The Effect of a Nighttime Curfew on the Spread of COVID-19 (January 30, 2023)
Summary
A research paper by Johannes Apel, Niklas Rohde and Jan Marcus of Universität Hamburg and Freie Universität Berlin, dated January 30, 2023, on the nighttime curfew in Hamburg, Germany. The curfew came into effect on April 2, 2021 and prohibited people from leaving home between 9 p.m. and 5 a.m., except for physical training, walking dogs and commuting to work. Using difference-in-differences and synthetic control methods, the authors compare the seven-day incidence rate in Hamburg with a pool of 35 German cities over an analysis period beginning February 1, 2021. The abstract states that the curfew was effective in reducing the number of Covid-19 cases. The paper closes with an appendix robustness table and an in-space placebo figure.
Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used
Cite this document
“The Effect of a Nighttime Curfew on the Spread of COVID-19 (January 30, 2023).”
Original at doi.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/the-effect-of-a-nighttime-curfew-on-the-spread-of-covid-19-1019c5ee44/
, SHA-256 364d97d1ed29a18e9002b6c3de54e8aeb02a20e5631ad016f07e3248c575cd60.Full text
The effect of a nighttime curfew on the spread of COVID-19
Johannes Apel,a Niklas Rohde,a Jan Marcusb
a
Universität Hamburg; b Freie Universität Berlin
January 30, 2023
Abstract
While nighttime curfews are less severe restrictions compared to around-the-clock curfews in
mitigating the spread of Covid-19, they are nevertheless highly controversial, with the scarce lit-
erature on their effectiveness providing mixed evidence. We study the effectiveness of the night-
time curfew in Hamburg, Germany’s second largest city, in mitigating the spread of Covid-19.
This curfew forbid people from leaving their home between 9 p.m. and 5 a.m. for non-essential
businesses. Applying both difference-in-differences and synthetic control group methods, we
find that the curfew was effective in reducing the number of Covid-19 cases. As it is unclear
whether and how the virus will mutate in the next time, policy-makers might have to resort
to non-pharmaceutical interventions again. Nighttime curfews should be kept in the toolbox of
policy-makers to fight Covid-19.
Keywords: Covid-19, Curfews, Non-pharmaceutical interventions, Synthetic control, Germany
1. Introduction
Governments around the world have implemented various non-pharmaceutical inter-
ventions (NPI) to slow the spread of Covid-19. Most of these interventions severely restrict
the lives of the affected populations (including, e.g., school closures, stay-at-home orders,
closings of restaurants, travel restrictions). Hence, governments constantly face a trade-
off between containing the pandemic and restricting civil rights. While strict lockdown
measures that allow people to leave the house only for essential businesses are found to
be effective in reducing the spread of Covid-19 (see, e.g., Flaxman et al., 2020; Hsiang
et al., 2020), they come with high social and economic costs (see, e.g., Bartik et al., 2020;
Chetty et al., 2020). For this reason, many governments (e.g., in Canada, France, Ger-
many, Greece, and Spain) implemented nighttime curfews, in which citizens are required
to stay at home during the night except for specific, well-defined reasons. The idea of
a nighttime curfew is to reduce private indoor gatherings, where the risk of infections is
way higher than outdoors as people are not allowed to leave their homes for non-essential
reasons.
While nighttime curfews are less severe restrictions than around-the-clock curfews,
they are highly controversial. First, nighttime curfews nevertheless substantially restrict
the affected individuals. Second, as these curfews do not affect daytime activities, they
may not be effective in fighting the spread of Covid-19. They might be even counterpro-
ductive if they increase contact density during earlier hours of the day (Sprengholz et al.,
2021). Generally, governments should not implement ineffective measures with substan-
tial side-effects. However, it can strengthen the public adherence to an intervention if
people are convinced of its effectiveness. For all these reasons, it is crucial to understand
the effectiveness of nighttime curfews.
This paper examines the effectiveness of nighttime curfews, based on the curfew in the
German city of Hamburg that came into effect on April 2, 2021. The curfew prohibited
1
people from leaving their home between 9 p.m. and 5 a.m., except for physical training,
walking dogs, and commuting to work. We compare the development of Covid-19 cases
in Hamburg, Germany’s second largest city with almost 2 million inhabitants, around
the introduction of the curfew with the development of Covid-19 cases in other German
regions over the same period. Applying both difference-in-differences and synthetic control
group methods, our empirical analyses show that the nighttime curfew was effective in
reducing the number of Covid-19 cases.
This study contributes to the literature on the effectiveness of different non-
pharmaceutical interventions to mitigate the spread of Covid-19 by focusing on nighttime
curfews, an intervention that restricts the lives of citizens in a less harsh manner than
full lockdowns and around-the-clock stay-at-home orders. Although many countries and
regions have implemented curfews at some point, there is surprisingly little evidence on
the effectiveness of nighttime curfews to contain the Covid-19 pandemic – and the existing
literature provides rather mixed evidence.
