Face mask use and physical distancing before and after
mandatory masking: Evidence from public waiting lines
Gyula Seres∗Anna Balleyer†Nicola Cerutti‡Jana Friedrichsen§
July 12, 2020
During the COVID-19 pandemic, the introduction of mandatory face mask usage
was accompanied by a heated debate. It was argued that community use of masks cre-
ates a false sense of security that could decrease social distancing, thus making matters
worse. We conducted a randomized ﬁeld experiment in Berlin, Germany, to investigate
whether masks lead to decreases in distancing and whether this mask eﬀect interacts
with the introduction of a mask mandate in Berlin. Joining lines in front of stores,
we measured the distance kept from the experimenter in two treatment conditions –
the experimenter wore a mask in one and no face covering in the other – both before
and after the introduction of mandatory mask use in stores. We ﬁnd no evidence that
mandatory masking has a negative eﬀect on distance keeping. To the contrary, in our
study, masks signiﬁcantly increase distancing and the eﬀect does not diﬀer between the
two periods. Further, we ﬁnd no evidence that the mask mandate aﬀected distancing.
However, our results suggest that the relaxation of shop opening restrictions had a
negative eﬀect on distancing.
Keywords: COVID-19; Face Masks; Social Distancing; Risk Compensation; Field Ex-
periment; Health Policy
JEL Codes: I12, D9, C93
The novel coronavirus SARS-CoV-2 that quickly spread to almost all countries in the world
has – by the end of June 2020 – lead to more than ten million conﬁrmed infections and
more than 500,000 deaths (CSSE, 2020; Dong et al., 2020). To address the imminent health
emergency, to make the growth rate of the virus sub-exponential (colloquially, to ﬂatten the
curve), and to mitigate hospital overload, most countries implemented complete or partial
lockdown policies including stay-at-home orders, travel bans, social distancing, and also
emphasized personal precautions in terms of hand hygiene and respiratory etiquette. While
the conjunction of these policies has been proven eﬀective and death rates are believed
to have been substantially higher in their absence, it has also become clear that both the
uncontrolled pandemic and the successful lockdown measures have had severe consequences
for the economy and society (Fernando E. Alvarez, 2020; Thunstr¨om et al., 2020).
∗Corresponding author. Humboldt-Universit¨at zu Berlin. email@example.com
†University of Groningen
‡Berlin School of Economics and Law
§Humboldt-Universit¨at zu Berlin, WZB Berlin Social Science Center, and DIW Berlin
¶Humboldt-Universit¨at zu Berlin.
Given that SARS-CoV-2 outbreaks might remain a possibility for a long time, societies need
to develop alternatives to a strict lockdown that allow for a safe life with the virus even
though neither an eﬀective treatment nor a vaccine is available. Mandated face mask use is
a non-pharmaceutical intervention that is potentially very potent in combating COVID-19
(van der Sande et al., 2008; Rengasamy et al., 2010; Suess et al., 2012; Saunders-Hastings
et al., 2017; Eikenberry et al., 2020; Mitze et al., 2020). However, health authorities and
politicians have been cautious in advising universal mask mandates with reference to a
potential backlash from an induced false sense of security (WHO, 2020; Synhetsstyrelsen,
2020; Norwegian Institute of Public Health, 2020). Such compensating behavior is found in
the context of road safety regulation, oﬀsetting the expected positive eﬀects from regulation
(Peltzman, 1975). Subsequently, although risk compensation is studied in the context of
HIV prevention (Eaton and Kalichman, 2007; Marcus et al., 2013; Wilson et al., 2014),
bicycle helmets (Adams and Hillman, 2001), and seat-belt laws (Houston and Richardson,
2007; Evans and Graham, 1991; Cohen and Einav, 2003), among others, there are mixed
results on the existence of risk compensatory behavior.
As robust evidence on the existence and extent of risk compensation in response to face
mask use is missing, citing it as an argument against the mandatory use of masks relies on
two implicit assumptions. First, risk compensation will actually happen in the context of
the current epidemic and in response to mask use. Second, risk compensation only matters
if its eﬀect is larger than the presumably positive direct eﬀects of a greater prevalence of
masks in the community that would follow from a mask mandate. Seres et al. (2020) run a
ﬁeld experiment measuring the eﬀect of face masks on distancing in outside waiting lines in
Berlin, Germany in April 2020. Their study provides evidence against risk compensation in a
context where masks were not mandatory. Subjects were observed to even stand further away
from an experimenter who was masked than from an unmasked one. Using an additional
survey, they show that this behavior might be triggered by second-order beliefs, meaning
that people expect individuals who wear a mask to prefer others to stay further away from
them. Seres et al. (2020) yield insights about behavioral eﬀects of masks, but their results
cannot be easily extrapolated to a situation with a mask mandate. Therefore, we address
the question of how a mask mandate inﬂuences the eﬀect of masks on distancing behavior.
Using the same methodology as Seres et al. (2020), we run a ﬁeld experiment outside waiting
lines in Berlin before and after a mask mandate was put into place. We ﬁnd that individuals
stand further away from someone wearing a mask than from an unmasked person both before
and after the introduction of the mask mandate. Thus, the mandate did not crowd out the
positive eﬀect of the face mask observed under voluntary masking. While we observe more
people wearing masks themselves after the introduction of the mask mandate than before,
we also ﬁnd that average distances to other persons are shorter after the mandate than
before. Using contextual data in the form of the number of open shops in the surrounding,
we argue that this eﬀect is not driven by the mask mandate but by concurrent changes in
the perceived risk from the virus. This is in line with an array of studies showing that the
adoption of precautionary behavior against COVID-19 crucially depends on the perceived
risk of becoming severely ill from the virus (Ajzenman et al., 2020; Allcott et al., 2020;
Grossman et al., 2020; Harper et al., 2020; Larsen et al., 2020; Rosenfeld et al., 2020; Wise
et al., 2020).
Our results complement further evidence from Germany, Italy, and the US. Empirical studies
examine social distancing in terms of time spent outside and proximity during this period.
Kovacs et al. (2020) use location data from Germany to show that the introduction of face
mask mandates in Germany did not lead to a compensatory eﬀect in individuals’ mobility
patterns in terms of time spent outside. In contrast to these, Yan et al. (2020) argue that US
Americans spent more time outside their homes after masks became mandatory in public
spaces. Their empirical strategy does not preclude that mobility would have changed in
this way also in the absence of masks. Another key dimension is observed distancing in
community settings. A very similar result to that of Seres et al. (2020) was obtained in a
ﬁeld experiment in Italy: Marchiori (2020) shows that wearing a face mask can substantially
improve adherence with the physical distancing regulations on pavements both in the absence
of a mask mandate and after its introduction. Based on this body of evidence, a direct
behavioral backlash on distancing from making face masks mandatory appears unlikely. For
maximal support from the population, such mandates should be clearly communicated as
necessary and as an additional safety measure.1
The rest of this paper proceeds as follows: Section 2 provides the setup, including the local
progress of the epidemic, and provides a general introduction to the policy environment. Sec-
tion 3 describes the experimental design. Section 4 formally states the hypotheses. Section
5 provides the main results of this paper. Section 7 concludes with further interpretations
of the main results and a discussion.
A face mask mandate was introduced in all German states toward the end of April and
coincided with the relaxation of other regulations. In Berlin, starting in mid-March 2020,
only supermarkets and stores selling basic necessities were allowed to open. Then, from
April 22, 2020 onward, small retail stores (<800m2) were allowed to reopen under certain
restrictions (e.g., limited number of customers). While masks had been previously dismissed
as an attractive policy option, the expectation of the increased movement of citizens and
potential crowding in cities as well as increasing public pressure, led to the introduction
of mandatory masking policies in all federal states with only slight variations regarding
the starting dates. The ob jective of these policies was to reduce the risk of contagion in
places that became increasingly frequented but where physical distance recommendations
are harder to uphold, such as shops and public transport. However, individuals may adjust
their precautions in other dimensions in response to such a mandate so that the net eﬀect
is, a priori, not clear. Therefore, we follow up on Seres et al. (2020) with an identical ﬁeld
experiment, conducted in Berlin after the introduction of compulsory masking, evaluating
the eﬀect of masks and the interaction with other policy changes. The detailed timeline of
the experiment and the restrictions are stated in Table 1.
