DISCUSSION PAPER SERIES

IZA DP No. 13319

Face Masks Considerably Reduce

COVID-19 Cases in Germany:
A Synthetic Control Method Approach

Timo Mitze
Reinhold Kosfeld
Johannes Rode
Klaus Wälde

JUNE 2020

Electronic copy available at: https://ssrn.com/abstract=3620634

DISCUSSION PAPER SERIES

IZA DP No. 13319

Face Masks Considerably Reduce

COVID-19 Cases in Germany:
A Synthetic Control Method Approach

Timo Mitze Klaus Wälde

University of Southern Denmark, RWI and Johannes Gutenberg University Mainz,
RCEA CESifo and IZA
Reinhold Kosfeld
University of Kassel
Johannes Rode
TU Darmstadt

JUNE 2020

Any opinions expressed in this paper are those of the author(s) and not those of IZA. Research published in this series may
include views on policy, but IZA takes no institutional policy positions. The IZA research network is committed to the IZA
Guiding Principles of Research Integrity.
The IZA Institute of Labor Economics is an independent economic research institute that conducts research in labor economics
and offers evidence-based policy advice on labor market issues. Supported by the Deutsche Post Foundation, IZA runs the
world’s largest network of economists, whose research aims to provide answers to the global labor market challenges of our
time. Our key objective is to build bridges between academic research, policymakers and society.
IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper
should account for its provisional character. A revised version may be available directly from the author.

ISSN: 2365-9793

IZA – Institute of Labor Economics

Schaumburg-Lippe-Straße 5–9 Phone: +49-228-3894-0
53113 Bonn, Germany Email: [email protected] www.iza.org

Electronic copy available at: https://ssrn.com/abstract=3620634

IZA DP No. 13319 JUNE 2020

ABSTRACT
Face Masks Considerably Reduce
COVID-19 Cases in Germany:
A Synthetic Control Method Approach1
We use the synthetic control method to analyze the effect of face masks on the spread of
Covid-19 in Germany. Our identification approach exploits regional variation in the point
in time when face masks became compulsory. Depending on the region we analyse, we
find that face masks reduced the cumulative number of registered Covid-19 cases between
2.3% and 13% over a period of 10 days after they became compulsory. Assessing the
credibility of the various estimates, we conclude that face masks reduce the daily growth
rate of reported infections by around 40%.

JEL Classification: I18, C23

Keywords: COVID-19, public health measures, face masks,
synthetic control method, Germany

Corresponding author:
Klaus Wälde
Gutenberg School of Management and Economics
Johannes Gutenberg Universität Mainz
Jakob-Welder-Weg 4
D-55131 Mainz
Germany
E-mail: [email protected]

1
Klaus Wälde has been acting as an IZA Visiting Research Fellow since March 2020.

Electronic copy available at: https://ssrn.com/abstract=3620634

1 Introduction
Many countries have experimented with several public health measures to mitigate the spread
of Covid-19. One particular measure that has been introduced are face masks. It is of obvious
interest to understand the contribution made by such a measure to reducing infections.
The effect of face masks on the spread of infections has been studied for a long time. The
usefulness in the clinical context is beyond dispute. There is also considerable evidence that
they helped in mitigating the spread of epidemics such as SARS 2003 or influenza (see below).
The effect of face masks worn in public on the spread of Covid-19 has not been systematically
analyzed so far. This is the objective of this paper.
There is a general perception in Germany that public wearing of face masks reduces incidences
considerably. This perception comes mainly from the city of Jena. After face masks were
introduced on 6 April 2020, the number of new infections fell almost to zero. Jena is not the
only city or region in Germany, however, that introduced face masks. Face masks became
compulsory in all federal states between 20 April and 29 April 2020. Six regions made masks
compulsory before the introduction at the federal level. These dates lay between 6 April and
25 April (see appendix A and Kleyer et al., 2020, for a detailed overview of regulations in
Germany). This leads to a lag between individual regions and the corresponding federal states
of between two and 18 days.
We derive findings by employing synthetic control methods (SCM, Abadie and Gardeazabal,
2003, Abadie et al., 2010, Abadie, 2019). Our identification approach exploits the previously
mentioned regional variation in the point in time when face masks became compulsory in
public transport and sales shops. We use data for 401 German regions to estimate the effect of
this public health measure on the development of registered infections with Covid-19. We
consider the timing of mandatory face masks as an exogenous event to the local population.
Masks were imposed by local authorities and were not the outcome of some process in which
the population was involved. 2 We compare the Covid-19 development in various regions to
their synthetic counterparts. The latter are constructed as a weighted average of control
regions that are similar to the regions of interest. Structural dimensions taken into account
include prior Covid-19 cases, their demographic structure and the local health care system.
We indeed find strong and convincing statistical support for the general perception that public
wearing of face masks in Jena strongly reduced the number of incidences. We obtain a synthetic
control group that closely follows the Covid-19 trend before introduction of mandatory masks
in Jena and the difference between Jena and this group is very large after 6 April. Our findings
indicate that the early introduction of face masks in Jena has resulted in a reduction of almost
25% in the cumulative number of reported Covid-19 cases after 20 days. The drop is greatest,
larger than 50%, for the age group 60 years and above. Our results are robust when we conduct
sensitivity checks and apply several placebo tests, e.g. tests for pseudo-treatment effects in
similarly sized cities in the federal state of Thuringia and for pseudo-treatment effects in Jena
before the treatment actually started. We also test for announcement effects.
Constructing control groups for other single regions is not always as straightforward as for Jena.
As a consequence, it is harder to identify the effect of face masks in these regions. When we

2
This is similar to the setup in Abadie et al. (2010), who study the effect of an increase in the tobacco tax in
California. The tobacco tax was decided upon by the state government.

Electronic copy available at: https://ssrn.com/abstract=3620634

move from single to multiple treatment effects, we find smaller effects. They are still
sufficiently large, however, to support our point that wearing face masks is a very cost-efficient
measure for fighting Covid-19. When we summarize all of our findings in one single measure
(we compare all measures in appendix B.4), we conclude that the daily growth rate of Covid-19
cases in the synthetic control group falls by around 40% due to mandatory mask-wearing
relative to the control group. 3
Concerning the literature (see appendix D for a more detailed overview), the effects of face
masks have been surveyed by Howard et al. (2020) and Greenhalgh et al. (2020). Greenhalgh
et al. (2020) mainly presents evidence on the effect of face masks during non-Covid epidemics
(influenza and SARS). Marasinghe (2020) reports that they “did not find any studies that
investigated the effectiveness of face mask use in limiting the spread of COVID-19 among those
who are not medically diagnosed with COVID-19 to support current public health
recommendations”.
In addition to medical aspects (like transmission characteristics of Covid-19 and filtering
capabilities of masks), Howard et al. (2020) survey evidence on mask efficiency and on the
effect of a population. They first stress that “no randomized control trials on the use of masks
<…> has been published”. The study which is “the most relevant paper” for Howard et al. (2020)
is one that analyzed “exhaled breath and coughs of children and adults with acute respiratory
illness” (Leung et al., 2020, p. 676), i.e. used a clinical setting. Concerning the effect of masks
on community transmissions, the survey needs to rely on pre-Covid-19 studies. We conclude
from this literature review that our paper is the first analysis that provides field evidence on
the effect of masks on mitigating the spread of Covid-19.

2 Identification, data and implementation

Identification. Our identification approach exploits the regional variation in the point in time
when face masks became mandatory in public transport and sales shops. Given the federal
structure of Germany, decisions are made by municipal districts (regions in what follows) and
federal states. We can exploit differences by, first, identifying six regions (equivalent to the EU
nomenclature of territorial units for statistics, NUTS, level 3) which made wearing face masks
compulsory before their respective federal states. For all other regions, mandatory mask-
wearing followed the decision of the corresponding federal state. Second, as Figure 1 shows,
variation across federal states also implies variations across regions.
To identify possible treatment effects from introducing face masks, we apply SCM for single
and multiple treated units. Our methodical choice is motivated as follows: First, the original
goal of SCM to “estimate the effects of <…> interventions that are implemented at an aggregate
level affecting a small number of large units (such as cities, regions, or countries)” (Abadie,
2019, p.3) clearly matches with our empirical setup. Compared to standard regression analyses,
SCM is particularly well suited for comparative case study analyses with only one treated unit
or a very small number thereof (Abadie and Gardeazabal, 2003, Becker et al., 2018). Second,
the method is flexible, transparent and has become a widely utilized tool in the policy
evaluation literature (Athey and Imbens, 2017) and for causal analyses in related disciplines

3
The main channel through which masks reduce transmission of SARS-CoV-2 is the reduction in aerosols and
droplets, as argued by Prather et al. (2020).