Sprengholz et al. (2021) examine public perceptions of nighttime curfews: The ma-
jority of participants in their survey did not perceive nighttime curfews as an effective
measure to contain the pandemic and stated that they would not reduce private contacts
due to a nighttime curfew. A substantial share of participants reported that they would
rather shift a private dinner meeting forward in time, suggesting that nighttime curfews
may increase contact density in the hours of the day when the curfew is not in place.
While these statements refer to stated behaviors and not to actual behavior, the study
by Sprengholz et al. (2021) highlights that nighttime curfews might backfire and actually
increase Covid infections.
However, several studies find that nighttime curfews reduce mobility. For instance,
Ghasemi et al. (2021) provide evidence that a nighttime curfew in Ontario/Canada sub-
stantially reduced nighttime mobility. Similarly, Velias et al. (2022) show that a temporal
extension of a nighttime curfew in the Attica region of Greece led to a small decrease
2
in mobility. These are important insights, but mobility data are not a direct measure of
the pandemic situation and reductions in mobility do not necessarily translate into fewer
Covid infections. For instance, nighttime curfews might induce two households to meet at
one household’s home instead of outdoors in a park. In this case, there would be a decline
in mobility (as one household stayed at home), while Covid cases might still increase as
the risk of infection is higher indoors than outdoors. Therefore, it is important to directly
examine measures of the pandemic situation like incidence rates or virus reproduction
numbers as well.
In a modeling study that examines several NPIs at the same time, Sharma et al.
(2021) find that curfews might be effective in slowing the spread of Covid-19. Andronico
et al. (2021) obtain similar results for French Guiana in another modeling study, while
Dimeglio et al. (2021) provide evidence that a temporal extension of a nighttime curfew
in Toulouse, France, actually increased infection rates slightly. de Haas et al. (2021) find
in an ex-post evaluation that nighttime curfews in the German state of Hesse did not
reduce Covid-19 infections. In Hesse, counties had to implement nighttime curfews when
the incidence rate exceeded a certain threshold. This is different to the situation that
we study: Hamburg made its own decision to implement a nighttime curfew. Further,
given that there were many other regions in Germany with similar infection rates, the
construction of a similar control group is feasible.
We contribute to this literature by focusing on a setting that offers several advantages
for a clean identification of curfew effects. First, there were many other regions in Germany
with a similar Covid-19 situation before the introduction of the curfew, allowing to use
a control group with a very similar pre-treatment level and development of Covid-19
cases. Second, due to the analysis at the sub-national level, many other factors that
affect the pandemic situation are very similar in Hamburg and the other German regions.
For instance, other interventions were implemented or agreed upon at the national level
(e.g., travel restrictions, vaccine availability). While our analysis relates to early 2021
3
when vaccines were not widely available, nobody knows whether and how the virus will
mutate in the future (Ledford, 2022) and it is not unlikely that policy-makers will need to
resort again to non-pharmaceutical interventions to contain the pandemic. Therefore, it is
important to know about the effectiveness of previous non-pharmaceutical interventions.
The remainder of the paper is structured as follows. Section 2 provides further in-
formation about the curfew in Hamburg, the data, and the applied statistical methods.
Section 3 presents the main empirical results, probes the robustness of the main findings,
and examines effect heterogeneity. Section 4 presents a discussion of the findings and
Section 5 concludes.
2. Materials and methods
2.1. Background
Hamburg introduced a nighttime curfew on April 2, 2021, as one of the first regions in
Germany. The curfew came into effect on Good Friday, just before the Easter weekend.The
curfew prohibited people from leaving their home between 9 p.m. and 5 a.m., except for
physical training, walking dogs, and commuting to work. In Hamburg, the nighttime
curfew was officially removed on May 12, one week after the incidence rate, the average
number of daily new infections per 100,000 inhabitants, fell below the symbolic value of
100 for the first time since the curfew implementation.
Hamburg’s curfew preempted the nationwide regulation for nighttime curfews in Ger-
many: From April 23, 2021, regions with an incidence rate above 100 had to implement
a nighttime curfew. The implementation of the curfews based on the federal regulation
is endogenous in the sense that the curfews automatically follow from the number of
weekly Covid-19 cases exceeding a specific threshold. This is an important difference
to the nighttime curfew in Hamburg that we analyze. While the implementation of the
curfew in Hamburg was also related to the development of the pandemic situation, it
was at Hamburg’s discretion to implement the curfew. At that time, many other cities
4
in Germany had similar or even higher incidence rates but did not implement a curfew.
Bavaria and Saxony implemented similar regulations before the nationwide regulations.
Therefore, we exclude cities from these two states from our main analyses. Our results
are, however, robust to including theses cities as well (see Section 3.2).
