Are Academics Messy?

Testing the Broken

Windows Theory with a Field Experiment in
the Work Environment ∗
JOÃO RAMOS,∆ BENNO TORGLER±
∆
PricewaterhouseCoopers Australia; ±Queensland University of Technology

We test the broken windows theory using a field experiment in a shared area of an academic workplace
(the department common room). More specifically, we explore academics’ and postgraduate students’
behavior under an order condition (a clean environment) and a disorder condition (a messy
environment). We find strong evidence that signs of disorderly behavior trigger littering: In 59% of the
cases, subjects litter in the disorder treatment as compared to 18% in the order condition. These results
remain robust in a multivariate analysis even when controlling for a large set of factors not directly
examined by previous studies. Overall, when academic staff and postgraduate students observe that
others have violated the social norm of keeping the common room clean, all else being equal, the
probability of littering increases by around 40%.

1. INTRODUCTION
An understanding of the triggers for antisocial and petty criminal behavior is
important to developing better communities. One theory that has strongly
influenced law enforcement strategies in several U.S. cities (e.g., New York,
Chicago, Baltimore, Boston and Los Angeles) is the broken windows theory
(BWT), which proposes that “signs of inappropriate behavior like graffiti or
broken windows lead to other inappropriate behavior (e.g., littering or stealing)”
(Keizer et al., 2008:1685). Based on this assumption, these communities have

∗
We would like to thank Nicholas McMeniman for his contribution to the conduct of this field
experiment. Thanks are also due to Ernst Fehr, Bruno Frey, Lorenz Goette, Alison Macintyre
and two anonymous referees for helpful comments and suggestions. Corresponding Author:
Benno Torgler: Queensland Behavioural Economics Group (QuBE), School of Economics and
Finance, Queensland University of Technology (Australia); EBS Business School, ISBS, EBS
Universität für Wirtschaft und Recht (Germany); and CREMA – Center for Research in
Economics, Management and the Arts (Switzerland); [Link]@[Link].

DOI:10.1515/1555-5879.1617

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
564 / REVIEW OF LAW AND ECONOMICS 8:3, 2012

attempted to maintain order by dealing more aggressively with minor offenses

(Harcourt and Ludwig, 2006). Harcourt and Ludwig (2006:272), however, criticize this
widespread policy on the grounds that “remarkably little is known about the
effects of broken windows.” In fact, some researchers contend that the
enthusiasm for the broken windows strategy is misplaced (Taylor, 2000; Harcourt,
2001), especially given Sampson and Raudenbush’s (1999) finding of only
moderate effects that were not robust. Harcourt and Ludwig (2006), find no
support for BWT when drawing on data from a social experiment in which low-
income families from communities with high rates of social disorder were
randomly assigned housing vouchers to move to less disadvantaged and
disorderly communities. Funk and Kugler (2003), on the other hand, apply a
dynamic approach to quarterly time series data from Switzerland and find
evidence that an increase in minor theft triggers a substantial increase in
subsequent, more severe crimes such as burglary or robbery (but not vice versa).
Reporting on six different controlled field experiments conducted in common
public spaces in the town of Groningen in the Netherlands, Keizer et al. (2008)
also present evidence that as a certain norm-violating behavior becomes more
common, it negatively influences conformity to other norms and rules. Hence,
not only are such empirical studies limited, but the results are mixed.
Research into BWT has also been criticized for being largely correlational and
for failing to provide concrete causal evidence (Keizer et al., 2008:1681),
shortcomings that indicate the potential usefulness of an experimental approach.
In fact, previous experimental investigations into littering and norms, albeit not
BWT-based (e.g., Cialdini, Reno, et al., 1990; Cialdini, Kallgren, et al., 1991), do show that
littering occurs more often in places already littered than in clean areas. This
finding suggests that if people notice that other individuals have been littering,
their willingness to litter increases, thereby reducing the moral constraints that
would ordinarily compel them to behave in a socially acceptable manner. Thus,
individual behavior is likely to be influenced by people’s perception of the
behavior of other citizens. Indeed, Torgler et al. (2009), work with a dataset of
over 32,000 observations, finding that if individuals believe littering in a public
place is common, then the justifiability of littering increases, but if they believe
others to be compliant, it decreases. It is unclear whether such results carry
over to littering in private places such as a workplace. Similarly, it should also
be noted that although BWT has previously been applied in shared public
settings or residential neighborhoods, it has yet to be tested in smaller, more
enclosed environments such as the workplace. Therefore, in this paper, we test
whether BWT can be applied in a quasi-private enclosed setting in a work
environment (i.e., a common room or lounge) using a controlled field
experiment with a relatively homogenous group of individuals (academics). The