The ﬁeld experiment took place in 2020 during the COVID-19 pandemic in Berlin, Germany.2
The ﬁrst part of the data was collected before the face mask mandate and the second half
after the introduction of the mandate. During the ﬁrst data collection period, acceptable
reasons to leave the place of residence were deﬁned at the state level, limiting the mobility of
the experimenters. To comply with public health recommendations, the choice of stores was
made to avoid long commuting from the experimenters’ homes. Figure 1 shows the locations
of businesses visited. There was no overlap in the list of stores between experimenters,
therefore, only one of them visited each store in the sample. To better proﬁt from the natural
experiment setting created by the mask mandate, in May, the experimenters revisited the
same stores as those in April. The store types, where observations took place, were previously
restricted to supermarkets, drug stores (except pharmacies), and post oﬃces to observe a
sample representing the population visiting public areas. During the pandemic, lines in
front of businesses were frequent in Berlin, but irregular. Therefore, only the existing lines
at the moment of data collection could be utilized for our experiment. We address potential
1Settele and Shupe (2020) provide evidence from survey data that support for policies critically depends
on the information and perceptions that individuals hold.
2According to Robert Koch Institute, one of the central bodies for the safeguarding of public health
in Germany (https://www.rki.de/), the state of Berlin had the seventh highest number of SARS-CoV-2
infections per 100,000 population of the 16 German states as of May 1, 2020, when the incidence of COVID-19
cases in Berlin was 157 per 100,000 inhabitants; close to the federal average 197 per 100,000 inhabitants.
14.03.2020 •Beginning of Corona Related
22.03.2020 •Tightest Restrictions in Place
Start: Data Collection 1 •18.04.2020
20.04.2020 •Retail <800m2Reopen
End: Data Collection 1 •24.04.2020
27.04.2020 •Mask Mandate in Public
29.04.2020 •Mask Mandate in Shops
02-04.05.2020 •Big Retails Reopen, Gatherings of
up to 50 People Allowed
Start: Data Collection 2 •12.05.2020
15.05.2020 •Restaurants and Cafes Reopen with
End: Data Collection 2 •20.05.2020
Table 1: Berlin COVID-19 Restrictions and Experiment Timeline
randomization concerns regarding store selection in Section 5.
3.2 Experimental Design
Our experiment has a 2 ×2 between-subject design with respect to using a mask and time
period. To study the eﬀect of masks on distancing, we use a between-subject design with
randomized face covering. In the Mask treatment, the experimenter was wearing a mask,
whereas in the NoMask treatment, no face covering was used. To investigate whether the
treatment eﬀect interacts with a set of policy changes implemented at the end of April 2020,
we ran the same design twice, once before and once after the introduction of a mask mandate
Our experiment was carried out by experimenters who measured the distance between them-
selves and others in lines in front of businesses. Data was collected in two periods between
April 18-24 and May 12-20. In both periods, 60 observations each were recorded by four
experimenters, adding up to 240 in each period and 480 in total. The pre-registered exper-
imental protocol is in Appendix A.3
The experimenters are independent researchers, two women and two men, aged between
31 and 35, who participated voluntarily and are credited as co-authors of this paper. 4
3The pre-mandate data was used in Seres et al. (2020), which had ﬁve experimenters. In this paper, we
use data of four experimenters who participated in both measurement periods. This study was pre-registered
with ﬁve experimenters; however, one was unable to participate in the second period.
4Experimenters being co-authors of the study might raise questions regarding the conscious or unconscious
eﬀects on outcomes. However, as stated before, our pre-mandate data was used in Seres et al. (2020) and
Figure 1: Map indicating the observation sites
Each recorded the observations individually in their own neighborhood. Two measures
were taken to reduce as much as possible potential noise from diﬀerent appearances of the
experimenters. First, each member of the team used a white FFP2 respiratory protection
mask, which was the most easily accessible type of mask in pharmacies during the ﬁrst
period of data collection.5Second, the dress-code was standardized to a pair of blue jeans
and a dark colored top (Balafoutas and Nikiforakis, 2012).
Each experimenter independently located a line outside a shop in their neighborhood and
determined an even number of observations to be collected there. The experimenter wore
a mask (treatment Mask) or not (treatment NoMask) based on the result of a coin toss.
Then, the experimenter joined the line, maintaining a distance of 150 cm from the previous
person, measured with a mobile device. While waiting for the subject, meaning the next
person arriving and joining the line behind the experimenter, she/he assumed a sideways po-
sition in the line, thus ensuring her/his face would be visible to the next person but avoiding
eye contact. Upon arrival of the subject, the experimenter measured the distance between
her/his own feet and the subject’s, subsequently left the line, and input the measured dis-
tance and demographic data of the subject into a previously prepared table. Particular
cases (e.g. groups of people, strollers) were uniformly measured according to the protocol
(Appendix A). The distance was recorded via a mobile augmented reality application, which
provides 1-centimeter precise measurements. No visual or audio recordings were taken to
comply with privacy laws. A measurement took about 5-20 seconds to complete. Distance
was only recorded if the subject assumed a steady position for the time of measurement
and it was clear for them where to stand. For groups, the measured subject was the per-
son closest to the experimenter. We made note of no case when a subject recognized the
measurement or reacted to it by moving away. Having completed the input of data, the
experimenter returned to the end of the line.
this study has been preregistered stating no expectations regarding the outcome. The post-mandate data
mirrors the ﬁndings of the ﬁrst period. Thereby, we rule out potential concerns related to experimenter
inﬂuence on the outcome.
5An FFP2 mask is a mechanical ﬁlter respirator as deﬁned by the EN 149 standard, similar to the N95
design. FFP2 and surgical masks are not visibly diﬀerent and we do not expect that carrying out the
experiment with surgical masks would have altered the results.
At any store and period, an equal number of observations with and without a mask were
collected. At each visit, the experimenter used a coin toss to determine with which of the
two treatments to start.
The experimenters also collected information on the subjects’ demographic proﬁle. In par-
ticular, subject’s age group, gender, the number of accompanying children and adults were
recorded as well as whether the subject was wearing a mask at the time of measurement.
Note that during the second round of data collection, i.e. after the introduction of the mask
mandate, all subjects presumably had a mask with them as a prerequisite to enter the store,
unlike before the mandate. However, no law mandated using the mask while waiting outside.
Additional controls for the setting include the length of the queue, store type and exact
location. In order to control for the impact of the store closure policy, we recorded the
number of businesses within a 50-meter radius around the location that were open at the
time of measurement during the second round of data collection in May that were legally
closed during the ﬁrst round of data collection in April. This variable shows substantial
variance as it ranges from 0 to 6 in the May sample. As we argue above, changes in
distancing may also be inﬂuenced by the general perception about the epidemic. As a
measure of this factor, we gathered daily data from Google trends that shows the relatively
number of searches for the novel coronavirus in Berlin.6
The study consists of two main observational periods, the ﬁrst taking place before the exoge-
nous policy changes, including the introduction of mandatory mask wearing in stores, and
the second one afterwards. Each period has a balanced number of observations per treatment
group. The introduction of a mandate by the state creates a natural experiment setting and
lets us understand the impact of the policy by analyzing pre- and post-intervention periods.
As the mandate was brought into force at the same time as the aforementioned measures, it
is hard to isolate its pure eﬀect. However, we believe that the mask mandate and relaxation
measures have diﬀerent eﬀects on our dependent variable, kept distance. From this point
on, the exogenous diﬀerence between periods is referred to as policy change and the diﬀerent
policies are underlined separately when necessary.
In the pre-policy period, wearing a mask was voluntary; whereas during the post-policy,
all subjects had to carry a mask with them as they were expected to wear it in the store.
