No Yes Yes Yes
Year fixed effects Yes Yes Yes Yes
Effect of one hour of −0.224 −0.116 −0.163 −0.103
commuting (−6.79) (−2.59) (−1.76) (−1.49)
No. of observations 25,712 25,712 9,818 11,052
No. of individuals 5,560 5,560 2,316 3,031
Data source: GSOEP.
Notes: Partial correlation coefficients are from least-squares estimations. t-Values are in parentheses.
a
The same control variables for individual characteristics as in Table 1 are included.
C The editors of the Scandinavian Journal of Economics 2008.
Stress that doesn’t pay: the commuting paradox 361
Compared to the results in Table 1, movers report a smaller reduction
in life satisfaction when commuting time is increased. The effect for one
hour is −0.115 units and thus about half the size of the effect found in the
baseline estimation. However, this effect is still substantial and statistically
significant. In the fixed-effects estimation for all movers, an F-test rejects
that commuting time and commuting time squared are jointly equal to
zero. Less can be said about whether there is a systematic difference in the
negative effect of commuting for people who only change their residence or
only change their job. While the effect seems larger for people who relocate
than for people who change their job, the standard errors are too large to
draw a statistically valid conclusion. People who change their residence
experience, on average, temporarily higher life satisfaction in their new
home.12
In sum, people who change their job and/or their residence still experience
reduced life satisfaction if their new arrangement involves longer
commuting. While the smaller effect size hints to some sort of moving
costs (that explain part of the overall negative correlation), the phenomenon
remains partly unexplained.
V. Towards Behavioral Explanations
There is yet another reaction to the general result of this study. Individuals’
decisions concerning commuting cannot be fully understood within the
traditional economics framework. It is an issue “[w]here economics stops
short” (Economist, 1998, special issue on commuting). Inspiration from
other social sciences may complement an economic analysis of commuting
behavior. Most prominent are insights from psychology that have been
successfully integrated into a new cross-disciplinary field of economics
and psychology, as in e.g. Rabin (1998), Camerer, Loewenstein and Rabin
(2003) and Frey and Stutzer (2007).
There are at least two lines of reasoning that could contribute to a better
understanding of people’s commuting behavior. First, people might not be
capable of correctly assessing the true costs of commuting for their wellbeing.
They might rely on inadequate intuitive theories when they predict
how they are affected by commuting. In particular, they may make mistakes
when they predict their adaptation to daily commuting stress.13
It has, for
12
Specifications (3) and (4) report effects for a change of residence as well as a change
of job because some people (who again move later on) have just moved before the initial
observation with reported information about commuting time.
13
Excellent overviews on people’s difficulties in predicting future utility, as well as on adaptation,
are provided in Frederick and Loewenstein (1999) and Loewenstein and Schkade
(1999).
C The editors of the Scandinavian Journal of Economics 2008.
362 A. Stutzer and B. S. Frey
example, been found that people do not get used to random noise; see
Weinstein (1982). In contrast, people adapt to a large extent to higher
income; see e.g. Stutzer (2004). In the case of overestimated adaptation,
people systematically choose too long commuting times. A similar reasoning
is followed in Simonsohn (2006). He argues that commuting behavior
can be better understood in a framework of constructed preferences.
People come up with some reference level of commuting time or commuting
radius that they are only prepared to give up after experiencing
negative effects on their well-being. In a challenging study on people who
move from one US city to another, Simonsohn finds that people from a
city where the average commuting time of the population is high (or low)
also choose to commute more (or less) than average at their new place
of residence (keeping individuals’ own past commuting experience constant).
In the latter model, people can thus either commute too much or too
little.
Second, people’s weak will-power might be another reason why long
commutes are not compensated.14
Those with limited self-control and
insufficient energy might be induced to not even try to improve their
lot. This view corresponds to what some lay people seem to think. The
decision to start searching for a job closer to home or an apartment that
reduces commuting time is again and again postponed to the following
week. However, this can only be a partial explanation as there are indications
of a negative effect of commuting on reported life satisfaction even
for those individuals who have either changed their residence and/or their
job. Still, some people might not only smoke more and save less than they
would actually like, but also commute more than what they consider to be
optimal.