Review of Law & Economics, © 2012 by De Gruyter

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
 Are Academics Messy? / 565

primary objective of the study is to gauge whether BWT can be applied more
generally to contexts other than those reported in the extant literature. We do
acknowledge that collective-action problems may arise related to the notion of
conditional cooperation (for experimental evidence, see, e.g., Fischbacher et al., 2001; Falk
and Fischbacher, 2002). A field experimental setting is particularly useful because,
by enabling researchers to focus on a homogenous group (in this case,
academics), it addresses the criticism that the differences across neighborhoods
in previous studies are driven by the unobservable individual characteristics of
neighborhood residents and related problems of self-selection (Harcourt and
Ludwig, 2006). For example, demographic factors, changes in drug markets,
organizational reforms within the police department, increased incarceration,
or a reduction in unemployment could also have contributed to the drop in
crime (Harcourt, 2001). Thus, prior studies that could not control for population
were also unable to isolate potential composition effects (e.g., specific groups
of individuals that may change their place of action when the environment is
messy). It might also be assumed that academics, being among society’s most
educated, might generally be sound in judgment and, in the context of littering,
engage in mostly compliant behavior. However, research shows that academics’
judgment is in fact skewed by the same self-enhancing sociocognitive
tendencies that influence the general population (Van Lange et al., 1997; Cross, 1977).
Exploring a private setting allows us to look at BWT effects in a non-anonymous
setting. In previous experimental settings such as that used by Keizer et al. (2008),
anonymity allowed subjects to easily get away with violating the norm, but more
importantly they were out of sight of the observer (experimenter). In this study on
the other hand, the observer was always present in the room. Thus, subjects were
aware that their norm-violating behavior could be observed and socially
sanctioned. Therefore, the results of our study show the impact of the disorder
cues. This is an important feature of the present work that addresses a limitation
of the previous work on the BWT. We will show that disorder also induces norm-
violating behavior when there is a high chance of being (socially) sanctioned.
Why, then, is it important to focus on littering? First, litter in public places has
been recognized as a major public health and safety hazard, one that diminishes
the aesthetic appearance of public places (Ackerman, 1997). Littering is also
considered to be one of the most neglected yet most visible forms of
environmental degradation (Finnie, 1973:123). Hansmann and Scholz (2003:753)
define litter simply as “the careless, incorrect disposal of minor amounts of
waste,” and items may be discarded either actively or passively (Sibley and Liu, 2003)
in such locations as parks, roads, paths, camping grounds, cafes, stores or other
public buildings. The most frequently discarded items include cigarettes, bottles
and other glass or plastic containers, napkins, bags, tissues, take-away food

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
566 / REVIEW OF LAW AND ECONOMICS 8:3, 2012

packages and snack wrappers, some of which are nondegradable and thus have
negative consequences for natural areas. Littering is not only visually ugly, it is
also potentially dangerous: a discarded live cigarette or a glass bottle can cause a
devastating forest fire (Crump et al., 1977). Hence, in addition to the costs of
employing someone to remove the litter, littering also engenders additional
environmental costs. Since the production of litter is a collective action, from a
socioeconomic viewpoint, refraining from littering can be seen as a cooperative
and social behavior. That is, the benefits derived from keeping outdoor public
places and the work environment clean are enjoyed by the wider community in
terms of the positive amenities of the area, whereas the costs of producing the
“public good character” are private (Anand, 2000).