Electronic copy available at: https://ssrn.com/abstract=3620634

(see, e.g., Kreif et al., 2015, for an overview of SCM in health economics, Pieters et al., 2017,
for a biomedical application). 4

April 27
Saarland, Baden Württemberg, Rheinland-Palatine, Bavaria, Lower
April 22 Saxony, Brandenburg, Bremen, Hamburg, Hessia, Mecklenburg-
Saxony-Anhalt Western Pomerania, Northrhine-Westphalia, Berlin ( public transport)

April 20 April 24
Saxony Thuringia April 29
Schleswig-Holstein, Berlin (shopping malls)

April 6 April 13 April April 27 May 4

April 6 April 14 April 17 20.04. April 25

Jena Nordhausen Rottweil Main-Kinzig-Kreis, Braunschweig
Wolfsburg

Figure 1: The timing of mandatory mask wearing in federal states (top) and individual regions (below)

SCM identifies synthetic control groups for the treated unit(s) to build a counterfactual. In our
case, we need to find a group of regions that have followed the same Covid-19 trend as treated
units before mandatory masks in the latter. This control group would then most likely have had
the same behavior as treated unit(s) in the absence of the mask obligation. We can then use
this group to ‘synthesize’ the treated unit and conduct causal inference. The synthetic control
group is thereby constructed as an estimated weighted average of all regions in which masks
did not become compulsory earlier on. Historical realizations of the outcome variable and
several other predictor variables that are relevant in determining outcome levels allow us to
generate the associated weights, which result from minimizing a pre-treatment prediction
error function (see Abadie and Gardeazabal, 2003, Abadie et al., 2010 and Abadie, 2019 for
methodical details).
Data. We use the official German statistics on reported Covid-19 cases from the Robert Koch
Institute (RKI, 2020). The RKI collects the data from local health authorities and provides
updates on a daily basis. Using these data (available via API), we build a balanced panel for 401
NUTS Level 3 regions and 95 days spanning the period from January 28 to May 1, 2020 (38,095
observations). We use the cumulative number of registered Covid-19 cases in each district as
main outcome variable. 5 We estimate overall effects for this variable together with
disaggregated effects by age groups (persons aged 15-34 years, 35-59 years and 60+ years). As
an alternative outcome variable, we also use the cumulative incidence rate. Table 1 shows
summary statistics for both variables for our sample period.
Table 1 also presents our other predictor variables. We focus on factors that are likely to
describe the regional number and dynamics of reported Covid-19 cases. Obviously, past values

4
Friedson et al. (2020) employ the SCM to estimate the effect of the shelter-in-place order for California in the
development of Covid-19. The authors find inter alia that around 1600 deaths from Covid-19 were avoided by this
measure during the first four weeks.
5
We are aware of the existence of hidden infections. As it appears plausible to assume that they are proportional
to observed infections across regions, we do not believe that they affect our results. We chose the date of
reporting (as opposed to date of infections) because not all reported infections include information about the date
of infection.

Electronic copy available at: https://ssrn.com/abstract=3620634

of (newly) registered Covid-19 cases are important to predict the regional evolution of Covid-
19 cases over time in an autoregressive manner. In addition, we argue that a region’s
demographic structure, such as the overall population density and age structure, and its basic
health care system, such as the regional endowment with physicians and pharmacies per
population, are important factors for characterizing the local context of Covid-19. Predictor
variables are obtained from the INKAR online database of the Federal Institute for Research on
Building, Urban Affairs and Spatial Development. We use the latest year available in the
database (2017). We consider it likely that regional demographic structures only gradually vary
over time such that they can be used to measure regional differences during the spread of
Covid-19 in early 2020.

Table 1: Summary Statistics of Covid-19 indicators (outcome variables) and predictors characterizing the
regional demographic structure and basic health care system
Mean S.D. Min. Max.
PANEL A: Data on registered Covid-19 cases
[1] Newly registered cases per day 4.13 10.66 0 310
[2] Cumulative number of cases 120.86 289.07 0 5795
[3] Cum. cases [2] per 100,000 inhabitants 59.87 106.80 0 1,530.32
PANEL B: Regional demographic structure and local health care system
Population density (inhabitants/km2) 534.79 702.40 36.13 4,686.17
Population share of highly educated* individuals (in %) 13.07 6.20 5.59 42.93
Share of females in population (in %) 50.59 0.64 48.39 52.74
Average age of females in population (in years) 45.86 2.11 40.70 52.12
Average age of males in population (in years) 43.17 1.83 38.80 48.20
Old-age dependency ratio (persons aged 65 years and 34.34 5.46 22.40 53.98
above per 100 of population age 15-64)
Young-age dependency ratio (persons aged 14 years 20.54 1.44 15.08 24.68
and below per 100 of population age 15-64)
Physicians per 10,000 of population 14.58 4.41 7.33 30.48
Pharmacies per 100,000 of population 27.01 4.90 18.15 51.68
Settlement type (categorial variable$) 2.59 1.04 1 4
Notes: * = International Standard Classification of Education (ISCED) Level 6 and above; $ = categories are based
on population shares and comprise 1) district-free cities (kreisfreie Großstädte), 2) urban districts (städtische
Kreise), 3) rural districts (ländliche Kreise mit Verdichtungsansätzen), 4) sparsely populated rural districts (dünn
besiedelte ländliche Kreise).

Implementation. The implementation of the SCM is organized as follows: As baseline analysis,

we focus on the single treatment case for the city of Jena for three reasons. First, as shown in
Figure 1, Jena was the first region to introduce face masks in public transport and sales shops
on April 6. This results in a lead time of 18 days relative to mandatory face masks in the
surrounding federal state Thuringia on April 24. By April 29, all German regions had introduced
face masks (exact dates are provided in appendix A). A sufficiently long lag between mandatory
face masks in the treated unit vis-à-vis the sample of control regions is important for effect
identification.

Electronic copy available at: https://ssrn.com/abstract=3620634

Second, the timing of the introduction of face masks in Jena is -by and large- not affected by
other overlapping public health measures related to the Covid-19 spread. Since March 22 the
German economy had been in a general “lock down” coordinated among all federal states. Only
from April 20 onwards has the economy been gradually reopening. Third, Jena is in various ways
a representative case for studying the Covid-19 development: On April 5, which is one day
before face masks became compulsory in Jena, the cumulative number of registered Covid-19
cases in Jena was 144. This is very close to the median of 155 for Germany. Similarly, the
cumulative number of Covid-19 incidences per 100,000 inhabitants was 126.9 in Jena compared
to a mean of 119.3 in Germany (compare Figure A1).
In our baseline configuration of the SCM, we construct the synthetic Jena by including the
number of cumulative Covid-19 cases (measured one and seven days before the start of the
treatment) and the number of newly registered Covid-19 cases (in the last seven days prior to
the start of the treatment) as autoregressive predictor variables. The chosen period shall
ensure that the highly non-linear short-run dynamics of regional Covid-19 cases are properly
captured. We use cross-validation tests to check the sensitivity of the SCM results when we
allow for a shorter training period in the pre-treatment phase by imposing longer lags. The
autoregressive predictors are complemented by the cross-sectional data on the region’s
demographic and basic health care structure.
Although the case study of Jena can be framed in a clear identification strategy, the Covid-19
spread in a single municipality may still be driven by certain particularities and random events
that may prevent a generalization of estimated effects. We therefore also test for treatment
effect in districts that introduced face masks after Jena but still before they became compulsory
in the corresponding federal state. More importantly, however, we apply a multiple treatment
approach that takes all regions as treated units which introduced face masks by April 22. This
results in 32 regions from Saxony and Saxony-Anhalt. All other regions apart from Thuringia
introduced face masks on April 27. We employ this delay to study the effects of mandatory
masks up to May 1st. We end on May 1st as we would expect that differences across treated
and non-treated regions should disappear 5-7 days after April 27. This delay results from a
median incubation time of 5.2 days (Linton et al., 2020 and Lauer et al., 2020) and around 2
days accounting for reporting to authorities (as assumed e.g. in Donsimoni et al., 2020a, b).
Although SCM appears to be a natural choice for our empirical identification strategy, we are
well aware of the fact that its validity crucially depends on important practical requirements
including the availability of a proper comparison group, the absence of early anticipation effects
or interference from other events (Cavallo et al., 2013, Abadie, 2019). In the implementation
of the single and multiple treatment SCM we check for these pitfalls through sensitivity and
placebo tests. We deal with these issues in our baseline case study for Jena as follows:
1. We have screened the introduction and easing of public health measures, as documented
in Kleyer et al. (2020), to ensure that no interference takes place during our period of study.
This is the case at least until April 20 when exit strategies from public health measures
started.
2. We make sure that the regions used to create the synthetic control, i.e. the donor pool, are
not affected by the treatment (Campos et al., 2015). We eliminate the two immediate
geographical neighbors of Jena from the donor pool to rule out spillover effects. We also
exclude those regions for which anticipation effects may have taken place because face
masks became compulsory in quick succession to Jena.