VI. Concluding Remarks
Commuting is for many people a time-consuming experience five days a
week. The journey from home to work and back is therefore an important
aspect of modern life; it affects people’s well-being and demands difficult
decisions about mobility on the labor and housing market.
Commuting is also interesting for economic research conceptually. The
decision to commute is hardly regulated. People are expected to freely
optimize. This environment allows for testing basic assumptions of the economic
approach, like market equilibrium. Positive and normative theories
14
The consequences of (economic) agents with self-control problems are discussed in e.g.
O’Donoghue and Rabin (1999) and Brocas and Carrillo (2003).
C The editors of the Scandinavian Journal of Economics 2008.
Stress that doesn’t pay: the commuting paradox 363
in urban and regional economics, as well as in public economics, rely on
a strong notion of equilibrium. It is assumed that people who can move
freely and change jobs arbitrage away any utility differentials between people,
whether they are due to residential characteristics or due to covering
distance, ceteris paribus.
In our test with panel data on subjective well-being for Germany, we
find, contrary to the prediction of equilibrium location theory, a large negative
effect of commuting time on people’s satisfaction with life. People
who commute 22 minutes (one way), which is the mean commuting time
in Germany, report, on average, a 0.103-point lower satisfaction with life.
This phenomenon is robust to a wide range of possible response biases,
and it is not explained by compensation at the level of households. If
people are aware of the full costs of commuting, the finding shows the importance
of moving costs of trapped individuals. However, an albeit small
effect also holds for people who either change their job or their place of
residence and so have the opportunity of re-optimizing their commuting
situation. There might, also for them, well be an explanation in terms of
economic costs not yet found and thus not yet incorporated into the analysis.
This cost factor would be interesting to know, because it potentially
relates to a sizable loss in well-being and should be explicitly modeled
in urban and public economics. Until an adequate rational-choice explanation
has been provided, we propose the general result to be a “commuting
paradox”.
Research along the lines studied in the field of economics and psychology
may well provide a better understanding of people’s decisions about
where to live and work and how long the commuting time may be. We
favor an explanation based on wrongly predicted adaptation. Decisions
about commuting involve a difficult trade-off between socially positively
sanctioned income and some loss of spare time that is difficult to assess.
Other behavioral anomalies may also play an important role in commuting
decisions. Limited will-power and loss aversion, however, may better
explain why people remain in an inferior status quo rather than why people
who spend more time commuting suffer lower well-being. It will be
a major challenge for future research to discriminate between alternative
“behavioral” explanations of the phenomenon.
For many people, commuting seems to encompass stress that does
not pay off. A better understanding of this phenomenon should provide
valuable insights on the institutional and behavioral restrictions to compensation.
Moreover, it may help commuters to increase their individual
well-being.
C The editors of the Scandinavian Journal of Economics 2008.
364 A. Stutzer and B. S. Frey
Appendix
Table A1. Descriptive statistics
Mean Std. dev. Fraction (%)
Satisfaction with life 7.143 1.67 Male 55.6
Commuting time 22.079 18.35 Female 44.4
(in minutes) Head of household or spouse 86.1
Commuting distance 12.698 19.11 Child of head of household 12.8
(in kilometers) Not child of head of household 1.1
Working hours 38.954 11.68 Single, no partner 18.8
(hours per week) Single, with partner 6.4
Real net labor income 1,326.601 933.06 Married 65.0
per month, 2000 euros Separated, with partner 0.3
Real net household 2,799.666 1,639.37 Separated, no partner 1.3
income per month, Divorced, with partner 2.5
2000 euros Divorced, no partner 3.9
Age 38.907 11.77 Widowed, with partner 0.3
Years of education 11.505 3.13 Widowed, no partner 1.2
No. of household members 3.117 1.35 Spouse living abroad 0.2
No children in household 54.1
1 child in household 23.6
2 children in household 16.8
3 or more children in household 5.5
Employed 83.5
Self-employed 16.5
Western Germany 79.8
Eastern Germany 20.2
National 83.1
EU foreigner 6.8
Other foreigner 10.1
First interview 4.9
Data source: GSOEP.