2. METHOD
To test the broken window theory, we conducted a small field experiment at the
School of Economics and Finance at the Queensland University of Technology
in Brisbane (Australia). Being members of the school, we could both control for
the impact of several variables that previous studies have neglected (for lack of
observability) and ensure that the subjects were unaware of being involved in
such a field experiment. The setting was the common room shared by almost all
the school’s faculty, administrative staff, and postgraduate students, and the
subjects were individuals that used the common room between 12:00 pm and
1:00 pm (i.e., during lunch time). One author and a graduate student sat in the
common room under the pretence of eating lunch or reading a newspaper,
while observing and surreptitiously recording the number of academics in the
room and their behavior. Knowing these academics personally allowed us to ex
post extrapolate personal characteristics seldom collected in previous field
experiments, including age, sex, field of research, and academic position. The
experiment was conducted in May 2009 over a period of six days. Given the
small size of the department, extending the period could have been problematic:
the validity of such experimentation requires that the subjects remain unaware
of being monitored and act naturally. In artificial laboratory environments, in
contrast, test subjects are keenly aware that their behavior is being monitored
and are prone to change their normal behavior, making the results difficult to
generalize (Levitt and List, 2009). Moreover, given the short time interval, we ran
the experiment in the order condition for three days followed by three days in
the disorder condition which avoided the spread of the disorder manipulation to
the control days (order condition).
Like Keizer et al. (2008), we distinguish between a contextual norm, whose
indications we manipulated and whose violation participants witnessed, and a
target norm, which participants themselves violated. The dependent variable is

Review of Law & Economics, © 2012 by De Gruyter

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
 Are Academics Messy? / 567

whether a common room user violates this target norm. We define the disorder
condition (treatment group) as one in which the contextual norm is violated
and the order condition (control group) as one in which it is not. We predicted
that participants would violate the target norm more frequently in the presence
of a contextual norm violation. An orderly environment is our control
treatment (see Figure 1). In this condition, clean cutlery, crockery, and drinking
glasses are stored in the common room cabinets and it is expected that any
used wares will be placed in the school’s dishwasher (in the same room) for
later washing. We therefore designate this behavior as the injunctive norm or
the most appropriate behavior in this situation. Any participant not placing
used common room utensils, plates, and so forth in the dishwasher is thus
considered to have littered and so violated the target norm.
We manipulated the indications of contextual norm violation in our disorder
condition by placing used cutlery, crockery, and drinking glasses in the common
room sink. To further reinforce this disorder condition, we made the common
room generally untidy by spreading newspapers, magazines, and sugar packets
around and placing litter on the floor near the trash can (see Figure 2). In addition
to making it immediately noticeable that the room was messier than usual, these
actions established evidence of a cross-norm inhibition effect, since not placing
cutlery in the dishwasher fosters violation of the norms related to room tidiness.

Figure 1: Order Condition Figure 2: Disorder Condition

3. RESULTS
We obtained 49 observations for each of the two treatments; the order
condition (ORDER, tidy common room) and the disorder condition
(DISORDER, untidy common room). Participants in the order condition
“littered” 18% of the time compared to 59% in the disorder condition (see
Figure 3). The results of a two-sample Wilcoxon rank-sum (Mann-Whitney)
test indicate that this difference between the order and disorder condition is
highly statistically significant at the 1% level (z = –4.125).

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
568 / REVIEW OF LAW AND ECONOMICS 8:3, 2012

Figure 3: Proportion of times that subjects littered

under the order treatment versus the disorder treatment

A total of 38 participants were observed, comprising 27 unique participants in

the disorder condition and 22 in the order condition. There are 11 participants
for whom we have observations in both conditions. We present a within-
subject analysis in Figure 4 that explores whether the same people behave
differently in the order and disorder condition. In 70% of the cases they litter
in the disorder condition compared with 22% in the order condition. This
difference is also highly statistically significant (z =-3.396).