At the time of the curfew implementation, B.1.1.7 (“Alpha”) was the dominant SARS-
CoV-2 variant in Germany. Across Germany, bars and restaurants were closed, it was
required to wear medical masks indoors, private gatherings were restricted, and testing
capacities were scarce. Further, vaccines against Covid-19 were not widely available in
Germany and were restricted to specific groups, including people older than 80, residents
in nursing homes, and medical staff. About 9.5 million inhabitants (11.3 %) had received
a first vaccine dosage, while about 4 million (5 %) had received two vaccinations (Robert
Koch Institute, 2021).
2.2. Data and sample
The curfew was implemented on April 2, 2021, and our main analysis period begins
on February 1, 2021. Hence, we have about two months of pre-treatment data, which
should be a sufficient period to compare the development of the outcome variable before
the implementation of the curfew. Having a longer pre-treatment period increases the risk
that other regional policies were implemented in some of the control units. In Section 3.2,
we show that our results are insensitive to different lengths of the pre-treatment period.
Our analysis period ends on April 30, one week after the national night curfew regulations
came into force on April 23. Hence, our analysis period consists of t = 89 days.
While Hamburg constitutes our treatment group, the pool of potential control units
consists of all German cities that have a population of at least 100,000 inhabitants and that
are independent counties (Kreisfreie Städte). Hence, we consider only urban regions in
order to make the pool of potential control units more comparable in terms of population
density and the provision of free Covid-19 testing stations. We further restrict this poten-
5
tial pool by excluding all cities that introduced nighttime curfews before the nationwide
regulation for nighttime curfews in our observation period. This is done by first checking
every Covid-related state-order (Corona-Landesverordnung) during our observation pe-
riod. Further, we double-checked all remaining cities by working through local newspapers
and local directives. We dropped Hagen, Halle (Saale), Heilbronn, Karlsruhe, Krefeld,
Köln, Leverkusen, Mannheim, Mülheim an der Ruhr, Offenbach, Osnabrück, Potsdam,
Remscheid, Salzgitter, Kaiserslautern, Koblenz, Ludwigshafen, Mainz, Stuttgart, Wolfs-
burg and Wuppertal because of local curfews as well as cities in Saxony and Bavaria
because of state-wide curfew regulations.
In the end, the pool of potential control units consists of 35 cities (see Table A.1 in the
Appendix for an overview over the included cities). Hence, our balanced panel consists of
N = 3204 observations (89 days x (35+1) cities).
The main outcome variable of this study is based on the number of officially reported
Covid-19 cases, including late reporting. More specifically, we use the so-called seven-day
incidence rate, which is the most commonly used metric in Germany to measure Covid-19
cases as this measure allows to smooth-out differential reporting behaviors on different
days of the week. For each city, we take the number of reported new infections across the
past seven days. We normalize this number by the size of the city’s population to take
into account differences in population size across cities. We focus on the incidence rate in
levels (and not in logs) as this is the metric used most commonly in the media and easier
to interpret. Moreover, several other NPI studies rely on Covid cases in levels as well (e.g.
Friedson et al., 2021; Isphording et al., 2021; Backhaus, 2022; Diederichs et al., 2022).
However, in Section 3.2, we show that our conclusions do not change when we apply
the log transformation to the incidence rate. The seven-day incidence rate is provided
by Corona Datenplattform (https://www.corona-datenplattform.de/), which collects
Covid-19 related data on behalf of the German Federal Ministry for Economic Affairs and
Climate Action.
6
2.3. Empirical strategy
We employ both difference-in-differences and synthetic control methods. Both meth-
ods compare the change in Covid-19 cases before and after the introduction of the night-
time curfew in Hamburg with the change in Covid-19 cases in a control group at the
same time. The two methods are strongly related but differ in the selection of the control
group.
For both methods, the causal interpretation depends on the assumption that Covid-
19 cases in Hamburg would follow the same trend as in the control group – if Hamburg
had not implemented the curfew. This is the common trend assumption. While this
assumption cannot be directly tested, we provide additional evidence for their plausibility
by examining whether Hamburg and the control group follow a similar trend before the
introduction of the curfew.
2.3.1. Difference-in-differences
We first estimate difference-in-differences (DiD) regressions, in which we regress the
seven-day incidence rate on (i) a binary indicator variable that takes on the value one
for all observations in Hamburg, our treatment group, (ii) a binary indicator variable
that takes on the values one for all observations after April 2, the date of the curfew
implementation, and zero else, and (iii) the interaction of the first two indicator variables.
The coefficient of this interaction term is the coefficient of interest and denotes the effect
of the curfew on Covid-19 cases.