References
Alonso, W. (1964), Location and Land Use: Toward a General Theory of Land Rent, Harvard
University Press, Cambridge, MA.
Brocas, I. and Carrillo, J. D. (2003), Information and Self-control, in I. Brocas and J. D.
Carrillo (eds.), The Psychology of Economic Decisions, Vol. 1: Rationality and Well-being,
Oxford University Press, Oxford.
Camerer, C., Loewenstein, G. and Rabin, M. (eds.) (2003), Advances in Behavioral Economics,
Russell Sage Foundation Press and Princeton University Press, New York and
Princeton, NJ.
Clark, W. A. V. and van Lierop, W. F. J. (1986), Residential Mobility and Household Location
Modelling, in P. Nijkamp (ed.), Handbook of Regional and Urban Economics, Vol. I,
North-Holland, Amsterdam.
Clark, A. E., Frijters, P. and Shields, M. A. (2008), Relative Income, Happiness and Utility:
An Explanation for the Easterlin Paradox and Other Puzzles, Journal of Economic
Literature. 46, 95–144.
Conley, J. P. and Konishi, H. (2002), Migration-proof Tiebout Equilibrium: Existence and
Asymptotic Efficiency, Journal of Public Economics 86, 243–262.
C The editors of the Scandinavian Journal of Economics 2008.
Stress that doesn’t pay: the commuting paradox 365
Crampton, G. R. (1999), Urban Labour Markets, in E. S. Mills and P. Cheshire (eds.),
Handbook of Regional and Urban Economics, Vol. III, North-Holland, Amsterdam.
Diener, E., Suh, E. M., Lucas, R. E. and Smith, H. L. (1999), Subjective Well-being: Three
Decades of Progress, Psychological Bulletin 125, 276–303.
Di Tella, R. and MacCulloch, R. (2006), Some Uses of Happiness Data in Economics, Journal
of Economic Perspectives 20, 25–46.
Economist (1998), Commuting, Survey, September 3, 1998.
EPA (Environmental Protection Agency) (2001), Commuter Choice Leadership Initiative:
Facts and Figures, EPA 420-F-01-023, EPA, Washington, DC.
Ferrer-i-Carbonell, A. and Frijters, P. (2004), How Important is Methodology for the Estimates
of the Determinants of Happiness?, Economic Journal 114, 641–659.
Frederick, S. and Loewenstein, G. (1999), Hedonic Adaptation, in D. Kahneman, E. Diener
and N. Schwarz (eds.), Well-being: The Foundations of Hedonic Psychology, Russell Sage
Foundation, New York.
Frey, B. S. and Stutzer, A. (2000), Happiness, Economy and Institutions, Economic Journal
110, 918–938.
Frey, B. S. and Stutzer, A. (2002a), What Can Economists Learn from Happiness Research?,
Journal of Economic Literature 40, 402–435.
Frey, B. S. and Stutzer, A. (2002b), Happiness and Economics: How the Economy and
Institutions Affect Human Well-being, Princeton University Press, Princeton, NJ.
Frey, B. S. and Stutzer, A. (2007), Economics and Psychology. A Promising New Crossdisciplinary
Field, MIT Press, Cambridge, MA.
Huriot, J.-M. and Thisse, J. F. (eds.) (2000), Economics of Cities: Theoretical Perspectives,
Cambridge University Press, Cambridge.
Kahneman, D., Krueger, A. B., Schkade, D. A., Schwarz, N. and Stone, A. A. (2004), A
Survey Method for Characterizing Daily Life Experience: The Day Reconstruction Method,
Science 306, 1776–1780.