Figure 4: Littered under the order treatment versus the disorder treatment using a
within-subject design (same individuals in both treatments)

The experiment was run across six days, and there was one day (Friday) where
the common room was once a disorder and once an order condition. This
allows us to compare littering on the same day which decreases uncertainty

Review of Law & Economics, © 2012 by De Gruyter

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
 Are Academics Messy? / 569

regarding compositional effects stemming from certain researchers coming

only on specific days. The results are presented in Figure 5. In 42% of the
cases, subjects littered in the order condition compared with 67% in the
disorder condition. Thus, we also observe substantial differences in this case
between the control and the treatment group.

Figure 5: Proportion of times that subjects littered under the order treatment
versus the disorder treatment on the same weekday (Friday)

Nonetheless, since our descriptive analysis only gives information on the raw
and not the partial effects, we also test whether the difference would remain
statistically significant in a multivariate analysis. Because the dependent variable
in the analysis – that is, whether individuals litter (value 1) or not (value 0) – is
nonlinear and binary, we calculate the marginal effects at the multivariate point
of means to find the quantitative effect of any given independent variable.
Table 1 presents the results for our five specifications. In specification (1), we
use only DISORDER as the independent variable. In specification (2), we add
in sociodemographic factors such as gender, age, and job characteristics;
namely, whether the individual has an economics or finance background
(ECONOMIST = 1), is academic staff (ACADEMIC STAFF = 1), or is a
postgraduate student (ACADEMIC STAFF = 0). Next, in specification (3), we
add in a PEOPLE PRESENT variable that measures whether the littering
behavior changes with the number of individuals in the room. As a further
robustness check, in regressions (4), (5), and (6), we integrate an additional
group of specifications in which the standard errors by subject are clustered:
such clustering not only reveals unobserved individual-specific characteristics,
but allows us to take into account multiple observations for subjects without

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
570 / REVIEW OF LAW AND ECONOMICS 8:3, 2012

losing degrees of freedom, therefore ruling out compositional effects. Because

the field experiment was conducted over a period of six days, we also
incorporate two dummy variables in specification (5) that control for a
MONDAY or FRIDAY effect. Finally, in specification (6), we control for the
POSITION (RANK) of the subjects (postgraduate student = 1; postdoctoral
fellow, lecturer, senior lecturer = 2; associate professor, professor = 3).

4. DISCUSSION
The results paint a robust picture that is consistent with previous results. The
coefficient of the disorder variable is always statistically significant (mostly at the
1% level) and the marginal effects are also quite large. All else being equal, a
disorder condition increases the probability of littering by between 26 and 45%.
When the contextual norm violation (signs of disorder) is present, subjects are
more likely to violate the target norm than when it is absent. This finding is in
contrast to the previous BWT studies discussed in our introduction that had
offered inconclusive evidence of a contextual norm violation effect in a public
space, neighborhood, or city. On the contrary, in our investigational
environment, signs of norm violation clearly promulgate further norm violations.
Our field experiment thus demonstrates that the broken windows theory holds
in relatively micro settings such as the workplace.
The fact that we knew the subjects also allowed us to identify several
interesting demographic characteristics whose inclusion was restricted in
previous BWT studies (e.g., Keizer et al., 2008). For example, our results show that
individuals aged 50 and over are more likely to litter than our reference group
(those under 30), with marginal effects around 60%. Likewise, senior staff
members are more likely to litter than junior staff, although the coefficient is
only statistically significant at the 10% level. We also find, however, that (in line
with previous findings on conditional norm violating behavior) the presence of
a large number of individuals in the room discourages littering, although this
coefficient is not always statistically significant. There is also evidence of a
Monday effect; that is, having spent the weekend at home (where violation of
the littering norm is most costly) impacts an individual’s behavior upon
returning to work at the start of the workweek. Apparently, spending the
weekend at home reinforces socially acceptable norms.
Next, we conduct several robustness tests (see Table 2). First, we use a
dummy to control for those individuals where we have more than one
observation (REPETITION, specification 7). Next, we also use a dummy
variable to control for those individuals that have been observed in both
treatments (WITHIN SUBJECT, specification 8). In addition, we conduct a
within-subject analysis looking only at those individuals who were present in