Our main DiD specification is based on a two-way fixed effects model with city and day
fixed effects. The city fixed effects take into account general differences in the outcome
variable between cities and the day fixed effects flexibly control for general changes in the
outcome variable over time. Further, we include separate indicators for the first ten days
of the curfew in Hamburg, the second ten days, and the remaining nine days, i.e. up to
seven days after the implementation of the nationwide curfew regulations. This allows
7
investigating the effect of the curfew at different points after its implementation. Due to
the incubation period and the lag in reporting, we do expect little or no effects in the
first 10 days after the curfew is in place. As in our setting the treatment starts at a single
point in time, there is no benefit of applying any of the recently developed two-way fixed
effects estimators that were specifically designed for settings with staggered treatment
timing (see, e.g., Callaway and Sant’Anna, 2021).
2.3.2. Synthetic control method
The Synthetic Control Method (SCM), first introduced by Abadie and Gardeazabal
(2003), gained wide popularity after Abadie et al. (2010) and is an extension of DiD
(Cunningham, 2021). The idea behind the SCM is that a combination of units often
provides a better comparison for the treatment unit than any single unit alone. Therefore,
the method uses a weighted average of units in the donor pool to construct a synthetic
version of the treatment group, i. e. a synthetic Hamburg. Several studies on NPIs to
contain the Covid-19 pandemic apply SCM (e.g., Mitze et al., 2020; Born et al., 2021).
Our aim is that the synthetic Hamburg resembles the real Hamburg closely with re-
spect to (i) the development of the incidence rate before the introduction of the curfew
and (ii) pre-determined characteristics that are likely to relate to the development of the
incidence rate after the curfew implementation. Therefore, we consider the incidence rate
on each of the 21 days before April 2, 2021, (curfew implementation) in the construction
of the synthetic Hamburg. We further include population density as a measure for the
transmission of the virus as well as median age and life expectancy as measures for viru-
lence. We also consider the share of people who recovered from Covid-19 as well as the
share of twice vaccinated people. Additionally, we take into account the poverty rate to
capture the ability to comply with movement restrictions as well as the share of voters
of the right-wing party Alternative für Deutschland (AfD; Alternative for Germany) as
a measure of the willingness to comply with Covid containment policies. The AfD is
8
connected to the anti-vaccination movement and publicly opposes most of the Covid-19
measures. Information on population and population density, people recovered, as well as
the election results are provided by Corona Datenplattform. All other control variables
(median age, life expectancy, poverty rate) originate from Federal Statistical Office of
Germany (2021).
The method of synthetic controls reweights the control cities in such a way that the
reweighted control group closely matches Hamburg with respect to the aforementioned
variables. Details of the SCM weighting procedure are described in Abadie et al. (2010)
and Abadie et al. (2015). We implement SCM using version 0.0.7 of Stata’s user-written
package synth (Abadie et al., 2010). Table A.1 in the Appendix shows that the syn-
thetic Hamburg has multiple donors. The major ones are Frankfurt (Main) with 38.3%,
Bremerhaven with 13.7%, Bonn with 11.6%, Ulm with 8.5% and Gelsenkirchen with 7.7%
contribution. On top of that, six more cities contribute a little into the synthetic Ham-
burg.
One criticism of SCM is that it might induce a regression-to-the-mean bias similar to
matched difference-in-differences estimators (Daw and Hatfield, 2018) if treated and con-
trol cities are drawn from populations with different distributions of the outcome variable
(Illenberger et al., 2020). While it is unclear why the distribution of the outcome variable
should be generally different in treated and control cities in our (pandemic) setting, as
a robustness exercise, we apply the recently developed synthetic difference-in-differences
(SDiD) estimator (Arkhangelsky et al., 2021), which Illenberger et al. (2020) discuss as
one potential solution to the regression-to-the-mean issue. SDiD aims to combine the
merits of DiD and SCM by not only reweighting the control cities to match the treatment
group but by also reweighting days from the pre-treatment period to match the average
outcome in the post-treatment period for each of the control cities. We implement SDiD
using Stata’s user-written program sdid in the version 1.2.0 (Pailañir and Clarke, 2022).
9
3. Results
3.1. Main Results
We begin the empirical analysis by plotting our outcome variable, the 7-day incidence
rate, separately for the treatment and control groups around the introduction of the
curfew.
Figure 1: The development of Covid-19 incidence rates in the treatment and control
groups
200
150
Incidence Rate
100
50
Feb 1 Mar 1 Apr 1 May 1
Hamburg Control cities
Note: The graph displays the development of our main outcome variable, the seven-day number of
reported new infections per 100,000 inhabitants (incidence rate) separately for Hamburg and the 35
control cities (as the average) before and after the introduction of the nighttime curfew in Hamburg.
The dashed vertical line indicates the date of the curfew implementation (April 2, 2021), while the
solid vertical line indicates the date one week after curfew implementation as the curfew is unlikely to
be effective immediately after is implementation due to Covid-19’s incubation period and subsequent
reporting lags.