Therefore, we hypothesize that as masks became a common sight of the city in the second
observational period, a mask mandate would increase the general public awareness of the
health hazard. Based on the two-process theory of reasoning (Stanovich and West, 2000;
Kahneman, 2011), compliance with distancing requires mental eﬀort. In our context, both
the mandate and seeing a masked experimenter may serve as a reminder inducing a conscious
decision-making process, System 2.7Thus, the mandate would not have a signiﬁcant impact
on the subjects entering the line behind the masked experimenter, whereas it would increase
the distance kept by the subjects behind the experimenter without a mask by already trig-
gering their System 2 through the higher presence of masked people on streets. Keeping
the natural experiment setting and our expectations in mind, we formed and preregistered
Seres et al. (2020) use a survey to understand the mechanism behind the mask increasing
physical distancing and conclude that it might be the following: people tend to believe that
a person wearing a mask prefers others to keep a greater distance. This mechanism could
still be at play after the introduction of the mask mandate because our experiment takes
6The chosen keyword is “Coronavirus”, as it is most commonly called colloquially in the German-speaking
7System 1 is a cognitive process deﬁned in the psychology literature as automatic, largely unconscious,
fast, and undemanding of computational capacity. In contrast, System 2 is demanding and relatively slow.
8For narrative purposes, the order and wording of hypotheses is diﬀerent from that of the preregistered
place outdoors and masks are only obligatory in stores. Another explanation for a distance
increase in response to face masks could be a reminder eﬀect: people seeing others wearing
masks may be reminded of the health risk from COVID-19 and, thereby, of the appropriate
measures to take to prevent an infection. Seres et al. (2020) ﬁnd no evidence in this regard.
However, the introduction of the mask mandate required people joining lines to carry a
mask with them, led to an increase in general mask usage in the waiting line, and could
potentially create a reminder eﬀect through the larger presence of masks. Put diﬀerently,
masks might work as a trigger activating System 2 cognition, thus leading people to keep
a greater distance just because they adjust their judgment based on the severeness of the
situation. Taking both mechanisms into consideration, the introduction of a mask mandate
may not change the distancing behavior of the subjects behind the masked experimenter. If
a mask still works as a respect signal, then they continue to signal the same. On the other
hand, if masks work as a general reminder, then the eﬀect of this reminder would spread to
the entire population without aﬀecting the behavior of the subjects joining the line behind
a person with a mask.
Hypothesis 1.A. Distance kept toward the masked experimenter in treatment Mask in
the waiting line does not change with the policy.
Hypothesis 1.B. Distance kept toward the unmasked experimenter in treatment NoMask
in the waiting line is greater after the policy change.
Unlike people behind the masked experimenter, based on the two-process theory, the man-
date can be expected to have a positive eﬀect on the distancing behavior of people behind
a person without a mask. According to this explanation, the mandate itself as well as its
direct consequence of seeing masks more often on the streets as well as in the waiting line
could potentially increase the general public awareness regarding COVID-19-related risk and
mitigation measures. The heightened awareness may then induce subjects to increase their
precautions and, thus, their distancing.
Hypothesis 2. After the policy change, distance kept toward the experimenter in the waiting
line is the same in treatments Mask and NoMask, i.e., subjects keep the same distance
from the masked and unmasked experimenter.
As a consequence of hypotheses 1.A and 1.B, we expected the subjects in the post policy
sample to keep on average the same distance from the experimenter in both treatment condi-
tions. The convergence of distances in the two treatments is driven by the expectation that
the mandate might increase distancing for people behind a person without a mask but leave
unchanged the distances behind a person wearing a mask. A convergence in distances across
the two treatments could alternatively result if masks lose their informational value with the
introduction of the mandate but awareness is unchanged. If wearing a mask is no longer per-
ceived as signaling a preferred larger distance, people behind the masked experimenter might
keep a shorter distance after the mandate than before and on average the same as if standing
behind the unmasked experimenter. This explanation is not hypothesized separately as it
directly contrasts with Hypotheses 1.A and 1.B.
Hypothesis 3. After the policy change, subjects wearing a mask do not keep a greater
distance from the experimenter than unmasked subjects (treatment conditions Mask and
Using the pre-mandate sample, Seres et al. (2020) conclude that subjects wearing a mask
keep a larger distance in general. However, they also argue that it might be due to a selection
eﬀect. Therefore, the mask mandate is expected to increase the representation of those who
wear a mask in the line, creating a diﬀerent selection compared to the pre-policy period.
Hence, distancing behavior of this post-policy sub-sample might resemble that of the rest of
the post-policy sample, therefore, we expect no diﬀerence in distancing behavior.
5 Empirical analysis
5.1 Sample Characteristics
The data set contains 480 observations. For descriptive statistics on subject characteristics,
see Table 2. Compared to the city’s age groups, our sample underrepresents 60+ population
(10.6% vs. 24.7%). This is not surprising due to the asymmetric eﬀect of the disease on
elderly (Verity et al., 2020). However, considering that social distancing is crucial in public,
our study aims to measure the eﬀect on people who leave their homes. Our sample is
representative in terms of gender according to a Chi-Square Goodness of Fit test (54.4%
vs. 50.8% in the population, χ2= 2.45, p = 0.117). Most subjects in our sample arrive
at the store alone, only 10.6% come with adult, and 6.1% with minor, companions. There
is a clear increase in mask use after the policy change as it soars from 17.1% pre-mandate
to 40.1% post-mandate. We also recorded the length of the line as the number of people
standing outside in front of the experimenter. The mean length is 5.63 individuals with a
standard deviation of 3.83.
Count NoMask Mask NoMask Mask P
Subject Without Mask 102 97 77 65 341
Subject With Mask 18 23 43 55 139
Accompanying Adult =0 107 105 108 109 429
Accompanying Adult =1 11 13 12 10 46
Accompanying Adult >1 2 2 0 1 5
Accompanying Child =0 111 112 113 116 451
Accompanying Child =1 7 7 7 4 25
Accompanying Child >1 2 2 0 0 4
Female Subject 61 65 65 70 261
Male Subject 59 55 55 50 219
Aged under 15 0 1 1 0 2
Aged between 15 and 25 13 19 14 15 61
Aged between 25 and 35 38 34 42 40 154
Aged between 35 and 45 35 29 33 33 130
Aged between 45 and 60 20 20 21 21 82
Aged above 60 14 17 9 11 51
Total 120 120 120 120 480
Table 2: Number of Subjects in diﬀerent treatment conditions.
Notes: Values show the number of observations with the given characteristics for categorical variables.
Age groups and gender reﬂect the experimenters’ impressions and are not to be interpreted as point
estimates. Subjects are counted with a mask if they were wearing one at the time of measurement.
5.2 Estimation Strategy
Our analysis seeks to understand whether and how the introduction of a mask mandate
changes the physical distancing behavior that face masks create. In doing so, we exploit
a natural experiment setting formed after the mask use in stores in Berlin was mandatory
starting from April 27th, 2020. Randomization of mask use by the experimenters allows us to
use a diﬀerence-in-diﬀerences approach, comparing pre- and post-policy periods for subjects
behind a masked or an unmasked experimenter. Due to the randomization of treatments,
we can assume parallel trends between treatment groups, thus arguing in favor of causal
evidence regarding permanence of masks’ behavioral eﬀect. On the other hand, we are
also well aware that the introduction of a mask mandate came along with other relaxation
measures. Thus, we interpret the pre- and post-policy diﬀerences as the joint eﬀect of the
mandate and relaxations, naming it accordingly in our model.
Pooling the entire sample, we estimate the following equation to identify the eﬀect of the
policy on distancing and its interaction with our Mask treatment:
Distancei=β0+β1M askEi+β2P olicyi+β3M askEi×P ol icyi
in which Distanceiis the distance kept by subject i,MaskEiis the indicator of the exper-
imenter wearing a mask, Policyiis an indicator for data collected after the policy changes
took place, thus distinguishing the two periods of data collection, and MaskS iis an in-
dicator for the subject wearing a mask. Hence, β2captures any eﬀect in distancing that
results from the conjunction of policy changes between the ﬁrst and the second data collec-
tion period but does not relate to the treatment, whereas the eﬀect of the mask mandate
jointly with other policy changes on the eﬀect from mask wearing is identiﬁed by β3. If we
cannot reject β3= 0, this implies that the face mask eﬀect on distancing is not signiﬁcantly
diﬀerent between two periods. Xiis a vector of all other covariates and controls used in
diﬀerent speciﬁcations. Standard errors εiare clustered according to store and date in order
to mitigate any potential correlation in error terms.9As the experimenter locations are not
overlapping in our sample, this approach also covers experimenter related correlations.