Review of Law & Economics, © 2012 by De Gruyter

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
 Are Academics Messy? / 571

both the order and the disorder treatment (specifications 9 to 12). In all
estimations the standard errors are adjusted for clustering on individuals. We can
see clearly that the marginal effects for our key variable DISORDER are
substantially larger in the within-subject design and range between 47 and 59
percent. The coefficient for DISORDER also remains statistically significant
when controlling for the two variables REPETITION and WITHIN SUBJECT.
Overall, the control variables show a similar picture as Table 1. People age 50
and over have a very high probability of littering. The Monday effect remains
statistically significant in the first two specifications in Table 2, but loses its
effect in the within-subject design, possibly because there are less observations.
The coefficient for PEOPLE PRESENT is on the border of being statistically
significant in the first two specifications. The variable POSITION was
statistically significant at the 10% level in Table 1 but loses its statistical
significance in Table 2.
In addition, we also ran regressions with subject-fixed effects, restricting
ourselves to the within-subject analysis. Our key variable DISORDER remains
statistically significant at the 1% level. For example, specification (12) with
individual fixed effects reports a z-value of 4.48 and a marginal effect of 47%.
A key comparative strength of this study is that we have substantial
information about the participants, thus we also explore whether factors such
as gender, age and status position interact with the contextual norm violation.
For example, are people in a “lower status position” (measured with the
variable POSITION, ACADEMIC STAFF or AGE) more influenced by cues
concerning contextual norm violation? We explore this aspect in a within-
subject design. None of the interaction terms were statistically significant.
However, we found some further interesting results when interacting
FEMALE and AGE (recoding the dummies to a single variable ranking from 1
to 4, 4=age50) with the variable PEOPLE PRESENT. In both cases the
interaction term was statistically significant, however, with a different sign.
When a person is female an additional person in the room reduces the
probability of littering. On the other hand, the interaction effect
AGE*PEOPLE PRESENT was positive. It seems that older people are
triggered to litter more if more people are around. This might be a case of
signalling seniority. However, on the other hand, the interaction effect
POSITION*PEOPLE PRESENT was not statistically significant.

5. LIMITATIONS
Although our results are robust, clear and significant, it should be noted that
this field experiment is subject to a size limitation, having been conducted over

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
572 / REVIEW OF LAW AND ECONOMICS 8:3, 2012

only six days among a small university faculty. This allowed the conduct of
such an experiment without generating suspicion. However, it might be
valuable to apply this approach for a longer observation period in a broader
setting; for example, a large industrial workplace in which accurate
identification of subject demographics is also feasible. A further limitation of
the study is that we are not able to rule out the possibility that the disorder
condition may have created the perception that the dishwasher was full. Hence,
it is conceivable that some individuals littered under the impression that they
could not have done better. There should also not be any expectation that the
problem will be solved by the cleaning service the next morning. The cleaning
service is only responsible for clearing the rubbish every morning and is not for
dealing with utensils and plates. The cleaners certainly do not use the
dishwasher or clean the sink, as it is perceived that this is the social
responsibility of staff members. One can therefore rule out the alternative
explanation that subjects of the study might be accustomed to a clean room
and when the room suddenly became messy, they may have wanted to make
the cleaning service aware that the room should be cleaned. Nevertheless,
when such a “dishwasher is full” signal is at work, it changes the anticipated
ease of conforming to the target norm between conditions (conforming to the
norm of putting the dishes in the dishwasher requires effort, as one must first
remove the clean dishes from the dishwasher). However, a disorder condition
without glasses etc. in the sink may not have avoided this limitation. As soon as
one experimental subject put their dishes in the sink the same situation would
have emerged. The lack of control over such dynamics is a key disadvantage
when using a field experiment. Moreover, although we cannot rule out
individuals’ perception of a full dishwasher, as an informal rule the
administrative staff members generally empty the dishwasher after their
morning tea (between 9 and 10 am) so that cutlery and dishes are available for
lunch. Thus, during the hours leading up to lunch the dishwasher would have
been empty. This reduces the likelihood of subjects perceiving that the
dishwasher is full and takes into account that the observed subjects use the
common room on a regular basis. In addition, we checked that the dishwasher
was empty before the experiment (for each day).
In addition, the BWT not only emphasizes that one norm violation fosters
the violation of other norms but that it fosters more serious norm violations.
Based on the data collected, we have not been able to analyze that element in
our paper and have only explored the spread of a norm violation in the domain
of (very) minor crime.