Electronic copy available at: https://ssrn.com/abstract=3620634

3. We account for early anticipation effects in Jena. Specifically, we take the announcement
that face masks will become compulsory one week before their introduction as an
alternative start of the treatment period.
4. We apply cross-validation tests to check for sensitivities related to changes in historical
values in the outcome variables used as predictors. We also run placebo-in-time tests to
check whether effects actually occurred even before the start of the treatment.
5. We test for the sensitivity of the results when changing the donor pool and run
comprehensive placebo-in-space tests as a mode of inference in the SCM framework.

Inference thereby relies on permutation tests and follows the procedures suggested by Cavallo
et al. (2013) and applied, for example, by Eliason and Lutz (2018) or Hu et al. (2018). For both
the single and multiple treatment applications we estimate placebo-treatment effects for each
district in which masks did not become compulsory early on. These placebo treatments should
be small, relative to the treated regions. We calculate significance levels for the test of the
hypothesis that the mask obligation did not significantly affect reported Covid-19 cases. This
provides us with p−values for each day, which capture the estimated treatment effect on
reported Covid-19 cases from placebo regions. The p-values are derived from a ranking of the
actual treatment effect within the distribution of placebo treatment effects. We follow the
suggestion in Galiani and Quistorff (2017) and compute adjusted p-values taking the pre-
treatment match quality of the placebo treatments into account. 6

3 The effects of face masks on Covid-19

Baseline results for Jena. Panel A in Figure 2 shows the SCM results for the introduction of face
masks in Jena on April 6. The visual inspection of the development of cumulative Covid-19 cases
shows that the fit of the synthetic control group is very similar to Jena before the treatment. 7
The difference in the cumulated registered Covid-19 cases between Jena and its corresponding
synthetic control unit after the start of the treatment can be interpreted as the treatment effect
on the treated.
The figure clearly shows a gradually widening gap in the cumulative number of Covid-19 cases
between Jena and the synthetic control unit. The size of the effect 20 days after the start of the
treatment (April 26) amounts to a decrease in the number of cumulative Covid-19 cases of 23%.
For the first 10 days, the decrease amounts to 13%. Expressed differently, the daily growth rate
of the number of infections decreases by 1.32 percentage points per day (see appendix B.4 for
computational details and an overview of all measures). If we look at the estimated differences
by age groups, Table A2 in the appendix indicates that the largest effects are due to the age
group of persons aged 60 years and above. Here the reduction in the number of registered
cases is even larger than 50%. For the other two age groups we find a decrease between 10
and 20%.

6
We conduct all estimations in STATA using “Synth” and “Synth Runner” packages (Abadie at al., 2020, Galiani and
Quistorff, 2017). Data and estimation files can be obtained from the authors upon request.
7
The pre-treatment root mean square prediction error (RMSPE) of 3.145 is significantly below a benchmark RMSPE
of 6.669, which has been calculated as the average RMSPE for all 401 regions in the pre-treatment period until
April 6. This points to the relatively good fit of the synthetic control group for Jena in this period.

Electronic copy available at: https://ssrn.com/abstract=3620634

If we consider a median incubation of 5.2 days plus a potential testing and reporting lag of 2-3
days, the occurrence of a gradually widening gap between Jena and its synthetic control three
to four days after the mandatory face masks seems fast. One might conjecture that an
announcement effect played a role. As shown in appendix B.7, online searches for (purchasing)
face masks peaked on April 22, when it was announced that face masks would become
compulsory in all German federal states. 8 A smaller peak (70% of the April 22 peak) of online
searches appeared on March 31. This is one day after Jena announced that masks would
become compulsory on April 6. The announcement was accompanied by a campaign “Jena zeigt
Maske” to communicate the necessity to wear face masks in public 9 and was widely discussed
all over Germany.
Panel B in Figure 2 therefore plots the results when we set the start of the treatment period to
the day of the announcement on 30 March. The visual inspection of the figure shows the
existence of a small anticipation effect (which is mainly driven by the relative development of
Covid-19 age group 15-34 years (Panel B in Figure A2). Yet, the gap to the synthetic control
significantly widens only approximately 10 days after the announcement. As this temporal
transmission channel appears plausible against the background of incubation times and given
that no other intervention took place around this time in Jena or the regions in the synthetic
control group, we take this as first evidence for a face mask-effect in the reduction of Covid-19
infections. Appendix B.6 shows similar SCM results for the incidence rate (overall and by age
groups). We find a reduction of approximately 30 cases per 100,000 of population.

Panel A: Introduction of face masks on April 6 Panel B: Announcement/Start of campaign on March 30

200

200
Cumulative number Covid-19 cases
150

150
100

100
50

50
0

March 27 April 6 April 16 April 26 53 March 30 April 6 April 19

Jena synthetic control unit Jena synthetic control unit

Figure 2: Treatment effects of mandatory face masks in Jena on April 6 and start of campaign on March
30 (see Table A3 and appendix B.2 for details)

8
See https://www.tagesschau.de/inland/corona-maskenpflicht-103.html. Last accessed May 05, 2020.
9
See https://www.jenaer-nachrichten.de/stadtleben/13069-jena-zeigt-maske-kampagne-f%C3%BCr-mund-
schutz-startet. Last accessed May 05, 2020.

Electronic copy available at: https://ssrn.com/abstract=3620634

Obviously, the estimated differences in the development of Jena vis-à-vis the synthetic Jena is
only consistently estimated if our SCM approach delivers robust results. Accordingly, we have
applied several tests to check for the sensitivity of our findings.
Cross-validation and placebo-in-time test. One important factor is that our results are not
sensitive to changes in predictor variables. We therefore perform cross-validation checks by
modifying the length of the training and validation periods before the start of the treatment.
Panel A in Figure 3 shows that lagging the autoregressive predictor variables further in time
only slightly changes our results. Importantly, we do not find a systematic downward bias of
our baseline specification (cumulative number of reported Covid-19 cases: one and seven days
before start of treatment; number of newly registered Covid-19 cases: last seven days before
start of treatment) compared to an alternative specification. The latter trains the synthetic
control on the basis of information on cumulative Covid-19 cases 7 and 14 days prior to the
treatment together with the development of newly register cases between day 7 and 14 prior
to the treatment. Given that regional Covid-19 cases developed very dynamically and non-
linearly in this period, this is an important finding in terms of the robustness of our results.

Panel A: Cross-validation for changes in predictors Panel B: Placebo-in-time test (20 days in advance)
250

200
Cumulative number Covid-19 cases

Cumulative number Covid-19 cases

150
200

100
150

50
100

March 30 April 6 April 26 March 16 March 26 April 5

Jena Jena
synthetic control unit (baseline) synthetic control unit
synthetic (lag: +1 day)
synthetic (lag: +3 days)
synthetic (lag: +7 days)

Figure 3: Cross-validation for changes in predictor variables and placebo-in-time test

Notes: In Panel A the baseline specification for the synthetic control group uses historical values of the outcome
variable in the following way: i) number of cumulative Covid-19 cases (measured one and seven days before the
start of the treatment), ii) the number of newly registered Covid-19 cases (in the last seven days prior to the start
of the treatment); the alternative specifications lag these values by 1, 3 and 7 days.