Figure 1 shows several noteworthy features. First, the development of the outcome
variable is rather similar in Hamburg and the control group before the implementation of
10
the curfew, supporting the common trend assumption. Only in the first half of February
2021 do the two trends differ. In Section 3.2, we show that our findings are robust to
excluding February from the pre-treatment period. Second, the incidence rate in the pre-
treatment period is higher in Hamburg compared to the control group. As the level of the
outcome variable might relate to the future development of the outcome, we show in the
next section that our findings are confirmed by SCM, where we work with a control group
that has a very similar pre-treatment outcome level as Hamburg. Third, already shortly
before the curfew, Covid-19 cases start to decrease in Hamburg. This is likely due to the
Easter holidays, which covers the week before and the week after Easter (i.e., from March
29 to April 9, 2021) in most states. During the Easter holidays, there is no Covid testing
in schools and many people take off days from work. Fourth, this decline in Hamburg
matches a similar decline in the control group. Fifth, developments in the treatment and
control groups are very similar in the first days after the implementation of the curfew.
This is not surprising given the incubation period and the lag in reporting and, hence,
further supports the common trend assumption. Lastly, and most importantly, several
days after the implementation of the curfew, trends diverge and the number of Covid
cases declines in Hamburg relative to the cases in the control group. This suggests that
the curfew reduced Covid cases.
We now turn to the DiD regression results in Table 1. The first column reports the
DiD coefficient of the baseline DiD. This column indicates that the curfew reduced the
incidence rate in Hamburg by 25.9 cases per 100,000 inhabitants. This effect is statistically
significant at the 1% level. The size of the estimated effect stays constant when we include
city and day fixed effects in column (2). Column (3) shows that the curfew did not reduce
Covid cases in the first ten days after its implementation. This was to be expected given
the incubation period and the lag in reporting. However, the estimated effect of the
nighttime curfew is large and statistically significant 11-20 days and 21-29 days after the
11
Table 1: The effect of Hamburg’s nighttime curfew: DiD estimation
Difference-in-differences Synthetic control Synthetic DiD
(1) (2) (3) (4) (5)
Curfew -25.9*** -25.9*** — — —
(5.4) (5.4)
Curfew1−10 — — 13.2*** 3.7 -1.0
(3.0) (6.8) (12.7)
Curfew11−20 — — -26.6*** -29.1** -45.1
(5.9) (10.5) (29.5)
Curfew20+ — — -68.4*** -83.1*** -86.8**
(9.0) (12.9) (43.3)
N 3204 3204 3204 1068 704
City fixed effects ✓ ✓ ✓
Day fixed effects ✓ ✓ ✓ ✓
Note: The table displays the effect of the nighttime curfew in Hamburg on the reported Covid-19 incidence
rate. Column (1) is based on the baseline DiD estimation, while columns (2) and (3) control for a full set
of day and city fixed effects. Column (3) splits the overall curfew indicator in indicators for the curfew
in the first 10 days, the second ten days, and the next nine days. Column (4) applies the city-specific
weights from the synthetic control approach to the specification in column (3), while column (5) weights
this specification with the weights from the synthetic difference-in-differences approach. Standard errors
in parentheses (* p < 0.1; ** p < 0.05; *** p < 0.01).
12
curfew implementation, suggesting that the night curfew had a substantial negative effect
on the Covid-19 incidence rate in Hamburg.
Figure 2: The evolution of the incidence rate in Hamburg and its synthetic version
200
150
Incidence rate
100
50
Feb 1 Mar 1 Apr 1 May 1
Hamburg synthetic Hamburg
Note: The graph displays the development of the incidence rate separately for Hamburg and its synthetic
counterpart. The dashed vertical line indicates the date of the curfew implementation (April 2, 2021),
while the solid vertical line indicates the date one week after the curfew implementation.
Figure 2 provides the SCM results graphically. It is evident that the level and trend
of the outcome variable are even more similar between Hamburg and its synthetic coun-
terpart in the pre-treatment period than in Figure 1. The trends are also very similar
in the days following the curfew, when the curfew is unlikely to have a direct impact
on Covid cases. This suggests that the synthetic control group manages to mimic quite
well what would have happened in Hamburg if there was no curfew in place. About 6-8
days after the curfew came into effect, the incidence rate increases much stronger in the
synthetic Hamburg than in the real Hamburg. In Hamburg the incidence rate decreases
13
after Mid-April, while it increases in the synthetic Hamburg until almost the end of April.
To quantify the SCM effects, column (4) in Table 1 applies the city-specific SCM weights
to our main specification. The SCM estimates are slightly more negative than the main
DiD results. When we apply the SDiD approach, the estimated effects are even more
negative for 11-20 days and 21-29 days after the curfew implementation. However, the
point estimate for 11-20 days after the ban is no longer statistically significant (with a
p-value of 0.126), which might be due to the fact that in our case SDiD makes only use
of three days in the pre-treatment period, thereby substantially reducing the sample size.