5.3 Main results
We structure the discussion of results along the hypotheses laid out above. We observe that
subjects keep a shorter distance from the experimenter in the data collected after the policy
change than before it on average. This also holds true in each treatment condition separately.
While subjects kept an average distance of 151.14 cm (SD=29.62) to the unmasked exper-
imenter in our pre-mandate sample, the average distance to the unmasked experimenter is
only 143.35 cm (SD=31.79) in the post-mandate sample. Similarly, subjects kept an average
distance of 159.85 cm (SD=31.79) to the masked experimenter in the pre-mandate sample,
but only 151.41 cm (SD=34.08) in the post-mandate sample. Thus, distances kept are, on
average, 7.79 cm (NoMask) and 8.41 cm (Mask) shorter in the post-policy period than
in the data collected before the policy change. Regression analysis conﬁrms this observa-
tion: from Table 3, we see that the coeﬃcient of the policy change is negative if we regress
observed distances on the policy change and controls for the subsample of subjects facing
the masked (column 1) and unmasked (column 2) experimenter. Therefore, we reject both
hypotheses 1.A and 1.B that subjects do not change or increase their distancing toward the
masked respectively unmasked experimenter due to the policy change.
Result 1. Distance kept toward the experimenter in the waiting line is weakly shorter after
the policy change in treatments Mask and NoMask.
In order to better understand whether the observed change in distancing between the pre-
and post-policy samples are driven by the mask mandate or by the relaxations in the restric-
tions in place, we estimate diﬀerent speciﬁcations of equation 1. The results and summarized
in columns 3 to 6 of Table 3.
In line with the observed diﬀerence in distancing, column (3) indicates that distances toward
both the unmasked and the masked experimenter are about 9 cm shorter after the policy
changes had taken eﬀect. While the coeﬃcient on the treatment dummy MaskE is positive
and signiﬁcant, the coeﬃcient of the interaction between MaskE and the policy change is
not signiﬁcant, suggesting that subjects keep signiﬁcantly larger distances to the masked
experimenter both before and after the policy changes took eﬀect.
We conclude that any diﬀerence from the policy changes must have aﬀected subjects facing
the masked and the unmasked experimenter equally. Using additional speciﬁcations, we
9The clustered standard errors are used to mitigate any potential serial correlation in the error terms
due to clustered sampling. As we are considering relatively small number of clusters (55 in total), we also
perform wild cluster bootstrap method as a robustness check following Cameron et al. (2008). Please see
Appendix for details.
Table 3: Distance OLS
(1) (2) (3) (4) (5) (6)
Sample Mask NoMask Pooled Pooled Pooled Pooled
(4.216) (4.266) (4.221) (4.273)
Policy -12.26∗∗ -8.213 -9.294∗-0.0833 -3.055 8.082
(4.380) (4.546) (4.610) (6.407) (9.891) (10.88)
MaskE×Policy -1.614 -2.250 -1.481 -2.098
(5.557) (5.458) (5.610) (5.502)
MaskS 13.57∗1.545 7.376∗7.623∗7.499∗7.785∗
(5.838) (4.052) (3.251) (3.298) (3.302) (3.356)
Stores -3.173∗∗ -3.255∗∗
Online Search 0.170 0.216
Pop. Density -2.282∗∗∗ -0.610 -1.423∗∗∗ -1.066∗∗ -1.435∗∗∗ -1.071∗∗
(0.411) (0.343) (0.309) (0.330) (0.300) (0.315)
Acc. Adult -5.087 -4.387 -5.442 -5.788 -5.403 -5.748
(6.952) (8.273) (4.999) (4.944) (4.992) (4.933)
Acc. Child -0.382 -5.921 -4.444 -3.789 -4.482 -3.821
(6.339) (4.175) (3.126) (3.254) (3.163) (3.296)
Ppl in Line 0.625 0.904 0.865∗∗ 0.938∗∗ 0.860∗∗ 0.934∗∗
(0.731) (0.484) (0.304) (0.309) (0.296) (0.303)
Constant 180.7∗∗∗ 158.7∗∗∗ 165.2∗∗∗ 156.5∗∗∗ 151.5∗∗∗ 138.9∗∗∗
(7.179) (5.412) (5.932) (6.701) (20.15) (18.32)
Demographics Yes Yes Yes Yes Yes Yes
Observations 240 240 480 480 480 480
R20.146 0.091 0.107 0.125 0.109 0.127
Notes: Ordinary least squares estimates. Dependent variable is distance kept from the experimenter.
Standard errors in parentheses are clustered by day and store. ∗p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001.
MaskE and MaskS are indicator variables for whether the experimenter or subject, respectively, used a
face mask. Acc. Adult and Acc. Child indicate whether the subject was accompanied by at least one
other adult or child, respectively. Density is population density based on the 2011 German Census data.
Controls include gender and age dummy variables. Standard errors are clustered in day and store level.
argue that the observed shift in behaviors can be explained by a combination of factors
including a change of general perception about the pandemic and relaxation of business
openings. Speciﬁcally, we include the variable Stores, measuring the number of businesses
that were legally closed in April but were open in May at the time of measurement within a
50-meter radius of the point of data collection in May, and the variable Online Search, repre-
senting the relative number of Berlin speciﬁc hits on Google search for the novel coronavirus
on the day of measurement.
Speciﬁcations (4)-(6) in Table 3 include the additional covariates individually and in combi-
nation. These results suggest that the decline in distancing after the policy change can be
explained by reopening stores rather than the introduction of the mask mandate.10. Accord-
ing to speciﬁcation (6), an additional newly open store near the location of measurement is
related to a decreased in distancing of 3.255 cm on average (p=0.002). However, we ﬁnd
no evidence that online search for pandemic-related content predicts the diﬀerence between
distances in the pre- and post-mandate samples well. Virus-related internet traﬃc positively
correlates with greater distancing but the eﬀect is not signiﬁcant (p=0.257).
Next, we investigate the eﬀect from the mask intervention in the post-mandate sample
(hypothesis 2). From table 3, it can be seen that the marginal eﬀect of MaskE is at least 9
10We do not include experimenter ﬁxed-eﬀects here because they are collinear with the variable Stores
and the model would be overidentiﬁed including both.
cm in all speciﬁcations and signiﬁcant. The eﬀect is robust across time: the interaction with
the policy change is negative, as predicted, but not statistically diﬀerent from zero. Thus,
we ﬁnd no evidence suggesting that the eﬀect has vanished with the introduction of a mask
mandate and thereby reject hypothesis 2. Further, in the post-mandate sample, subjects
kept greater distances from the masked than from the unmasked experimenters, on average.
Result 2. In the post-mandate sample, subjects maintain a signiﬁcantly larger distance from
the experimenter wearing a mask.
We now turn to hypothesis 3 that, in the post-mandate period, the sub-sample of subjects
wearing a mask themselves does not react diﬀerently to the masked experimenter than
the rest of the sample. We ﬁrst note that the share of masked subjects is indeed much
higher post-mandate than pre-mandate, even though masks were never mandatory in outside
waiting lines (see table 2).11 Indeed, this increase suggests that more people regularly wear
masks, such that the subsample of mask wearers post-mandate is less selected and more
similar to the population of unmasked individuals. We ﬁnd that subjects with mask keep a
signiﬁcantly larger distance than unmasked subjects in the pre-mandate sample (two-sample
t=-2.3788, p=0.0091), but this diﬀerence vanishes in the post-mandate sample (two-sample
t=-0.6327, p=0.2638). Thus, our data is consistent with hypothesis 3.