Another important factor for the validity of the results is that we do not observe an anticipation
effect for Jena prior to the announcement day. We test for a pseudo-treatment in Jena over a
period of 20 days before the introduction of face masks. This period is equally split into a pre-

Electronic copy available at: https://ssrn.com/abstract=3620634

and pseudo post-treatment period. As Panel B in Figure 3 shows, there is no strong deviation
from the path of the synthetic control group. This result needs to be interpreted with some
care as the regional variation of Covid-19 cases in Germany is very heterogeneous the longer
we go back in time. This is indicated by the generally lower fit of the synthetic control group in
matching the development in Jena in mid-March when the absolute number of Covid-19 cases
was still low.
Changing the donor pool. In addition, we also check for the sensitivity of the results when
changing the donor pool. This may be important as our baseline specification includes the
region of Heinsberg in the donor pool used to construct the synthetic Jena (with a weight of
4.6%; compare Table A3). As Heinsberg is one of the German regions which was significantly
affected by the Covid-19 pandemic during the Carnival season, this may lead to an
overestimation of the effects of face masks. Accordingly, appendix B.8 presents estimates for
alternative donor pools. Again, we do not find evidence for a significant bias in our baseline
specification. By tendency, the treatment effect becomes larger, particularly if we compare
Jena only to other regions in Thuringia (to rule out macro-regional trends) and to a subsample
of larger cities (kreisfreie Städte). Both subsamples exclude Heinsberg. We also run SCM for
subsamples excluding Thuringia (to rule out spillover effects) and for East and West Germany
(again in search for specific macro regional trends). Generally, these sensitivity tests underline
the robustness of the estimated treatment effect for Jena.
Placebo-in-space tests. A placebo test in space checks whether other cities that did not
introduce face masks on April 6 have nonetheless experienced a decline in the number of
registered Covid-19 cases. If this had been the case, the treatment effect may be driven by
other latent factors rather than face masks. Such latent factors may, for instance, be related to
the macro-regional dynamics of Covid-19 in Germany. Therefore, appendix B.9 reports pseudo-
treatment effects for similarly sized cities in Thuringia assuming that they have introduced face
masks on April 6 although −in fact− they did not. As the figure shows, these cities show either
a significantly higher or similar development of registered Covid-19 compared to their synthetic
controls. This result provides further empirical support for a relevant effect in the case of Jena.
As a more comprehensive test, we also ran placebo-in-space tests for all other regions that did
not introduce face masks on April 6 or closely afterwards. Again, we estimate the same model
on each untreated region, assuming it was treated at the same time as Jena. The empirical
results in Figure 4 indicate that the reduction in the reported number of Covid-19 cases in Jena
clearly exceeds the trend in most other regions − both for the overall sample in Panel A and the
subsample of large cities (kreisfreie Städte) in Panel B.
As outlined above, one advantage of this type of placebo-in-space-test is allows us to conduct
inference. Accordingly, Panel C and Panel D report adjusted p-values that indicate the
probability if the treatment effect for Jena was observed by chance given the distribution of
pseudo-treatment effects in the other German regions (see Galiani and Quistorff, 2017). For
both samples, the reported p-values indicate that the reduction in the number of Covid-19
cases in Jena did not happen by chance but can be attributed to the introduction of face masks,
at the latest - roughly two weeks after the start of the treatment. This timing is again in line
with our above argument that a sufficiently long incubation time and testing lags need to be
considered in the evaluation of treatment effects.

10

Electronic copy available at: https://ssrn.com/abstract=3620634

 Panel A: Placebo in space (all NUTS3 regions on April 6) Panel B: Placebo in space (lager cities [krsfr. Städte] on April 6)

300
400
Diff. (treated - synthetic control)

Diff. (treated - synthetic control)

200
200

100
0
0

-200 -100
-200

March 30 April 26 April 6 April 26

Jena Other NUTS3 regions Jena Only larger cities (krsf. Städte)

Panel C: Significance levels for full sample [Panel A] Panel D: Significance levels for sample of large cities [Panel B]

.5
.5

.4
.4
Adjusted P-values

Adjusted P-values

.3
.3
.2

.2
.1

.1
0

0 5 10 15 20 5 10 15 20

Number of days after introduction of face masks Number of days after introduction of face masks

Figure 4: Comprehensive placebo-in-space tests for the effect of face masks on Covid-19 cases
Notes: Graphs exclude the following regions with a very large number of registered Covid-19 cases: Hamburg
(2000), Berlin (11000), Munich (9162), Cologne (5315) and Heinsberg (5370). In line with Abadie et al. (2010), we
only include placebo effects in the pool for inference if the match quality (pre-treatment RMSPE) of the specific
control regions is smaller than 20 times the match quality of the treated unit. P-values are adjusted for the quality
of the pre-treatment matches (see Galiani and Quistorff, 2017).

Treatment in other districts. Jena may be a unique case. We therefore also study treatment effects
for other regions that have antedated the general introduction of face masks in Germany. Further
single unit treatment analyses are shown in appendix C. Multiple unit treatments are studied in two
ways. The first sample covers all 401 regions and 32 treated units. The second focused on the
subsample of 105 larger cities (kreisfreie Städte), of which 8 are treated units. Treated regions
introduced face masks by April 22. The multiple treatment approach, visible in Figure 5, points to a
significant face mask-effect in the reduction of Covid-19 infections. The adjusted p-values indicate
that the estimated treatment effects are not random.
Face masks may have made a particular difference in the spread of Covid-19, particularly in
larger cities with higher population density and accordingly higher intensity of social
interaction. 10 Over a period of 10 days, we observe an average reduction of 12.3 cases between
treated and control regions. Relative to the average number of cumulative Covid-19 cases on
May 1 in control regions (295.6), this amounts to a reduction of 4.2% of cases. The daily growth

10
This is perfectly in line with Prather et al. (2020) given the reduction in aerosols and droplets via using masks.

11

Electronic copy available at: https://ssrn.com/abstract=3620634

rate of the number of infections correspondingly shrinks by 0.42 percentage points. For the
entire sample, the reduction in the daily growth rate is estimated to be 0.23 percentage points
(see again appendix B.4 for an overview of all measures).
Left Panels: Full sample of NUTS3 regions Right Panels: Subsample of larger cities [krsf. Städte]
Cumulative number Covid-19 cases

260

300
Cumulative number Covid-19 cases
240

280
treated
220

260
synthetic
200

240
control
unit
180

220
-10 -5 0 5 10 -10 -5 0 5 10

Difference treated - synthetic control

Difference treated - synthetic control

5
0

0
-2

-5
-10
-4

-15
-6

-10 -5 0 5 10 -10 -5 0 5 10
Number of days before/after introduction of face masks Number of days before/after introduction of face masks
.1
.5

Adjusted P-values
Adjusted P-values

.02 .04 .06 .08

.4
.3
.2
.1
0

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Number of days after introduction of face masks Number of days after introduction of face masks

Figure 5: Average treatment effects for introduction of face masks (multiple treated units)
Notes Statistical inference for adjusted p-values has been conducted on the basis of a random sample of 1,000,000
placebo averages.

4 Conclusion
We set out by analyzing the city of Jena. The introduction of face masks on 6 April reduced the
number of new infections over the next 20 days by almost 25% relative to the synthetic control
group. This corresponds to a reduction in the average daily growth rate of the total number of
reported infections by 1.32 percentage points. Comparing the daily growth rate in the synthetic
control group with the observed daily growth rate in Jena, the latter shrinks by around 60% due
to the introduction of face masks. This is a sizeable effect. Wearing face masks apparently
helped considerably in reducing the spread of Covid-19. Looking at single treatment effects for
all other regions puts this result in some perspective. The reduction in the growth rate of
infections amounts to 20% only. By contrast, when we take the multiple treatment effect for
larger cities into account, we find a reduction in the growth rate of infections by around 40%.

12

Electronic copy available at: https://ssrn.com/abstract=3620634

What would we reply if we were asked what the effect of introducing face masks would have
been if they had been made compulsory all over Germany? The answer depends, first, on which
of the three percentage measures we found above is the most convincing and, second, on the
point in time when face masks are made compulsory. The second aspect is definitely not only
of academic interest but would play a major role in the case of a second wave.
We believe that the reduction in the growth rates of infections by 40% to 60% is our best
estimate of the effects of face masks. The most convincing argument stresses that Jena
introduced face masks before any other region did so. It announced face masks as the first
region in Germany while in our post-treatment period no other public health measures were
introduced or eased. Hence, it provides the most clear-cut experiment of its effects. Second, as
stated above, Jena is a fairly representative region of Germany in terms of Covid-19 cases.
Third, the smaller effects observed in the multiple treatment analysis may also result from the
fact that −by the time that other regions followed the example of Jena− behavioral adjustments
in Germany’s population had also taken place. Wearing face masks gradually became more
common and more and more people started to adopt their usage even when it was not yet
required.
We should also stress that 40 to 60% might still be a lower bound. The daily growth rates in the
number of infections when face masks were introduced was around 2 to 3%. These are very
low growth rates compared to the early days of the epidemic in Germany, where daily growth
rates also lay above 50% (Wälde, 2020). One might therefore conjecture that the effects might
have been even greater if masks had been introduced earlier.
We simultaneously stress the need for more detailed analyses. First, Germany is only one
country. Different norms or climatic conditions might change the picture for other countries.
Second, we have ignored spatial dependencies in the epidemic diffusion of Covid-19. This might
play a role. Third, there are various types of face masks. We cannot identify differential effects
since mask regulations in German regions do not require a certain type. This calls for further
systematic causal analyses of the different health measure implemented to fight the spread of
Covid-19. Our results provide some initial empirical evidence on this important matter.