Review of Law & Economics, © 2012 by De Gruyter

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
 Are Academics Messy? / 573

6. CONCLUSIONS
The novel contribution of this study is the use of a small controlled field
experiment in a shared academic workplace area (common room) to explore
whether the broken windows theory can help explain littering behavior. Our
results strongly suggest that signs of disorder in the common room lead to a
substantial increase in the probability of subjects violating the target norm (i.e.,
littering). Specifically, the descriptive analysis shows that 59% of subjects in the
disorder condition littered compared to only 18% in the order condition.
Another strength of this analysis, compared to previous research, is its ability
to control for a set of independent factors in a multivariate analysis in which
the statistically significant difference between the control and treatment groups
holds constant. Specifically, when academics see that other academics have
violated the social norm of keeping the common room clean, all else being
equal, the probability of their littering increases by around 40%.
The results can be interpreted as “lower-bound” effects as the behavior was
observed in a non-anonymous setting with one of the experimenters always
present in the common room. One can expect that individuals behave better if
they can be observed by others (in this case the experimenters). As such the
effects might be bigger in an environment where subjects feel anonymous.
In sum, our results suggest that preventing signs of disorder may be an
effective method of maintaining social norms of compliance. Our study also
contributes to the literature on workplace deviance. It has been shown that
workplace deviance such as theft, fraud, vandalism, sabotage, and voluntary
absenteeism are a pervasive and expensive problem for organizations (Bennett
and Robinson, 2000). We report that signs of disorder induce violations of a work
floor norm. Thus, the question arises whether it applies also to other company
norms and future research could provide further insights on how disorder
influences workplace deviance.

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
574 / REVIEW OF LAW AND ECONOMICS 8:3, 2012

Table 1: The Impact of Disorder on Littering

Probit Model (1) (2) (3) (4) (5) (6)
INDEPENDENT FACTORS
DISORDER 1.134*** 1.280*** 1.067*** 1.067*** 0.737** 0.809***
(4.11) (4.01) (3.14) (3.93) (2.50) (2.95)
0.408 0.453 0.381 0.381 0.259 0.284
MALE -0.277 -0.284 -0.284 -0.228 -0.251
(-0.73) (-0.73) (-0.69) (-0.48) (-0.52)
-0.105 -0.107 -0.107 -0.083 -0.092
ECONOMIST -0.272 -0.319 -0.319 -0.289 -0.186
(-0.75) (-0.86) (-0.73) (-0.60) (-0.37)
-0.104 -0.121 -0.121 -0.106 -0.068
ACADEMIC STAFF 0.425 0.611 0.611 0.563
(1.20) (1.63) (1.62) (1.33)
0.160 0.228 0.228 0.203
AGED 30−39 0.164 0.089 0.089 -0.108 -0.206
(0.44) (0.23) (0.18) (-0.21) (-0.39)
0.062 0.033 0.033 -0.038 -0.072
AGED 40−49 -0.589 -0.654 -0.654 -0.861* -0.911*
(-1.12) (-1.21) (-1.35) (-1.71) (-1.77)
-0.199 -0.215 -0.215 -0.253 -0.264
AGED 50 AND OVER 1.599*** 1.719*** 1.719*** 1.758*** 1.692***
(3.28) (3.41) (3.41) (3.38) (3.35)
0.571 0.604 0.604 0.619 0.601
PEOPLE PRESENT -0.155* -0.155 -0.177* -0.164
(-1.89) (-1.58) (-1.72) (-1.64)
-0.057 -0.057 -0.063 -0.059
MONDAY -1.223* -1.195**
(-1.96) (-1.99)
-0.342 -0.337
FRIDAY 0.391 0.345
(1.43) (1.44)
0.145 0.128
POSITION (RANK) 0.532*
(1.67)
0.191
Prob > chi2 0.000 0.000 0.000 0.000 0.000 0.000
Pseudo R2 0.137 0.286 0.315 0.315 0.373 0.380
# of observations 98 98 98 98 98 98
Notes: z-values are in parentheses; marginal effects are in italics. *, **, and *** indicate
statistical significance at the 5%, 1%, and 0.1% levels, respectively. The following variables
comprise the reference group: ORDER TREATMENT, FEMALE, FINANCE,
POSTGRADUATE STUDENT, AGED BELOW 30, TUESDAY−THURSDAY.