3.2. Robustness
This section examines the robustness of our findings with respect to (i) different time
periods, (ii) different control groups, and (iii) various estimation issues. Columns (1) and
(2) of Table A.2 change the length of the pre-treatment period by minus and plus one
month, respectively. Column (3) disregards the week before and the week after Easter
(i.e., March 29 to April 11, 2021) as the implementation of the curfew in Hamburg was just
before the Easter weekend. These two weeks are school holidays in most German states,
in which many people take days off from work for holidays or family visits. Goodman-
Bacon and Marcus (2020) discuss spillover effects as a threat for the identification of causal
effects of NPIs. For this purpose, column (4) excludes all cities from the northern German
states (Schleswig-Holstein, Bremen, Lower Saxony, Mecklenburg-West Pomerania) from
the control group. Note that the neighboring regions of Hamburg are not part of our
main control group. The nearest city in the main control group is Kiel, which is about
100 kilometers away from Hamburg. All other control cities are even further away.
Columns (5) and (6) add further cities to the control group. Column (5) also considers
the eight cities with a population of more than 100,000 that are not independent counties,
while column (6) also includes cities from Bavaria and Saxony, the two states that imple-
mented similar curfew regulations before the nationwide regulations. Curfews in cities in
14
these states were implemented both in the pre-treatment period and the post-treatment
period as well as in parallel to the curfew in Hamburg. Column (7) weights all cities by
their population, while column (8) takes the natural logarithm of the incidence rate to
analyze whether our findings are robust to this transformation of the dependent variable.
The latter robustness test is particularly relevant as the common trend assumption gen-
erally holds either in levels or in logs. When using the outcome in levels, we implicitly
assume that the outcome develops linearly, while the log-specification is more directed
to exponential growth rates. While the latter might be seen as more appropriate for a
pandemic with exponential growth rates, Figure 1 shows that the linearity assumption
provides a good approximation for the actual development of the incidence rate in the
weeks preceding the curfew. It is reassuring that the results in column (8) show that the
conclusion of our study does not depend on the scaling of the outcome variable. The
point estimates suggest that the curfew reduced the incidence rate by about 25% in the
second 10 days and by about 45% 21-29 days after the curfew implementation. Column
(9) uses daily new cases as outcome variable in order to facilitate a comparison of results
to other studies. The obtained daily point estimates imply that in the first 29 days af-
ter its implementation the curfew reduced cumulative Covid cases per 100,000 by about
160. As Hamburg has a population of about 1.84 million inhabitants, this means that the
curfew prevented almost 3,000 reported infections in this time span.
All robustness checks confirm our main findings: The curfew did not reduce reported
Covid cases during the first ten days after its implementation, while it substantially
reduced reported Covid cases in the subsequent period.
Additionally, we perform placebo tests for the SCM results, following the idea of
the “In-space-placebo”-tests by Abadie et al. (2010). More specifically, we construct a
synthetic version of each city in the control group and plot the outcome difference between
each city and its synthetic version over time. Figure A.1 in the appendix shows the curfew
effect in Hamburg (black thick line) compared to the placebo curfew effects in all other
15
cities (gray thin lines). It is striking that in the second half of the post-treatment period
the curfew effect in Hamburg is more negative than all the placebo curfew effects.
Abadie et al. (2010) and Abadie et al. (2015) suggest to use the distribution of these
placebo effects for an alternative method of statistical inference, similar to randomization
inference. If there was no curfew effect, the probability that Hamburg would exhibit the
most negative difference between the actual city and its synthetic counterpart would be
p = 1/36 = 0.028. This is highly unlikely and, therefore, increases our confidence that
the obtained effect is indeed due to the curfew and not due to random.
3.3. Effect heterogeneity
Table 2: Heterogeneity in the curfew effect
By age By sex
Main 5-14 15-34 35-59 60-79 80+ Women Men
(1) (2) (3) (4) (5) (6) (7) (8)
Curfew1−10 13.2 62.7*** 10.6 6.5 -2.6 1.9 15.1* 9.9
(8.2) (13.8) (12.0) (10.0) (6.8) (14.0) (8.3) (8.8)
Curfew11−20 -26.6*** 6.5 -37.6*** -37.9*** -20.2*** -11.6 -25.0*** -30.4***
(8.2) (13.8) (12.0) (10.0) (6.8) (14.0) (8.3) (8.8)
Curfew20+ -68.4*** -113.8*** -87.8*** -75.2*** -32.0*** -15.7 -69.9*** -65.8***
(8.6) (14.4) (12.6) (10.4) (7.1) (14.7) (8.7) (9.2)
N 3204 3204 3204 3204 3204 3204 3204 3204
Note: The table displays the effect of the nighttime curfew on the reported Covid-19 incidence rate
separately for different age groups and for women and men. Standard errors in parentheses (* p < 0.1;
** p < 0.05; *** p < 0.01).
Information on the number of reported Covid cases is provided separately by age
groups and sex, allowing to examine whether the curfew affected these groups differently.