Koslowsky, M., Kluger, A. N. and Reich, M. (1995), Commuting Stress: Causes, Effects, and
Methods of Coping, Plenum Press, New York.
Layard, R. (2005), Happiness: Lessons from a New Science, Penguin, London.
Lepper, H. S. (1998), Use of Other-reports to Validate Subjective Well-being Measures, Social
Indicators Research 44, 367–379.
Loewenstein, G. and Schkade, D. (1999), Wouldn’t It Be Nice? Predicting Future Feelings,
in D. Kahneman, E. Diener and N. Schwarz (eds.), Well-being: The Foundation of Hedonic
Psychology, Russell Sage Foundation, New York.
McMillen, D. P. and Singell, L. D. (1992), Work Location, Residence Location, and the
Intraurban Wage Gradient, Journal of Urban Economics 32, 195–213.
Muth, R. F. (1969), Cities and Housing, University of Chicago Press, Chicago.
Novaco, R. W., Stokols, D. and Milanesi, L. C. (1990), Subjective and Objective Dimensions
of Travel Impedance as Determinants of Commuting Stress, American Journal of
Community Psychology 18, 231–257.
O’Donoghue, T. and Rabin, M. (1999), Doing It Now or Later, American Economic Review
89, 103–124.
Rabin, M. (1998), Psychology and Economics, Journal of Economic Literature 36, 11–46.
Schwarz, N. and Strack, F. (1999), Reports of Subjective Well-being: Judgmental Processes
and their Methodological Implications, in D. Kahneman, E. Diener and N. Schwarz
(eds.), Well-being: The Foundations of Hedonic Psychology, Russell Sage Foundation,
New York.
Simonsohn, U. (2006), New-Yorkers Commute More Everywhere: Contrast Effects in the
Field, Review of Economics and Statistics 88, 1–9.
Small, K. A. (1992), Urban Transportation Economics, Harwood, Chur.
C The editors of the Scandinavian Journal of Economics 2008.
366 A. Stutzer and B. S. Frey
So, K. S., Orazem, P. F. and Otto, D. M. (2001), The Effects of Housing Prices, Wages,
and Commuting Time on Joint Residential and Job Location Choices, American Journal
of Agricultural Economics 83, 1036–1048.
Stutzer, A. (2004), The Role of Income Aspirations in Individual Happiness, Journal of
Economic Behavior and Organization 54, 89–109.
Stutzer, A. and Frey, B. S. (2004), Reported Subjective Well-being: A Challenge for Economic
Theory and Economic Policy, Schmollers Jahrbuch 124, 1–41.
Tiebout, C. M. (1956), A Pure Theory of Local Expenditure, Journal of Political Economy
64, 416–424.
Timothy, D. and Wheaton, W. C. (2001), Intra-urban Wage Variation, Employment Location,
and Commuting Times, Journal of Urban Economics 50, 338–366.
van Ommeren, J. (2000), Commuting and Relocation of Jobs and Residences, Ashgate,
Aldershot.
van Ommeren, J., Rietveld, P. and Nijkamp, P. (1997), Commuting: In Search of Jobs and
Residence, Journal of Urban Economics 42, 402–421.
van Ommeren, J., van den Berg, G. J. and Gorter, C. (2000), Estimating the Marginal Willingness
to Pay for Commuting, Journal of Regional Science 40, 541–563.
Weinberg, D. H., Friedman, J. and Mayo, S. K. (1981), Intraurban Residential Mobility: The
Role of Transactions Costs, Market Imperfections, and Household Disequilibrium, Journal
of Urban Economics 9, 332–348.
Weinstein, N. D. (1982), Community Noise Problems: Evidence against Adaptation, Journal
of Environmental Psychology 2, 87–97.
Wildasin, D. E. (1987), Theoretical Analysis of Local Public Economics, in E. S. Mills (ed.),
Handbook of Regional and Urban Economics, Vol. II, North-Holland, Amsterdam.
First version submitted August 2005;
final version received January 2008.
C The editors of the Scandinavian Journal of Economics 2008.