Review of Law & Economics, © 2012 by De Gruyter

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
 Are Academics Messy? / 575

Table 2: Within-Subject Analysis and Dealing with Repeated Observations

Probit Model
(7) (8) (9) (10) (11) (12)
Dep. Variable: Littering
Within-subject analysis (only participants where observations are
available in both treatments)
INDEPENDENT FACTORS
DISORDER 0.776** 0.817*** 1.316*** 1.668*** 1.457*** 1.281***
(2.52) (2.61) (5.63) (5.56) (4.10) (4.48)
0.274 0.285 0.486 0.592 0.531 0.474
MALE -0.297 -0.230 -0.399 -0.284 -0.318
(-0.61) (-0.46) (-0.82) (-0.60) (-0.53)
-0.109 -0.084 -0.158 -0.113 -0.126
ECONOMIST -0.152 -0.133 0.425 0.352 0.525
(-0.30) (-0.25) (1.08) (0.85) (0.94)
-0.055 -0.048 0.167 0.139 0.204
ACADEMIC STAFF 0.248 0.182 0.080 0.326 0.067
(0.43) (0.35) (0.22) (0.73) (0.07)
0.090 0.065 0.032 0.129 0.027
AGED 30−39 -0.143 -0.163 1.137* 1.037* 0.862
(-0.26) (-0.29) (1.75) (1.73) (1.25)
0.191 -0.057 0.428 0.394 0.333
AGED 40−49 -0.985* -0.999** -0.361 -0.480 -0.558
(-1.90) (-1.99) (-0.77) (-1.34) (-1.24)
0.126 -0.281 -0.141 -0.184 -0.210
AGED 50 AND OVER 1.784*** 1.725** 1.315** 1.421*** 1.460***
(3.26) (2.89) (2.48) (3.32) (2.72)
0.625 0.610 0.469 0.497 0.513
PEOPLE PRESENT -0.175* -0.156 -0.217 -0.236
(-1.74) (-1.53) (-1.22) (-1.33)
-0.063 -0.056 -0.087 -0.094
MONDAY -1.277** -1.376** -0.796
(-1.98) (-2.17) (-1.30)
-0.356 -0.369 -0.295
FRIDAY 0.320 0.355 0.344
(1.35) (1.48) (0.92)
0.119 0.131 0.136
POSITION (RANK) 0.398 0.351 0.194
(0.96) (0.88) (0.32)
0.143 0.125 0.077

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
576 / REVIEW OF LAW AND ECONOMICS 8:3, 2012

REPETITION -0.435 -0.687

(-0.87) (-1.06)
-0.165 -0.263
WITHIN SUBJECT 0.402
(0.90)
0.143
Clustering over individuals YES YES YES YES YES YES
Prob > chi2 0.000 0.000 0.000 0.000 0.000 0.000
Pseudo R2 0.383 0.392 0.178 0.274 0.327 0.361
# of observations 98 98 50 50 50 50
Notes: z-values are in parentheses; marginal effects are in italics. *, **, and *** indicate
statistical significance at the 5%, 1%, and 0.1% levels, respectively. The following variables
comprise the reference group: ORDER TREATMENT, FEMALE, FINANCE,
POSTGRADUATE STUDENT, AGED BELOW 30, TUESDAY−THURSDAY, NO
REPETITIONS (dummy=1: individuals with only one observation), NOT WITHIN SUBJECT
(dummy=1: individuals who were observed in only one treatment).