Table 2 shows the results for different age groups. The curfew did not reduce the incidence
rate in any of the age groups in the first ten days after its implementation. However, over
the next ten days we observe strong declines in the incidence rate for the age groups
15-34 and 35-59, the age groups in which people most often have private gatherings in the
evening. The effects for the other age groups are either insignificant (ages 5-14 and 80+)
16
and/or clearly smaller (ages 60-79). In the following period (more than 20 days after the
curfew implementation), we see stronger reductions across all age groups. This is in line
with the idea of a trickle-down effect in the sense that non-infections in one age group
will be also beneficial for other age groups in subsequent periods.
Finally, Table 2 shows that the effects are similar for women and men and, hence,
provides little evidence for effect heterogeneity with respect to sex.
4. Discussion
This paper addresses the effectiveness of nighttime curfews in mitigating the spread
of Covid-19. Based on the curfew in Germany’s second largest city, Hamburg, in April
2021, the paper provides evidence that this curfew substantially reduced Covid-19 cases.
About one month after its implementation, the curfew decreased cumulative Covid cases
by about 160 per 100,000 population; put differently, the curfew prevented about 3,000
reported infections in Hamburg in this time span. We provide several arguments that
support a causal interpretation of our results. First, before the implementation of the
curfew, Hamburg and the control group exhibited very similar trends with respect to the
outcome variable, making it more plausible that trends would have been similar in the
period after if Hamburg had not implemented the curfew (common trend assumption).
Second, our finding is confirmed both by difference-in-differences and synthetic control
group methods. Third, we observe several patterns in the data that are in line with the
effectiveness of the curfew: The curfew does not reduce reported Covid cases in the first
days after its implementation, when the curfew is unlikely to be effective due to Covid-19’s
incubation period and subsequent reporting lags. Further, we find that the curfew has
the strongest effect in the age groups that are most likely to have private gatherings in
the evening.
In order to better understand the magnitude of our estimates, it is helpful to compare
the estimated curfew effect to the effect of other NPIs. However, such a comparison is
17
complicated by several factors. First, different studies use different outcome measures
to evaluate the effectiveness of specific NPIs (e.g., case numbers, reproduction number,
death rates). Second, studies that look at a wide range of different NPIs do not take
into account the endogeneity of the policy and co-treatments in the form of other NPIs
implemented at the same time or only a few days apart. Therefore, we focus on studies
that evaluate the effectiveness of single interventions as these studies often apply more
convincing strategies to identify causal effects. Third, studies focus on different stages
of the pandemic and the magnitude of the NPI effects might differ substantially across
the different stages (e.g., due to different outcome levels, different variants of the virus,
vaccine availability etc.). For all these reasons, the following comparisons have to be taken
with a grain of salt.
Friedson et al. (2021) analyze the shelter-in-place order in California in March 2020,
which was the first in the U.S. and included the closure of non-essential businesses and
a stay-at-home order for all non-essential activities. They find that this shelter-in-place
order reduced cumulative Covid cases per 100,000 population by 160.9 to 194.7 about one
month after its implementation. While our estimates for the curfew effect imply a similar
magnitude on cumulative cases (see Section 3.2) as the lower bound of the shelter-in-place
order effect, case numbers were way lower in California in March 2020, meaning that the
proportionate effect of the SIPO was clearly larger. Backhaus (2022) studies the partial
relaxation of travel restrictions in Europe in summer 2020, exploiting the different start of
school holidays across German states. He finds that daily new cases increased by about 0.5
per 100 000 population 16-30 days after the school break. This point estimate is less than
one tenth of the daily curfew effect (see Table A.2). Also exploiting differences in school
holidays across German states, Isphording et al. (2021) show that school re-openings in
summer 2020 did not increase the number of Covid infections. All in all, the comparison
with previous findings on different NPIs suggests that the curfew is more effective than
18
international travel restrictions and school closures but less effective than stay-at-home
orders (but also less restrictive).
Our results refer to a situation when vaccines were not widely available and a variant
of SARS-CoV-2 dominated (Alpha) that was more contagious than the original type but
clearly less contagious than the Omicron variant that dominated in most countries at the
beginning of 2022. A crucial question is what can be learned from our results for the fu-
ture handling of the pandemic. It is unclear whether and how the virus will mutate in the
next months and weeks (Ledford, 2022). It is not unlikely, however, that a new mutation
of the virus will emerge with a high degree of immune evasion that causes severe disease.