Result 3. In the post-policy sample, subjects wearing a mask do not keep a larger distance
from the experimenter.
Even though the self-selection into mask-wearing does not allow for a causal interpretation
of the estimate coeﬃcient on MaskS, we further note that the coeﬃcient is signiﬁcantly
positive in speciﬁcations (3) to (6), contradicting the hypothesis that subjects may engage
in risk compensation and reduce their distancing in response to the protection oﬀered by a
Table 3 reveals further interesting patterns. Population density of the neighborhood de-
creases distancing, 1000 inhabitants/km2decreases distancing by more than 1cm.12 Sub-
jects arriving in a group keep a shorter distance, but the diﬀerence is not signiﬁcant in any
speciﬁcation, neither for adult nor for minor companions.13 The number of people in line
in front of the experimenter has a small but signiﬁcant eﬀect on distancing (in (6): -0.93,
p=0.003). To learn if wearing a mask makes subjects not to stand behind the experimenter,
we test if the sample correlation coeﬃcient between this and the treatment variable is sig-
niﬁcant. The reasoning is that this behavior would increase the time between observations,
resulting in shorter lines. This claim is rejected (r= 0, p = 0.117).
5.4 Further Results
German health authorities and oﬃcial mandates to limit the spread of the coronavirus spec-
ify that individuals should keep a distance of at least 150 cm to each other. In addition to
our main analysis, we investigate how our treatment of masking the experimenter and the
introduction of the mask mandate in Berlin aﬀect compliance with this required minimum
distance. We ﬁnd that compliance is higher toward the masked experimenter in both ob-
servation periods. Before the mask mandate, compliance is 54.17% if the experimenter does
not wear a mask and 69.17% if she/he does. After the mandate is introduced, compliance
is 40% if the experimenter does not wear a mask and 49.17% if he/she does.
The 150cm rule may look arbitrary as the recommendations of safe distances vary substan-
tially between countries.14 Hence, we also consider if compliance with alternative threshold
values increases with masking the experimenter. Figure 2a demonstrates that the choice
11The use of masks is signiﬁcantly higher post mandate, γ2=0.9036 p=0.014.
12Area is deﬁned by postal code.
13Further demographic controls gender and age dummies are not signiﬁcant.
14For example, as of June 2020, the U.S. Center for Disease Control and Prevention (CDC) recommends
a 6-feet distance (=182.88cm).
0 .2 .4 .6 .8 1
50 100 150 200 250
CDF NoMask CDF Mask
(a) Cumulative distribution function.
0 .005 .01 .015
50 100 150 200 250
(b) Kernel density estimates
Figure 2: Cumulative distribution functions of distances kept by the subject from the ex-
perimenter in NoMask (blue) and Mask (red) conditions (in centimeter). Cumulative
distributions are exact and densities are estimated univariate Epanechnikov kernel density
of the critical value does not change our conclusion that mask improve distancing from
the experimenter. It is evident from the ﬁgure that subjects in the Mask condition are
more likely to exceed any relevant threshold value, i.e. compliance is generally higher there
than in the NoMask condition. Using non-parametrically estimated kernel density func-
tions, we conﬁrm a positive shift in distancing (Fig. 2b, D=0.175, P=0.01, two-sided
Equation (1) is correctly speciﬁed only if the treatment dummy MaskE does not inﬂuence
the subject’s decision of putting on a mask, MaskS. We believe that the exogeneity of
MaskS is given as subjects decide about their use of a mask before seeing the experimenter.
However, this decision may be reversed upon seeing the experimenter. We therefore test the
independence claim is tested with the following logit binary choice model:
P r(M askS = 1) = exp(γ0+M askE +γ2M andate+γ3M askE×M andate×M andate+φXi+εi)
1+exp(γ0+MaskE+γ2M andate+γ3M askE ×M andate+φXi+εi)(2)
Using the same set of covariates as in Model (4) in Table 3, we ﬁnd that the coeﬃcient of
MaskE is not signiﬁcant (γ1=0.2358, p=0.524). We conclude that the subjects decide about
wearing a mask independently of whether the experimenter wears a mask or not.
5.5 External Validity
According to the medical literature, airborne contagion is a primary source of transmission
of SARS-CoV-2 and, thus, distancing between individuals is important to prevent the spread
of the virus (Zhang et al., 2020). Hence, a successful mitigation strategy needs to understand
and take into account how policy aﬀects distancing patterns. In this study, we analyze how
the introduction of a mask mandate aﬀected distancing in order to contribute knowledge
in this respect. An overall evaluation of policies on distancing needs to take into account
as many facets of individual behavior as possible because restrictions as well as re-openings
alter the choice set of customers, resulting in changes in behavior that aﬀect the exposure
to infectious particles.
Putting our design in perspective, the data was collected in an environment where trans-
mission is possible (Qian et al., 2020)16 but at the same time, wearing a mask is optional
15A parametric test yields similar results. We estimated a logit model analogous to model (6) in Table 3
where now the dependent variable is compliance with the 150cm threshold. The estimated coeﬃcient of Mask
Experimenter is positive and signiﬁcant (β1=0.7145, p=0.02). The interaction between Mask Experimenter
and the policy change is insigniﬁcant as in the main model (β3=-0.2893, p=0.535). See Appendix for details.
16Airborne lifetime of small speech droplets can reach 8-14 minutes according to Stadnytskyi et al. (2020),
but air movements can dilute the concentration of virus and make transmission substantially less likely.
even under the mandate. The places where we collected data fall into a category of settings
in which distancing is recommended by authorities with the pretext that it helps preventing
contagion. Our ﬁndings do not necessarily generalize to the eﬀect of the mask mandate in
stores or high-risk areas where the mask mandate made wearing a mask mandatory. How-
ever, our results are fully in line with evidence from mobility patterns in Germany, which
have not changed negatively with the introduction of a mask mandate (Kovacs et al., 2020).
We seek to extend the literature on masks by investigating how face mask policies aﬀect
distancing and, speciﬁcally, how they interact with the eﬀect that face masks have on dis-
tancing behavior. We follow up on the claim that face masks make individuals prone to less
rigorous compliance with other contagion prevention recommendations, such as physical dis-
tancing, a claim that is frequently heard in the discussion on face masks but has so far not
received empirical support. Seres et al. (2020) show that individuals keep a greater distance
from someone who is masked, contradicting a negative eﬀect of masking. However, ex ante,
it is not clear that this would still be true in the presence of a mask mandate. With data
from periods before and after the introduction of a mask mandate in Berlin, we show that
the positive eﬀect of masking observed in Seres et al. (2020) persisted in the presence of a
mask mandate. In this section, we evaluate our ﬁndings in the light of motivation crowding,
risk compensation, two-process theory, and cognitive dissonance.
Based on motivation crowding theory, the introduction of a face mask policy may ultimately
alter the behavioral response of individuals beyond the wearing of masks, e.g. distancing
behavior. We argue that a large part of the precautions that individuals engage in to ﬂatten
the curve are intrinsically motivated. This is in line with evidence on the eﬀect of perceived
risk on precautions taken and also with the observation that mobility in many place was
substantially reduced even before oﬃcial stay-at-home orders became eﬀective. According
to motivation crowding theory, a mask mandate may then crowd in or crowd out intrinsic
motivation to comply with measures to prevent the virus spread, depending on circumstances
(Frey and Jegen, 2001). A crowding-in in motivation and, thus, an eﬀect that reinforces the
eﬀectiveness of the mask mandate with respect to virus spread can be expected if individuals
perceive the policy as supporting their intrinsic motivation. However, if individuals perceive
it as negating their intrinsic eﬀorts or as making those redundant, the policy may induce
countervailing behavior change by crowding out intrinsic motivation (Frey and Jegen, 2001;
Festr´e and Garrouste, 2015). Further, policies may crowd out compliance with existing
norms because they are perceived as rules that substitute social norms (Ostrom, 2000).
Our data does not support the notion of crowding out from the introduction of the mask
mandate as we attribute the observed decrease in distancing in the post-mandate sample to
the accompanying policy relaxations.