References
Abadie A. (2019), Using Synthetic Controls: Feasibility, Data Requirements, and Methodological
Aspects. Article prepared for the Journal of Economic Literature.
https://economics.mit.edu/files/17847
Abadie A., & Gardeazabal J. (2003), The Economic Costs of Conflict: A Case Study of the Basque
Country. American Economic Review, 93(1): 113–132.
https://www.aeaweb.org/articles?id=10.1257/000282803321455188
Abadie A., A. Diamond, & J. Hainmueller (2010), Synthetic Control Methods for Comparative Case
Studies: Estimating the Effect of California’s Tobacco Control Program. Journal of the American
Statistical Association, 105(490): 493–505. https://doi.org/10.1198/jasa.2009.ap08746
Abadie, A., Diamond, A., & Hainmueller, J. (2020), Synth: Stata module to implement synthetic control
methods for comparative case studies. Revised version 2020-05-09.
https://econpapers.repec.org/software/bocbocode/s457334.htm
Becker S., Heblich S., & Sturm D. (2018), The Impact of Public Employment: Evidence from Bonn,
CESifo Working Paper Series 6841, CESifo Group Munich. http://ftp.iza.org/dp11255.pdf
Campos N., Coricelli F., & Moretti L. (2019), Institutional integration and economic growth in Europe.
Journal of Monetary Economics, 103: 88–104. https://doi.org/10.1016/j.jmoneco.2018.08.001

13

Electronic copy available at: https://ssrn.com/abstract=3620634

Donsimoni, J. R., R. Glawion, B. Plachter & K. Wälde (2020a), Projecting the Spread of COVID-19 for
Germany, German Economic Review, 21: 181-216 https://www.iza.org/publications/dp/13094
Donsimoni, J. R., R. Glawion, B. Plachter, K. Wälde & C. Weiser (2020), Should Contact Bans Be Lifted in
Germany? A Quantitative Prediction of Its Effects, CESifo Economic Studies, forthcoming.
https://idw-online.de/de/attachmentdata79709.pdf
Friedson, A., D. McNichols, J.J. Sabia & D. Dave (2020), Did California’s Shelter-In-Place Order Work?
Early Coronavirus-Related Public Health Effects, IZA DP No 13160.
https://www.iza.org/publications/dp/13160
Galiani, S., & B. Quistorff (2017), The synth_runner package: Utilities to automate synthetic control
estimation using synth. The Stata Journal, 17(4), 834–849.
https://doi.org/10.1177%2F1536867X1801700404
Greenhalgh, T., M. B. Schmid, T. Czypionka, D. Bassler & L. Gruer (2020), Face masks for the public
during the covid-19 crisis, BMJ 2020;369:m1435. https://doi.org/10.1136/bmj.m1435
Howard, J., A. Huang, Z. Li, Z. Tufekci, V. Zdimal, H-M. v.d. Westhuizen, A. v. Delft, A. Price, L. Fridman,
L-H. Tang, V. Tang, G. L. Watson, C.E. Bax, R. Shaikh, F. Questier, D. Hernandez, L.F. Chu, C.M.
Ramirez & A. W. Rimoin (2020), Face Masks Against COVID-19: An Evidence Review, Preprints
2020, 2020040203. https://www.doi.org/10.20944/preprints202004.0203.v1
Hu L., R. Kaestner, B. Mazumder, S. Miller & A. Wong (2018), The effect of the affordable care act
Medicaid expansions on financial wellbeing, Journal of Public Economics 163:99-112.
https://doi.org/10.1016/j.jpubeco.2018.04.009.
Kleyer, C., R. Kosfeld, T. Mitze & K. Wälde (2020), Public health measures concerning Covid-19 in
Germany: a systematic overview, mimeo.
Kreif, N., R. Grieve, D. Hangartner, A. J. Turner, S. Nikolova, & M. Sutton (2016). Examina-tion of the
synthetic control method for evaluating health policies with multiple treated units. Health
Economics 25(12): 1514–1528. https://doi.org/10.1002/hec.3258
Lauer, S.A., K.H. Grantz, Q. Bi, F.K. Jones, Q. Zheng, H.R. Meredith, A.S. Azman, N.G. Reich & J. Lessler
(2020), The Incubation Period of Coronavirus Disease 2019 (COVID-19) From Publicly Reported
Confirmed Cases: Estimation and Application, Annals of Internal Medicine 172: 577–582.
https://doi.org/10.7326/M20-0504
Leung, N. H. L., D. K. W. Chu, E. Y. C. Shiu, K-H. Chan, J. J. McDevitt, B. J. P. Hau, H-L. Yen, Y. Li, D. K. M.
Ip, J. S. M. Peiris, W-H. Seto, G. M. Leung, D. K. Milton & B. J. Cowling (2020) Respiratory virus
shedding in exhaled breath and efficacy of face masks, Nat Med 26, 676–680.
https://doi.org/10.1038/s41591-020-0843-2
Linton, N.M., T. Kobayashi, Y. Yang, K. Hayashi, A.R. Akhmetzhanov, S.-M. Jung, B. Yuan, R. Kinoshita &
H. Nishiura (2020), Incubation period and other epidemiological characteristics of 2019 novel
Coronavirus infections with right truncation: A statistical analysis of publicly available case data.
Journal of Clinical Medicine, 9(2): 538. https://doi.org/10.3390/jcm9020538
Marasinghe, K.M. (2020), Concerns around public health recommendations on face mask use among
individuals who are not medically diagnosed with COVID-19 supported by a systematic review
search for evidence., PREPRINT (Version 3) available at Research Square
https://doi.org/10.21203/rs.3.rs-16701/v3.
Pieters, H., R. Curzi, A. Olper & J. Swinnen (2017). Effect of democratic reforms on child mortality: A
synthetic control Analysis. The Lancet Global Health, 4: e627-e632.
https://doi.org/10.1016/S2214-109X(16)30104-8
Prather, K. A., C.C. Wang and R. T. Schooley, 2020, Reducing transmission of SARS-CoV-2, Science
https://doi.org/10.1126/science.abc6197
Robert Koch Institute (2020): Covid-19 Infektionen, General Website (NPGEO Corona Hub):
https://npgeo-corona-npgeo-de.hub.arcgis.com/
Wälde, K. (2020), Corona-Blog, Einschätzung vom Freitag, 20. März 2020,
https://www.macro.economics.uni-mainz.de/2020/03/20/einschatzung-vom-freitag-20-marz/

14

Electronic copy available at: https://ssrn.com/abstract=3620634

 Supplementary Appendix
for

Face Masks Considerably Reduce

Covid-19 Cases in Germany
__
A synthetic control method approach

Timo Mitze(a), Reinhold Kosfeld(b), Johannes Rode(c) and Klaus Wälde(d)

(a) University of Southern Denmark, RWI and RCEA, (b) University of Kassel, (c) TU Darmstadt

(d) Johannes Gutenberg University Mainz, CESifo and Visiting Research Fellow IZA

15

Electronic copy available at: https://ssrn.com/abstract=3620634

A. Timing of introduction of mandatory face masks

Table A1: Overview of dates when masks became compulsory in federal states and districts
Introduction Difference
Public Individual of face in days to
Federal State transport Sales shops NUTS3 region masks state
Baden-Wurttemberg 27.04.2020 27.04.2020 LK Rottweil 17.04.2020 10
Bavaria 27.04.2020 27.04.2020
Berlin 27.04.2020 29.04.2020
Brandenburg 27.04.2020 27.04.2020
Bremen 27.04.2020 27.04.2020
Hamburg 27.04.2020 27.04.2020
Hesse 27.04.2020 27.04.2020 Main-Kinzig-Kreis 20.04.2020 7
Mecklenburg-West Pomer. 27.04.2020 27.04.2020
Lower Saxony 27.04.2020 27.04.2020 Wolfsburg 20.04.2020 7
Braunschweig 25.04.2020 2
North Rhine-Westphalia 27.04.2020 27.04.2020
Rheinland-Pfalz 27.04.2020 27.04.2020
Saarland 27.04.2020 27.04.2020
Saxony 20.04.2020 20.04.2020
Saxony-Anhalt 22.04.2020 22.04.2020
Schleswig-Holstein 29.04.2020 29.04.2020
Thuringia 24.04.2020 24.04.2020 Jena 06.04.2020 18
Nordhausen 14.04.2020 10
Notes: A comprehensive overview of all public health measures introduced in German federal states and individual
regions is given in Kleyer et al. (2020).