REFERENCES
Ackerman, Frank. 1997. Why Do We Recycle? Markets, Values, and Public Policy.
Washington, DC: Island Press.
Anand, P.B. 2000. “Cooperation and the Urban Environment: An Exploration,” 36
Journal of Development Studies 30-58.
Bennett, Rebecca J., and Sandra L. Robinson. 2000. “Development of a Measure of
Workplace Deviance,” 85 Journal of Applied Psychology 349-360.
Cialdini, Robert B., Raymond R. Reno, and Carl A. Kallgren. 1990. “A Focus Theory
of Normative Conduct: Recycling the Concept of Norms to Reduce Littering
in Public Places,” 58(6) Journal of Personality and Social Psychology 1015-1026.
_______, Carl A. Kallgren, and Raymond R. Reno. 1991. “A Focus Theory of Normative
Conduct: A Theoretical Refinement and Reevaluation of the Role of Norms in
Human Behavior,” 24 Advances in Experimental Social Psychology 201-234.
Cross, K. Patricia. 1977. “Not Can, but Will College Teaching be Improved?” 17 New
Directions for Higher Education 1-15.
Crump, S. Larry, Dennis L. Nunes, and E.K. Crossman. 1977. “The Effects of Litter on
Littering Behavior in a Forest Environment,” 9 Environment and Behavior 137-146.
Falk, Armin, and Urs Fischbacher. 2002. “ ‘Crime ’ in the Lab-Detecting Social
Interaction,” 46 European Economic Review 859-869.
Finnie, William C. 1973. “Field Experiments in Litter Control,” 5 Environment and
Behavior 123-144.
Fischbacher, Urs, Simon Gächter, and Ernst Fehr. 2001. “Are People Conditionally
Cooperative? Evidence from a Public Goods Experiment,” 71 Economics
Letters 397-404.

Review of Law & Economics, © 2012 by De Gruyter

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM
 Are Academics Messy? / 577

Funk, Patricia, and Peter Kugler. 2003. “Dynamic Interactions Between Crimes,” 79
Economics Letters 291-298.
Hansmann, Ralf, and Roland W. Scholz. 2003. “A Two-Step Informational Strategy for
Reducing Littering Behavior in a Cinema,” 35 Environment and Behavior 752-762.
Harcourt, Bernard E. 2001. Illusion of Order: The False Promise of Broken Windows Policing.
Cambridge, MA: Harvard University Press.
_______ and Jens Ludwig. 2006. “Broken Windows: New Evidence from New York City
and a Five-City Social Experiment,” 73 University of Chicago Law Review 271-320.
Keizer, Kees, Siegwart Lindenberg, and Linda Steg. 2008. “The Spreading of
Disorder,” 322(5908) Science 1681-1685.
Levitt, Steven D., and John A. List. 2009. “Field Experiments in Economics: The Past,
the Present and the Future,” 53 European Economic Review 1-18.
Sampson, Robert J., and Stephen W. Raudenbush. 1999. “Systematic Social
Observation of Public Spaces: A New Look at Disorder in Urban
Neighborhoods,” 105 American Journal of Sociology 603-651.
Sibley, Chris G., and Jiu H. Liu. 2003. “Differentiating Active and Passive Littering: A
Two-Stage Process Model of Littering Behavior in Public Spaces,” 35
Environment and Behavior 415-433.
Taylor, Ralph B. 2000. Breaking Away From Broken Windows: Evidence from Baltimore
Neighborhoods and the Nationwide Fight Against Crime, Grime, Fear and Decline.
Boulder, Co.: Westview Press.
Torgler, Benno, Bruno S. Frey, and Clevo Wilson. 2009. “Environmental and Pro-
Social Norms: Evidence on Littering,” 9(1) B.E. Journal of Economic Analysis &
Policy Article 18.
Van Lange, Paul A., Toon W. Taris, and Roos Vonk. 1997. “Dilemmas of Academic
Practice: Perceptions of Superiority Among Social Psychologists,” 27(6)
European Journal of Social Psychology 675-685.

Brought to you by | Queensland University of Technology

Authenticated | [Link]
Download Date | 3/19/13 12:11 AM