Further, individuals may perceive the introduction of a face-masking mandate as an indi-
cation that alternative precautions – e.g., avoiding unnecessary contacts and trips, keeping
safe physical distances to others – had become less relevant. Speciﬁcally, individuals may
perceive face masks as an eﬀective means of reducing the overall infection risk as evidence
for this becomes available. If individuals show risk compensation behavior and decrease
their compliance with complementary measures such as distance-keeping, the expected ben-
eﬁcial eﬀect from compulsory masking would be (partially) negated. Ironically, introducing
a face mask policy (rather than just recommending face mask use) may introduce a general
increase in risk compensation behavior, which was the reason why face masks were not rec-
ommended as a protective measure initially by WHO and other health bodies. Our results
speak against such a direct backlash from a mask mandate but also suggest that distancing
is sensitive to contextual changes such as increased shop openings.
On a diﬀerent note, we stated our hypotheses on policy eﬀects based on the premise that
both the masked experimenter and the mask mandate, which implies that subjects in the
waiting line carry a mask with them, serve as triggers: According to the two-process theory
of reasoning (Stanovich and West, 2000; Kahneman, 2011), these triggers would induce
the necessary mental eﬀort to comply with the recommended distancing. What we ﬁnd is
only partially consistent with this. The positive eﬀect of the mask-wearing experimenter
on distancing is present before and after the mask mandate, suggesting that the mask in
treatment Mask has a trigger eﬀect. But we do not ﬁnd evidence that the mask mandate
serves as a trigger as distancing if anything decreases after the mandate.
According to the theory of cognitive dissonance (Festinger, 1957), the interaction of people
with the outside world highly depends on mental inconsistencies. In our case, the relaxations
that came with the policy change at the end of April 2020 have plausibly suggested to
people that the severity of the situation had decreased. Thus, avoiding a mask became less
costly for human psychology after the policy change and the mask mandate may not have
created enough of a general awareness to counteract this change in beliefs. When faced
with the masked experimenter directly in front of them, though, subjects in our experiment
apparently adjust their perception and better comply with the recommended distancing,
leading to the observed positive mask eﬀect.
This study utilizes a ﬁeld experiment in which we measure distancing in lines to stores.
The experimenters entered these lines randomly with and without a face masks during the
COVID-19 pandemic. Measurement was carried out twice: before and after the mask man-
date in stores. The main ﬁndings show that the eﬀect of masks worn by the experimenters
is positive signiﬁcant and not signiﬁcantly diﬀerent in the two time periods. We do not
ﬁnd evidence of risk compensation: Subjects wearing a mask do not keep a shorter distance
and the mask mandate did not change this. Using pre-mandate ﬁeld data and survey, Seres
et al. (2020) conclude that mask-driven distancing is implied by a signaling channel. A
mask mandate does not impact this conclusion. Assuming that distancing is an eﬀective
measure against transmission of SARS-CoV2, we ﬁnd no evidence that mandating masks
has a negative spillover eﬀect.
Adams, J. and M. Hillman (2001). The risk compensation theory and bicycle helmets. Injury
Prevention 7 (2), 89–91.
Ajzenman, N., T. Cavalcanti, and D. Da Mata (2020). More than words: Leaders’ speech
and risky behavior during a pandemic. Available at SSRN 3582908 .
Allcott, H., L. Boxell, J. Conway, M. Gentzkow, M. Thaler, and D. Y. Yang (2020). Polar-
ization and public health: Partisan diﬀerences in social distancing during the coronavirus
pandemic. NBER Working Paper 26946.
Balafoutas, L. and N. Nikiforakis (2012). Norm enforcement in the city: A natural ﬁeld
experiment. European Economic Review 56 (8), 1773–1785.
Cameron, A. C., J. B. Gelbach, and D. L. Miller (2008). Bootstrap-based improvements for
inference with clustered errors. Review of Economics and Statistics 90 (3), 414–427.
Cohen, A. and L. Einav (2003). The eﬀects of mandatory seat belt laws on driving behavior
and traﬃc fatalities. Review of Economics and Statistics 85 (4), 828–843.
CSSE (2020). COVID-19 Dashboard by the Center for Systems Science and Engineering
(CSSE) at Johns Hopkins University (JHU).
Dong, E., H. Du, and L. Gardner (2020). An interactive web-based dashboard to track
COVID-19 in real time. The Lancet Infectious Diseases 20 (5), 533–534.
Eaton, L. A. and S. C. Kalichman (2007). Risk compensation in HIV prevention: im-
plications for vaccines, microbicides, and other biomedical HIV prevention technologies.
Current HIV/Aids Reports 4 (4), 165–172.
Eikenberry, S. E., M. Mancuso, E. Iboi, T. Phan, K. Eikenberry, Y. Kuang, E. Kostelich,
and A. B. Gumel (2020). To mask or not to mask: Modeling the potential for face mask
use by the general public to curtail the covid-19 pandemic. Infectious Disease Modelling.
Evans, W. N. and J. D. Graham (1991). Risk reduction or risk compensation? The case of
mandatory safety-belt use laws. Journal of Risk and Uncertainty 4 (1), 61–73.
Fernando E. Alvarez, David Argente, F. L. (2020). A simple planning problem for COVID-19
lockdown. NBER Working Paper 26981.
Festinger, L. (1957). A theory of cognitive dissonance, Volume 2. Stanford University Press.
Festr´e, A. and P. Garrouste (2015). Theory and evidence in psychology and economics about
motivation crowding out: A possible convergence? Journal of Economic Surveys 29 (2),
Frey, B. S. and R. Jegen (2001). Motivation crowding theory. Journal of Economic Sur-
veys 15 (5), 589–611.
Grossman, G., S. Kim, J. Rexer, and H. Thirumurthy (2020). Political partisanship inﬂu-
ences behavioral responses to governors’ recommendations for COVID-19 prevention in
the united states. Available at SSRN 3578695 .
Harper, C. A., L. P. Satchell, D. Fido, and R. D. Latzman (2020). Functional fear predicts
public health compliance in the covid-19 pandemic. International Journal of Mental Health
Houston, D. J. and L. E. Richardson (2007). Risk compensation or risk reduction? Seatbelts,
state laws, and traﬃc fatalities. Social Science Quarterly 88 (4), 913–936.
Kahneman, D. (2011). Thinking, fast and slow. Macmillan.
Kovacs, R., M. Dunaiski, and J. Tukiainen (2020). Compulsory face mask policies do not
aﬀect community mobility in germany. OSF, https://doi.org/10.31219/osf.io/m3sv8.
Larsen, M. V., M. B. Petersen, and J. Nyrup (2020). Do survey estimates of the public’s
compliance with COVID-19 regulations suﬀer from social desirability bias? PsyArXiv,
Marchiori, M. (2020). Covid-19 and the social distancing paradox: dangers and solutions.
arXiv preprint arXiv:2005.12446 .
Marcus, J. L., D. V. Glidden, K. H. Mayer, A. Y. Liu, S. P. Buchbinder, K. R. Amico,
V. McMahan, E. G. Kallas, O. Montoya-Herrera, J. Pilotto, et al. (2013). No evidence
of sexual risk compensation in the iPrEx trial of daily oral HIV preexposure prophylaxis.
PloS One 8 (12).
Mitze, T., R. Kosfeld, J. Rode, and K. W¨alde (2020). Face masks considerably reduce
COVID-19 cases in Germany: a synthetic control method approach. IZA Discussion
Paper No. 13319.
Norwegian Institute of Public Health (2020). Hand hygiene, cough etiquette,
face masks, cleaning and laundry - advice and information to the gen-
eral public. https://www.fhi.no/en/op/novel-coronavirus-facts-advice/facts-and-general-
Ostrom, E. (2000). Collective action and the evolution of social norms. Journal of Economic
Perspectives 14 (3), 137–158.
Peltzman, S. (1975). The eﬀects of automobile safety regulation. Journal of Political Econ-
omy 83 (4), 677–725.
Qian, H., T. Miao, L. Li, X. Zheng, D. Luo, and Y. Li (2020). Indoor transmission of
SARS-CoV-2. medRxiv, https://doi.org/10.1101/2020.04.04.20053058.