16

Electronic copy available at: https://ssrn.com/abstract=3620634

B. Background and additional estimates for SCM application to Jena

This appendix presents supporting findings for the comparative case study of Jena.

B.1. Covid-19 cases and cumulative incidence rate in Jena and Germany on April 5

Panel A: Cumulative number Covid-19 cases (April 5) Panel B: Cumulative Incidence Rate (April 5)
600

300
400

200

Jena
200

100

Jena
0

Figure A1: Box plots for distribution of Covid-19 cases across German NUTS3 regions (April 5)

17

Electronic copy available at: https://ssrn.com/abstract=3620634

B.2. Evaluation of pre-treatment predictor balance and prediction error (RMSPE)

This appendix shows the balancing properties of the SCM approach together with the root
mean square percentage error (RMSPE) as a measure for the quality of the pre-treatment
prediction.

Table A2: Pre-treatment predictor balance and RMSPE for SCM in Figure 2

Introduction of Announcement/
Treatment:
face masks start of campaign
Jena Synthetic Jena Synthetic
control group control group
Cumulative number of registered Covid-19
cases (one and seven days before start of 129.5 129.2 93 92.7
treatment, average)
Number of newly registered Covid-19
cases (last seven days before the start of 3.7 3.8 5 5.2
the treatment, average)
Population density (Population/km2) 38.4 22.8 968.1 947.9
Share of highly educated population (in %) 968.1 1074.3 38.4 26.3
Share of female in population (in %) 50.1 50.1 50.1 50.1
Average age of female population (in
43.5 43.7 43.5 43.9
years)
Average age of male population (in years) 40.5 40.6 40.5 40.8
Old-age dependency ratio (in %) 32.1 29.3 32.1 29.8
Young-age dependency ratio (in %) 20.3 19.6 20.3 19.5
Physicians per 10,000 of population 20.5 19.8 20.5 20.8
Pharmacies per 100,000 of population 28.8 28.7 28.8 28.6
Settlement type (categorial variable) 1 1.3 1 1.9
RMSPE (pre-treatment) 3.145 4.796
Notes: Donor pool includes all other German NUTS3 regions except the two immediate neighboring regions of
Jena (Weimarer Land, Saale-Holzland-Kreis) as well as the regions Nordhausen and Rottweil since the latter regions
introduced face masks in short succession to Jen on April 14 and April 17.

18

Electronic copy available at: https://ssrn.com/abstract=3620634

B.3. Selected control regions and their associated sample weights

Table A3: Distribution of sample weights in donor pool for synthetic Jena
Introduction of face masks (Panel A in Figure 2)
ID NUTS 3 region Weight
13003 Rostock 0.326
6411 Darmstadt 0.311
3453 Cloppenburg 0.118
7211 Trier 0.117
6611 Kassel 0.082
5370 Heinsberg 0.046
Note: Donor pools corresponds to SCM estimation in Panel A
of Figure 2. Sample weights are chosen to minimize the
RMSPE ten days prior to the start of the treatment.

B.4. Growth rates

Jena has 142 registered cases on April 6 compared to an estimated number of 143 cases in the
synthetic control group. On April 26 Jena counts 158 cases and the synthetic control group
reports 205 (again estimated) cases. The daily growth rate in Jena is denoted by xJena and can
be computed from 142 [1+xJena]20 = 158. The daily growth rate in the control group is denoted
by xcontrol and can be computed from 143 [1+xcontrol]20 = 205. Hence, the introduction of the face
mask is associated with a decrease in the number of infections of xcontrol – xJena percentage
points per day.

Table A4: Summary of treatment effects of face mask introduction in Germany

Single Multiple Multiple

Treatment treatments treatments
(Jena) (all districts) (cities)
Percentage change in cumulative number of
-22.9% n.a. n.a.
Covid-19 cases over 20 days
Absolute change in cumulative number of
-23 -5.8 -12.3
Covid-19 cases over 10 days
Percentage change in cumulative number of
-12.8% -2.3% -4.2%
Covid-19 cases over 10 days
Difference in daily growth rates of Covid-19
-1.32% -0.23% -0.42%
cases in percentage points
Reduction in daily growth rates of Covid-19
60.1% 18.94% 37.28%
cases in percent

These numbers are computed in an Excel-file available on the web pages of the authors.

19

Electronic copy available at: https://ssrn.com/abstract=3620634

 B.5. SCM results by age groups

Panel A: 15-34 yrs. / Introduction Panel B: 15-34 yrs. / Announcement

Cumulative number Covid-19 cases

Cumulative number Covid-19 cases

100
80

80
70

60
60

40
20
50

April 27 April 6 April 16 April 26 March 17 March 30 70 April 16 April 26

Jena synthetic control unit Jena synthetic control unit

Panel C: 35-59 yrs. / Introduction Panel D: 35-59 yrs. / Announcement

Cumulative number Covid-19 cases

Cumulative number Covid-19 cases

70

80
60

60
50

40
40

20
30

0
March 27 April 6 April 16 April 26 March 17 March 30 April 6 April 16 April 26

Jena synthetic control unit Jena synthetic control unit

Panel E: 60+ yrs. / Introduction Panel F: 60+ yrs. / Announcement

Cumulative number Covid-19 cases
Cumulative number Covid-19 cases

30
50
40

20
30

10
20
10

March 27 April 6 April 16 April 26 March 17 March 30 April 6 April 16 90

Jena synthetic control unit Jena synthetic control unit

Figure A2: Treatment effects for introduction and announcement of face masks in Jena
Notes: Predictor variables are chosen as for overall specification shown in Figure 2.

Table A5: Sample weights in donor pool for synthetic Jena (cumulative Covid-19 cases; by age groups)
Age Group 15-34 years Age Group 35-59 years Age Group 60 years and above
ID NUTS 3 region Weight ID NUTS 3 region Weight ID NUTS 3 region Weight
1001 Flensburg 0.323 6411 Darmstadt 0.528 6411 Darmstadt 0.522
7211 Trier 0.207 16055 Weimar 0.16 16055 Weimar 0.244
Neustadt a.d.
13003 Rostock 0.184 14511 Chemnitz 0.15 7316 0.068
Weinstraße
5370 Heinsberg 0.142 8221 Baden-Baden 0.07 9562 Erlangen 0.06
Hochtaunus-
3453 Cloppenburg 0.107 6434 0.062 3356 Osterholz 0.056
kreis
Offenbach am
6413 0.038 8435 Bodenseekreis 0.029 5515 Münster 0.027
Main
5370 Heinsberg 0.001 9188 Starnberg 0.022
Note: Donor pools corresponds to SCM estimations in Figure A2. Sample weights are chosen to minimize the
RMSPE ten days prior to the start of the treatment.

20

Electronic copy available at: https://ssrn.com/abstract=3620634

 B.6. Effects on cumulative number of infections per 100,000 inhabitants

Panel A: Overall sample Panel B: Persons aged 15-34 years

180

80
Cumulative Incidence Rate

Cumulative Incidence Rate

160

70
140
120

60
100

50
80

March 27 April 6 April 16 April 26 March 27 April 6 April 16 April 26

Jena synthetic control unit Jena synthetic control unit

Panel C: Persons aged 35-59 years Panel D: Persons aged 60 years and above
60

40
Cumultative Incidence Rate
Cumulative Incidence Rate

50

30
40

20
30

10
20

March 27 April 6 April 16 April 26 March 27 April 6 April 16 April 26

Jena synthetic control unit Jena synthetic control unit

Figure A3: Treatment effects for introduction of face masks on cumulative incidence rate
Notes: See Table 1 for a definition of the incidence rate. Predictor variables are chosen as for overall specification
shown in Figure 2.

Table A6: Sample weights in donor pool for synthetic Jena (cumulative incidence rate)
ID NUTS 3 region Weight
6411 Darmstadt 0.46
15003 Magdeburg 0.171
5370 Heinsberg 0.133
13003 Rostock 0.093
5515 Münster 0.066
11000 Berlin 0.035
12052 Cottbus 0.032
6611 Kassel 0.011
Note: Donor pools corresponds to SCM estimation in Figure A3. Sample
weights are chosen to minimize the RMSPE ten days prior to the start of
the treatment.