Rengasamy, S., B. Eimer, and R. E. Shaﬀer (2010). Simple respiratory protec-
tion—evaluation of the ﬁltration performance of cloth masks and common fabric materials
against 20–1000 nm size particles. Annals of Occupational Hygiene 54 (7), 789–798.
Roodman, D., M. Ørregaard Nielsen, J. G. MacKinnon, and M. D. Webb (2019). Fast and
wild: Bootstrap inference in stata using boottest. The Stata Journal 19 (1), 4–60.
Rosenfeld, D. L., H. Rothgerber, and T. Wilson (2020). Politicizing the covid-
19 pandemic: ideological diﬀerences in adherence to social distancing. PsyArXiv,
Saunders-Hastings, P., J. A. Crispo, L. Sikora, and D. Krewski (2017). Eﬀectiveness of
personal protective measures in reducing pandemic inﬂuenza transmission: A systematic
review and meta-analysis. Epidemics 20, 1–20.
Seres, G., A. Balleyer, N. Cerutti, A. Danilov, J. Friedrichsen, Y. Liu, and M. S¨uer (2020).
Face masks increase compliance with physical distancing recommendations during the
covid-19 pandemic. Retrieved from osf.io/db8sj.
Settele, S. and C. Shupe (2020). Lives or livelihoods? Perceived tradeoﬀs and public demand
for non-pharmaceutical interventions. CEBI Working Paper 19/20.
Stadnytskyi, V., C. E. Bax, A. Bax, and P. Anﬁnrud (2020). The airborne lifetime of small
speech droplets and their potential importance in SARS-CoV-2 transmission. Proceedings
of the National Academy of Sciences 117 (22), 11875–11877.
Stanovich, K. E. and R. F. West (2000). Individual diﬀerences in reasoning: Implications
for the rationality debate? Behavioral and Brain Sciences 23 (5), 645–665.
Suess, T., C. Remschmidt, S. B. Schink, B. Schweiger, A. Nitsche, K. Schroeder,
J. Doellinger, J. Milde, W. Haas, I. Koehler, et al. (2012). The role of facemasks and
hand hygiene in the prevention of inﬂuenza transmission in households: results from a
cluster randomised trial; berlin, germany, 2009-2011. BMC infectious diseases 12 (1), 26.
Synhetsstyrelsen (2020). FAQ. 6. Should I be wearing a mouth or face mask if I am healthy?
Thunstr¨om, L., S. C. Newbold, D. Finnoﬀ, M. Ashworth, and J. F. Shogren (2020). The
beneﬁts and costs of using social distancing to ﬂatten the curve for covid-19. Journal of
Beneﬁt-Cost Analysis, 1–27.
van der Sande, M., P. Teunis, and R. Sabel (2008). Professional and home-made face masks
reduce exposure to respiratory infections among the general population. PLoS One 3 (7).
Verity, R., L. C. Okell, I. Dorigatti, P. Winskill, C. Whittaker, N. Imai, G. Cuomo-
Dannenburg, H. Thompson, P. G. Walker, H. Fu, et al. (2020). Estimates of the severity
of coronavirus disease 2019: a model-based analysis. The Lancet Infectious Diseases.
WHO (2020). Advice on the use of masks in the context of COVID-19. WHO reference
number: WHO/2019-nCoV/IPC Masks/2020.3.
Wilson, N. L., W. Xiong, and C. L. Mattson (2014). Is sex like driving? HIV prevention
and risk compensation. Journal of Development Economics 106, 78–91.
Wise, T., T. D. Zbozinek, G. Michelini, C. C. Hagan, et al. (2020). Changes in risk perception
and protective behavior during the ﬁrst week of the COVID-19 pandemic in the United
States. PsyArXiv, https://doi.org/10.31234/osf.io/dz428.
Yan, Y., J. Bayham, E. P. Fenichel, and A. Richter (2020). Do face
masks create a false sense of security? A COVID-19 dilemma. medRxiv,
Zhang, R., Y. Li, A. L. Zhang, Y. Wang, and M. J. Molina (2020). Identifying airborne
transmission as the dominant route for the spread of covid-19. Proceedings of the National
Academy of Sciences.
There are two appendices. The ﬁrst appendix contains the pre-registered protocol for the
ﬁeld experiment. (Appendix A). The second consists of robustness tests (Appendix B).
A Experimental Protocol
17 Disclaimer: Experimenters signed up to this experiment on voluntary basis under the
condition that they do not belong to any risk groups. In order to prevent posing any further
risk on any of the parties, the Robert Koch Institute’s health recommendations are strictly
The instructions for the recording of data follow. Please read the whole document and follow
all points very carefully.
Code of Conduct
As experimenter, you will need an FFP2 respiratory protection mask for this experiment.
Each time, before you go to an experiment location, you will take two full-body (self-
)portrait photos of yourself: One with and one without a mask. The primary purpose of the
photos is recording variables describing your appearance if this is requested by the reviewers.
To decrease the noise due to experimenter appearance, you are expected to wear a pair of
blue jeans and a dark colored (black, dark gray, or navy blue) top without any visible text
or logo.19 Your outﬁt and mask type have to match that you used for the preceding data
Location You may choose a location that satisﬁes the following list of conditions.
•The establishment is an open supermarket, a drug store (except pharmacy), or a post
•There must be a queue outside with people waiting for entering the store. The queue
must stand on a ﬂat surface with no obstructing objects. Make sure that the queue
is clearly visible and it is clear for the arriving subject that you are the last person in
the line and approximately where they should stand.
•You can record the data anytime between May 12-20 between 08:00-20:00 during day-
light with good visibility. In order to secure good visibility conditions, do not record
data when it is raining.
•You should avoid stores that have heavy traﬃc that would make measurement diﬃcult.
For instance, if there is another store or a subway exit next door, people in the queue
might change their position frequently, making recording data problematic.
•The time gap between people who are let in the store must be suﬃciently long. The
measurement may take a couple of seconds, and you may be asked to move forward if
the queue moves; the subject can also move before you can record the distance between
you. The speed is usually slower at post oﬃces than at supermarkets.
•The location you choose should be limited to those you visited during the previous
17There are minor diﬀerences in the two protocols. These changes are clearly marked in the text.
18The Robert Koch Institute (RKI) is the German government’s key scientiﬁc institution in the ﬁeld
of biomedicine. It is one of the central bodies for the safeguarding of public health in Germany. See
19Please consult us if you do not own these items.
Data Recording Method You will need a smartphone with an installed augmented-reality
tape-measure app that is capable of measuring small distances in centimeters with small
measurement errors. The error is measured individually on the same device you use on
location. Place two ﬂat objects on the ground at any location with a clear surface exactly
100 cm from each other. Similarly to the protocol on location, measure this distance with
the application. Do the same measurement ﬁve times with diﬀerent positions of the objects.
You may proceed with this hardware and application if the error is within a 3% margin
Preparation for Data Recording In total, you are expected to perform 60 independent
observations. Before each session, you set an even target of observations you are planning
to record. Half of them you execute with your mask on, the other half without. The order
you decide randomly using a fair coin or any random number generator. Example: You set
the number to 20. After tossing the coin, you start with 10 observations with your mask on.
After ﬁnishing with this, you remove the mask and perform another 10 without it. Finally,
you leave the location.
The purpose of changing your appearance only once is to limit the number of times you
may accidentally touch your face. You can safely avoid this if you remove the mask by only
touching the strings. You should proceed the same way if you start your work without your
mask on. To learn about the safe way of wearing a mask, please consult the website of the
Robert Koch Institute.
Data Recording Procedure Due to lock-down measures in place, you will work alone and
record the data individually. After choosing the location, go to the end of the queue outside
and carefully follow this protocol.
1. Go to the queue and stand 150 centimeters (1.5 meter) away from the last person.20
Measure this using the same application.
2. Turn sideways, neither facing the queue nor the subject arriving after you. Make sure
that you can see both.
3. If necessary, calibrate your application such that it is ready for measurement. Do not
open other applications at this point.