21

Electronic copy available at: https://ssrn.com/abstract=3620634

Table A7: Sample weights in donor pool for synthetic Jena (cumulative incidence rate; by age groups)
Age Group 15-34 years Age Group 35-59 years Age Group 60 years and above
ID NUTS 3 region Weight ID NUTS 3 region Weight ID NUTS 3 region Weight
5370 Heinsberg 0.377 6411 Darmstadt 0.419 6411 Darmstadt 0.448
13003 Rostock 0.288 14511 Chemnitz 0.184 14612 Dresden 0.313
1001 Flensburg 0.14 14612 Dresden 0.154 9188 Starnberg 0.071
6611 Kassel 0.138 8221 Heidelberg 0.138 16054 Suhl 0.069
11000 Berlin 0.058 9188 Starnberg 0.088 5515 Münster 0.06
5370 Heinsberg 0.016 8221 Heidelberg 0.039
Note: Donor pools corresponds to SCM estimations in Figure A3. Sample weights are chosen to minimize the
RMSPE ten days prior to the start of the treatment.

B.7. Google trends and announcement effects

100

March 31 April 22
80
60
40
20
0

March 4 March 18 April 1 April 15 April 29

Search: "Face Mask" Search: "Buy Face Mask" Search: "Buy Mask"

Figure A4: Online search for face masks and purchase options according to Google Trends
Note: Online search for keywords (in German) as shown in the legend as Face Mask (“Mund.-Nasen-Schutz”),
Buy Face Mask (“Mundschutz kaufen”) and Buy mask (“Maske kaufen”); alternative keywords show similar peaks
but with a lower number of hits; based on data from Google Trends (2020).

22

Electronic copy available at: https://ssrn.com/abstract=3620634

B.8. Changes in donor pool for synthetic Jena

300
250
200
150
100
50

March 30 April 6 April 26

Jena synth. only Thuringia

synth. ex. Thuringia synth. only larger cities
synth. only East Germany synth. ony West Germany

Figure A5: Treatment effects for changes in donor pool used to construct synthetic Jena
Notes: See main text for a detailed definition of the respective donor pools. Predictor variables are chosen as for
overall specification shown in Figure 2.

Table A8: Sample weights for alternative donor pools used to construct synthetic Jena
Only Thuringia Excluding Thuringia Only larger cities
ID NUTS 3 region Weight ID NUTS 3 region Weight ID NUTS 3 region Weight
16076 Greiz 0.533 13003 Rostock 0.318 6411 Darmstadt 0.504
16051 Erfurt 0.467 6411 Darmstadt 0.302 13003 Rostock 0.304
7211 Trier 0.129 5113 Essen 0.192
3453 Cloppenburg 0.122
6611 Kassel 0.083
5370 Heinsberg 0.046
Only East Germany Only West Germany
ID NUTS 3 region Weight ID NUTS 3 region Weight
16051 Erfurt 0.865 6411 Darmstadt 0.242
14612 Dresden 0.124 3402 Emden 0.198
11000 Berlin 0.011 6611 Kassel 0.169
7211 Trier 0.168
4012 Bremerhaven 0.167
5370 Heinsberg 0.057
Note: Donor pools corresponds to SCM estimations in Figure A5. Sample weights are chosen to minimize the
RMSPE ten days prior to the start of the treatment.

23

Electronic copy available at: https://ssrn.com/abstract=3620634

 B.9. Place-in-space tests for other major cities in Thuringia
Cumulative number Covid-19 cases
140

150
Cumulative number Covid-19 cases
120

100
100

50
80
60

0
March 27 April 6 April 16 April 26 March 27 April 6 April 16 April 26

Erfurt synthetic control unit Gera synthetic control unit

Cumulative number Covid-19 cases

Cumulative number Covid-19 cases

100

200
150
80

100
40 60

50
20

March 30 April 6 April 16 April 26 March 30 April 6 April 16 April 26

Eisenach/Suhl/Weimar synthetic control unit Gotha synthetic control unit

Figure A6: Placebo tests for the effect of face masks in other cities in Thuringia on April 6.
Notes: For the placebo tests in the other cities in Thuringia the same set of predictors as for Jena (Figure 2) has
been applied. The reported regions cover all kreisfreie Städte plus Gotha (Landkreis). The cities Weimar, Suhl and
Eisenach have been aggregated since the number of reported Covid-19 is low in these cities.

24

Electronic copy available at: https://ssrn.com/abstract=3620634

Table A9: Sample weights in donor pool for synthetic control groups (other cities in Thuringia)
Erfurt Gera
ID NUTS 3 region Weight ID NUTS 3 region Weight
13003 Rostock 0.28 15001 Dessau-Roßlau 0.501
16055 Weimar 0.244 16054 Suhl 0.222
3356 Osterholz 0.212 7318 Speyer 0.162
7313 Landau in der Pfalz 0.154 8231 Pforzheim 0.061
6413 Offenbach am Main 0.078 7311 Frankenthal (Pfalz) 0.046
5370 Heinsberg 0.029 8211 Baden-Baden 0.005
5515 Münster 0.004 9662 Schweinfurt 0.003
14521 Erzgebirgskreis 0.001
Weimar/Suhl/Eisenach Gotha
ID NUTS 3 region Weight ID NUTS 3 region Weight
15001 Dessau-Roßlau 0.263 15081 Altmarkkreis 0.23
12052 Cottbus 0.236 16077 Altenburger Land 0.164
13004 Schwerin 0.202 15086 Jerichower 0.161
9361 Amberg 0.177 3402 Emden 0.111
14626 Görlitz 0.069 16071 Weimarer Land 0.108
9363 Weiden i.d. Opf. 0.036 16074 Saale-Holzland-Kreis 0.063
14521 Erzgebirgskreis 0.008 16061 Eichsfeld 0.058
9184 München 0.005 16070 Ilm-Kreis 0.055
6411 Darmstadt 0.005 3453 Cloppenburg 0.027
15003 Magdeburg 0.017
4012 Bremerhaven 0.007
Note: Donor pools corresponds to SCM estimations in Figure A6. Sample weights are
chosen to minimize the RMSPE ten days prior to the start of the treatment.

25

Electronic copy available at: https://ssrn.com/abstract=3620634

C. The effect in other German cities and regions (single treatment analyses)
In addition to Jena, we test for treatment effects in Nordhausen, Rottweil, Main-Kinzig-Kreis,
and Wolfsburg (compare Figure 1). We ignore Braunschweig here as the introduction became
effective only two days in advance of its federal state.
Cumulative number Covid-19 cases

Cumulative number Covid-19 cases

60

600
50

500
40

400
30

300
20

April 6 April 14 April 26 April 6 April 17 April 26

Nordhausen synthetic control unit Rottweil synthetic control unit

Cumulative number Covid-19 cases

Cumulative number Covid-19 cases
600

300
500

280
400

260
300

240
200

220

April 6 April 20 April 26 70 April 20 April 26

Main-Kinzig-Kreis synthetic control unit Wolfsburg synthetic control unit

Figure A7: Treatment effects for introduction of face masks in other cities
Notes: Nordhausen (Thuringia, April 14, top left), Rottweil (Baden Württemberg, April 17, top right), Wolfsburg
(Lower Saxony, April 20, middle left), Main-Kinzig-Kreis (Hessia, April 20, middle right). Predictor variables are
chosen as for overall specification shown in Figure 2.

As the figure shows, the result is 2:1:1. Rottweil and Wolfsburg display a positive effect of
mandatory mask wearing, just as Jena. The results in Nordhausen are very small or unclear. In
the region of Main-Kinzig, it even seems to be the case that masks increased the number of
cases relative to the synthetic control group. As all of these regions introduced masks after
Jena, the time period available to identify effects is smaller than for Jena. The effects of
mandatory face masks could also be underestimated as announcement effects and learning
from Jena might have induced individuals to wear masks already before they became
mandatory. Finally, the average pre-treatment RMSPE for these four regions (7.150) is larger
than for the case of Jena (3.145). For instance, in the case of the region of Main-Kinzig it is more
than three times as high (9.719), which indicates a lower pre-treatment fit. The obtained
treatment effects should then be interpreted with some care as the pre-sample error could also
translate into the treatment period. In order to minimize the influence of a poor pre-treatment
fit for some individual regions, the main text therefore compares the results in Jena mainly with
a multiple unit treatment approach.

26

Electronic copy available at: https://ssrn.com/abstract=3620634

Table A10: Sample weights in donor pool for synthetic controls (other treated NUTS3 regions)
Nordhausen Rottweil
ID NUTS 3 region Weight ID NUTS 3 region Weight
16069 Hildburghausen 0.228 8327 Tuttlingen 0.324
6636 Werra-Meißner-Kreis 0.209 5966 Olpe 0.216
16064 Unstrut-Hainich-Kreis 0.168 8136 Ostalbkreis 0.2
16054 Suhl 0.109 16071 Weimarer Land 0.063
3402 Emden 0.093 14521 Erzgebirgskreis 0.06
12073 Uckermark 0.071 3102 Salzgitter 0.043
12053 Frankfurt (Oder) 0.07 16061 Eichsfeld 0.035
3354 Lüchow-Dannenberg 0.051 9187 Rosenheim 0.031
9279 Dingolfing-Landau 0.025
3455 Friesland 0.003
Main-Kinzig-Kreis Wolfsburg
ID NUTS 3 region Weight ID NUTS 3 region Weight
8136 Ostalbkreis 0.193 8212 Karlsruhe 0.357
1062 Stormarn 0.168 8221 Heidelberg 0.189
5966 Olpe 0.113 8211 Baden-Baden 0.158
6433 Groß-Gerau 0.105 10046 St. Wendel 0.128
9473 Coburg 0.092 14511 Chemnitz 0.071
5562 Recklinghausen 0.063 5117 Mülheim an der Ruhr 0.059
7313 Landau in der Pfalz 0.059 5315 Köln 0.028
9171 Altrötting 0.056 15003 Magdeburg 0.007
7338 Rhein-Pfalz-Kreis 0.047 9663 Würzburg 0.004
6437 Odenwaldkreis 0.041
8236 Enzkreis 0.041
3159 Göttingen 0.023
Note: Donor pools corresponds to SCM estimations in Figure A7. Sample weights are chosen to
minimize the RMSPE ten days prior to the start of the treatment.

27

Electronic copy available at: https://ssrn.com/abstract=3620634

D. A brief survey of public health measures against Covid-19
Our approach goes in line with various studies that have already tried to better understand the
effect of public health measures on the spread of Covid-19 (Barbarossa et al., 2020, Hartl et al.,
2020, Donsimoni et al., 2020, Dehning et al., 2020, Gros et al., 2020, Adamik et al, 2020).
However, these earlier studies all take an aggregate approach in the sense that they look at
implementation dates for a certain measure and search for subsequent changes in the national
incidence. There are some prior analyses that take a regional focus (Khailaie et al. 2020) but no
attention is paid to the effect of policy measures. 11
There are also many cross-country analyses, both in a structural SIR (susceptible, infectious and
removed) sense (Chen and Qiu, 2020) and with an econometric focus on forecasting the end of
the pandemic (Ritschl, 2020). Others draw parallels between earlier pandemics and Covid-19
(Barro et al., 2020). These studies do not explicitly take public health measures into account.
Some studies discuss potential effects of public health measures and survey general findings
(Wilder-Smith et al. 2020, Anderson et al., 2020, Ferguson et al, 2020) but do not provide direct
statistical evidence on specific measures.
The synthetic control method (SCM) has been applied by Friedson et al. (2020) to estimate the
effect of the shelter-in-place order for California, USA, in the development of Covid-19. The
authors find inter alia that around 1600 deaths from Covid-19 have been avoided by this
measure during the first four weeks. The effects of face masks have been surveyed by Howard
et al. (2020) and Greenhalgh et al. (2020). Greenhalgh et al. (2020) mainly presents evidence
on the effect of face masks during non-Covid epidemics (influenza and SARS). Marasinghe
(2020) reports that they “did not find any studies that investigated the effectiveness of face
mask use in limiting the spread of COVID-19 among those who are not medically diagnosed with
COVID-19 to support current public health recommendations”.
In addition to medical aspects (like transmission characteristics of Covid-19 and filtering
capabilities of masks), Howard et al. (2020) survey evidence on mask efficiency and on the
effect of a population. They first stress that “no randomized control trials on the use of masks
<…> has been published”. The study which is “the most relevant paper” for Howard et al. (2020)
is one that analyzed “exhaled breath and coughs of children and adults with acute respiratory
illness” (Leung et al., 2020, p. 676), i.e. used a clinical setting. Concerning the effect of masks
on community transmissions, the survey needs to rely on pre-Covid-19 studies.
We conclude from this literature review that our paper is the first analysis that provides field
evidence on the effect of masks on mitigating the spread of Covid-19.

11
In a short note, Hartl and Weber (2020) apply panel methods based on time dummies to understand the relative
importance of various public health measures. They employ data at the federal state level and not at the regional
level. As a detailed model description is not available, an appreciation of results is difficult at this point.

28

Electronic copy available at: https://ssrn.com/abstract=3620634

References (not appearing in the main text)
Adamik, B., M. Bawiec, V. Bezborodov, W. Bock, M. Bodych, J. Burgard, T. Götz, T. Krueger, A.
Migalska, B. Pabjan, T. Ozanski, E. Rafajlowicz, W. Rafajlowicz, E. Skubalska-Rafajlowiczc, S.
Ryfczynska, E. Szczureki & P. Szymanski (2020), Mitigation and herd immunity strategy for COVID-
19 is likely to fail, medRxiv preprint, https://doi.org/10.1101/2020.03.25.20043109
Anderson, R.M., H. Heesterbeek, D. Klinkenberg & T. D. Hollingsworth (2020), How will country-based
mitigation measures influence the course of the COVID-19 epidemic?, The Lancet 395(10228),
931-34. https://doi.org/10.1016/s0140-6736(20)30567-5
Barbarossa, M., J. Fuhrmann, J. Meinke, S. Krieg, H. Vinod Varma, N. Castelletti & T. Lippert (2020), The
impact of current and future control measures on the spread of COVID-19 in Germany, medRxiv
preprint, https://doi.org/10.1101/2020.04.18.20069955
Barro, R., J. Ursua & J. Wenig (2020), The Coronavirus and the Great Influenza Epidemic Lessons from
the ‘SpanishFlu’ for the Coronavirus’s Potential Effectson Mortality and Economic Activity,
CESifoWorkingPaper 8166. https://www.cesifo.org/en/publikationen/2020/working-
paper/coronavirus-and-great-influenza-epidemic-lessons-spanish-flu.
Chen, X. & Z. Qiu (2020), Scenario analysis of non-pharmaceutical interventions on global Covid-19
transmissions, CEPR Press, Covid Economics, Vetted and Real-Time Papers, 7.
https://cepr.org/sites/default/files/news/CovidEconomics7.pdf
Dehning, J., J. Zierenberg, F.P. Spitzner, M. Wibral, J.P. Neto, M. Wilczek & V. Priesemann (2020),
Inferring COVID-19 spreading rates and potential change points for case number forecasts,
Science. https://doi.org/10.1126/science.abb9789
Ferguson, N.M. et al. (2020), Impact of non-pharmaceutical interventions (NPIs) to reduce COVID-19
mortality and healthcare demand, Imperial College COVID-19 Response Team.
https://doi.org/10.25561/77482
Gros, C., R. Valenti, K. Valenti & D. Gros (2020), Strategies for controlling the medical and socio-
economic costs of the Corona pandemic. https://arxiv.org/abs/2004.00493
Hartl, T., & E. Weber (2020m. Welche Maßnahmen brachten Corona unter Kontrolle?
https://www.oekonomenstimme.org/artikel/2020/05/welche-massnahmen-brachten-corona-
unter-kontrolle/
Hartl, T., Wälde, K. & E. Weber (2020), Measuring the impact of the German public shutdown on the
spread of Covid-19, Covid Economics 1: 25-32.
https://cepr.org/sites/default/files/news/CovidEcon1%20final.pdf
Khailaie, S., T. Mitra, A. Bandyopadhyay, M. Schips, P. Mascheroni, P. Vanella, B. Lange, S. Binder & M.
Meyer-Hermann (2020), Estimate of the development of the epidemic reproduction number Rt
from Coronavirus SARS-CoV-2 case data and implications for political measures based on
prognostics, medRxiv preprint, https://doi.org/10.1101/2020.04.04.20053637.
Ritschl, A. (2020), Visualising and forecasting Covid-19, Covid Economics 5: 79-105.
https://cepr.org/sites/default/files/news/CovidEconomics5.pdf
Wilder-Smith, A., C. Chiew & V. Lee (2020), Can we contain the COVID-19 outbreak with the same
measures as for SARS? The Lancet. Infectious Diseases
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7102636/

29

Electronic copy available at: https://ssrn.com/abstract=3620634