4. If someone is approaching, turn your back against the queue and face the subject
before they arrive. Make sure that your face is visible, but look at your device the
whole time. Keep a neutral facial expression and do not make eye contact.
5. The app measures distance by pinning two points on the ground. These two points
are the closest points of yours and the subject’s shoes. You pin the tip of their shoe
ﬁrst when they arrive, and the tip of your shoe second.
6. Record the length and exit the queue.
7. After this, record all remaining variables, starting with the number of people in the
queue who were standing before you outside at the point of measurement. After this,
go back to the end of the queue until you reach your target number of observations.
Further Points to Consider
If there is a group, the subject is the person closest to you, irrespective of age. Exceptions:
If the closest person is an infant in a stroller or a person in a wheelchair, the closest point is
where the front wheel touches the ground. If this reference point belongs to a stroller, the
person you record is the one handling the stroller.
20Recommended minimum safe distance by the Federal Government of Germany and the Robert Koch
Do not record an observation if you are unable to pinpoint the position of the subject
accurately (i.e. the subject can keep jogging in place, move back or forward before you
can ﬁnish pinning) or if the subject engages in an activity that would trigger distancing
according to local social norms (i.e. smoking, talking on the phone, eating).
There are 3 time slots per day: morning 8:00-12:00, mid-day 12:00-16:00, and early evening
16:00-20:00. Do not record more than 50% of the observations in one period of time (e.g.
morning), even if they are recorded on diﬀerent days.
Do not attempt to make any media record of the subject or any other individual near you
as this may be unwelcome without consent. If you meet hostile or unfriendly reactions or
you are questioned by someone, you can reveal your identity and that you are conducting a
publicly funded scientiﬁc study. If this hinders or inﬂuences recording data, or puts you in
an uncomfortable situation, leave the location.
You are asked to identify if there is a shop/establishment nearby that is open at the time
point of measurement and accepting customers, but was legally not allowed to open in April
because of the business type (e.g. nail salon, certain types of retail store). To qualify, it has
to be visible and within a 50-meter radius from the point of data collection.
Data and Variables
In this part, you can ﬁnd the list of variables with the corresponding codes. Your task is to
complete the spreadsheet for each observation. You will receive the spreadsheet by email.
If you ﬁnished recording, send the ﬁle to firstname.lastname@example.org.
MaskE Treatment variable. Experimenter 0=without 1=with mask.
Distance Distance to the subject. Measured in centimeter (cm).
GenderS Binary variable. Subject gender 0=male 1=female.
AgeS Guessed age category of the subject. 0= below 14, 1=14-25, 2=25-
35, 3=35-45, 4=45-60, 5=60+. If it is uncertain, write your best
MaskS Binary variable. Subject 0=without, 1=with a manufactured
mask, 2=with homemade mask or improvised cover of mouth and
nose (e.g. scarf ).
CompanyAdult Number of accompanying adults, 0=no adult. Adult, if age>14.
CompanyChild Number of accompanying children, 0=no child. Child, if age<14.
TotalNumofPeople The total number of people outside in front of you in the queue
at the moment of measurement. Do not include people inside.
SocialNormS The presence of social norm violations (i.e. smoking, food, other).
Address Address of the experiment. For example, “Spandauer Strasse 1,
Store Type of the store. 1=post oﬃce, 2=supermarket, 3=drug store,
4=other (please add a note)
Local At least one business open nearby (50m) that was not allowed to
open in April but is open to customers at the time of measurement.
ID Surname of experimenter.
Date Date of the month. E.g. if the date is April 20, write 20.
Time Time of the day (i.e. 1400, 1430, etc.).
Note 1 Additional remarks, may be left empty.
Note 2 Additional remarks, may be left empty.
B Further robustness checks
B.1 Collection method
The observations were collected in sessions. Each session is deﬁned as the target number
of observations that the experimenter aims to obtain when initially approaching the line.
The experimenter wore a mask (treatment group) for half of the observations and collected
the other half without wearing the mask (control group). The order of the treatment and
control group was randomized through a coin toss. To ensure the conditions in which the
measurements were taken did not diﬀer, we calculated the time distance between measure-
ments in the same session.21 The average time between observations was 320 seconds, with a
standard deviation of 335 seconds and no signiﬁcant diﬀerence between the treatment group
and the control group (Mann-Whitney U test z = -0.926, p = 0.3547). We believe, therefore,
that no subject refrained from joining the line because of the experimenter wearing – or not
wearing – a mask. On average, 5.63 people (SD=3.83) were present in the line, excluding
the experimenter. Pre-mandate, the average line comprised 6.48 (SD=4.11) subjects, while
post-mandate the average length dropped to 4.78 (SD=3.33). We did not detect a signiﬁcant
diﬀerence between the length of lines between the treatment group and the control group
(Mann-Whitney U test z = 0.188, p = 0.8511). The age of subjects wearing a mask pre- and
post-intervention, as highlighted in ﬁgure 3a, is substantially diﬀerent. The older portion
of the sample was much more likely to wear masks (38.71%) even before their use in shops
was made compulsory. The percentage of mask wearers in the other age categories, instead,
rose in the post-intervention period, reaching an average of 40.45% subjects aged 0 to 60
from the previous average of 13.88% in the pre-intervention period.
0 .1 .2 .3 .4 .5
0−15 15−2525−35 35−45 45−60 60+ 0−15 15−25 25−35 35−45 45−60 60+
Subjects with masks by age category (%)
(a) Mask usage by age.
50 100 150 200 250 300
0−15 15−25 25−35 35−45 45−60 60+ 0−15 15−25 25−35 35−45 45−60 60+
Distance by age category (cm)
(b) Distance held by the subjects, by age category.
Figure 3: Distance and mask wearing by age categories.
During our measurements, we were not able to accurately record the type of mask used
by the subjects. FFP2 masks, while diﬀerent in substance from a surgical mask due to
their ﬁltering properties, are optically diﬃcult to distinguish from other types of masks. We
expect the subjects looking at the experimenters to have encountered the same diﬃculty.
B.2 Wild cluster bootstrap
Cluster-robust standard errors may be inaccurate when calculated on small numbers of
clusters (see, e.g., Cameron et al., 2008). While our data is divided in 55 clusters and, thus,
is less prone to this type of inaccuracy, here we report the p-values calculated through a wild
21One of the experimenters did not record the exact time of the observations, therefore the corresponding
data was removed from this analysis. We include only measurements taken at most 60 minutes apart from
the previous, as time distances longer than 30 minutes come from breaks taken by the experimenters or
absence of a line at the time of measure. 7 observations were recorded at more than 60 minutes from the
previous, 5 in the control group and 2 in the treatment group.
Table 4: Wild bootstrap p-values
Distance Bootstrap p-value
Mask Experimenter 9.392∗[0.048]
Mandate 8.082 [0.505]
Mask Experimenter ×-2.098 [0.715]
Newly Open Stores -3.255∗∗ [0.022]
Online Search 0.216 [0.339]
Mask Subject 7.785∗∗ [0.028]
Population Density -1.071∗∗ [0.004]
Accompanying Adult -5.748 [0.284]
Accompanying Child -3.821 [0.244]
# of People in Line 0.934∗∗ [0.009]
Female Subject -0.717 [0.832]
Aged under 15 3.835 [0.739]
Aged between 15 and 25 -8.159 [0.289]
Aged between 25 and 35 -0.986 0.842
Aged between 35 and 45 -0.670 [0.891]
Aged between 45 and 60 3.626 [0.483]
Constant 138.9∗∗∗ [0.000]
Notes: Ordinary least squares estimates. Clustered errors (day, store) in parentheses. ∗p < 0.05, ∗∗
p < 0.01, ∗∗∗ p < 0.001. Mask Experimenter and Mask Subject are indicator variables for whether
the experimenter or subject, respectively, used a face mask. Female Sub ject=1 if the subject is female.
Accompanying Adult and Accompanying Child indicate whether the subject was accompanied by at
least one other adult or child, respectively. Population density is based on the 2011 German Census
data. Wild cluster bootstrap p-values in square brackets.