As it may take time to adjust the vaccine production to these new mutants, policy-makers
might have to resort to non-pharmaceutical interventions again. Moreover, there might be
situations where the number of sick people approaches a threshold where critical infras-
tructure like hospitals, police force, public transport or child care and education threatens
to collapse. Such a collapse could not only lead to higher death rates and the postpone-
ment of necessary medical interventions but also to severe economic costs. Therefore, it is
important to know about the effectiveness of previous non-pharmaceutical interventions
in curbing the pandemic and controlling disease intensity. Because of the high share of
vaccinated people and the case of multiple infections, the aims of non-pharmaceutical in-
terventions shift from providing mostly medical protection to the maintenance of critical
infrastructure. In line with this, several virologist (e.g., Germany’s leading coronavirus
expert Christian Drosten) think that in autumn/winter 2022/23 policy-makers will have
to resort to harsher measures again (Der Spiegel, 2022). Further, several studies pro-
vide evidence that political trust induces higher rates of compliance with Covid-related
containment policies (see e.g. Bargain and Aminjonov, 2020; Brodeur et al., 2021). How-
ever, in order to have trust in policy-makers, it is crucial that citizens are convinced that
governments implement effective and reasonable policies. Therefore, it is important to
provide evidence about the effectiveness of specific policies.
19
5. Conclusion
We conclude that the April 2021 nighttime curfew in Hamburg was effective in miti-
gating the spread of Covid-19 and that nighttime curfews should be kept in the toolbox of
policy-makers to fight Covid-19. Curfews might be particularly useful in order to prevent
situations, in which critical infrastructure like hospitals and the police force are overbur-
dened due to a high share of infections among employees. While nighttime curfews restrict
affected individuals, these restrictions are less severe than around-the-clock curfews.
20
References
Abadie, A., Diamond, A., and Hainmueller, J. (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.
Abadie, A., Diamond, A., and Hainmueller, J. (2015). Comparative politics and the
synthetic control method. American Journal of Political Science, 59(2):495–510.
Abadie, A. and Gardeazabal, J. (2003). The economic costs of conflict: A case study of
the basque country. American Economic Review, 93(1):113–132.
Andronico, A., Tran Kiem, C., Paireau, J., Succo, T., Bosetti, P., Lefrancq, N., Nacher,
M., Djossou, F., Sanna, A., Flamand, C., Salje, H., Rousseau, C., and Cauchemez, S.
(2021). Evaluating the impact of curfews and other measures on SARS-CoV-2 trans-
mission in French Guiana. Nature Communications, 12(1):1634.
Arkhangelsky, D., Athey, S., Hirshberg, D. A., Imbens, G. W., and Wager, S. (2021).
Synthetic difference-in-differences. American Economic Review, 111(12):4088–4118.
Backhaus, A. (2022). International travel in times of the covid-19 pandemic: The case of
german school breaks. Economics & Human Biology, 44:101090.
Bargain, O. and Aminjonov, U. (2020). Trust and compliance to public health policies in
times of COVID-19. Journal of Public Economics, 192:104316.
Bartik, A. W., Bertrand, M., Cullen, Z., Glaeser, E. L., Luca, M., and Stanton, C. (2020).
The impact of COVID-19 on small business outcomes and expectations. Proceedings of
the National Academy of Sciences, 117(30):17656–17666.
Born, B., Dietrich, A. M., and Müller, G. J. (2021). The lockdown effect: A counterfactual
for Sweden. Plos one, 16(4):e0249732.
Brodeur, A., Grigoryeva, I., and Kattan, L. (2021). Stay-at-home orders, social distancing,
and trust. Journal of Population Economics, 34(4):1321–1354.
Callaway, B. and Sant’Anna, P. H. (2021). Difference-in-differences with multiple time
periods. Journal of Econometrics, 225(2):200–230.
Chetty, R., Friedman, J. N., Hendren, N., Stepner, M., et al. (2020). The economic
impacts of covid-19: Evidence from a new public database built using private sector
data. NBER Working Paper, 27431.
Cunningham, S. (2021). Causal inference: The mixtape. Yale University Press.
Daw, J. R. and Hatfield, L. A. (2018). Matching and regression to the mean in difference-
in-differences analysis. Health services research, 53(6):4138–4156.
21
de Haas, S., Goetz, G., and Heim, S. (2021). Measuring the effects of COVID-19-related
night curfews: Empirical evidence from Germany. MAGKS Joint Discussion Paper
Series in Economics.
Der Spiegel (2022). Drosten warnt vor Corona-Leichtsinn: Wir haben keinen infektions-
freien Sommer. https: // www. spiegel. de/ wissenschaft/ christian-drosten-
ueber-corona-das-virus-ist-nicht-absolut-harmlos-geworden-a-ff87c8eb-
d574-45cb-ab49-96200795ae9c .
Diederichs, M., van Ewijk, R., Isphording, I. E., and Pestel, N. (2022). Schools under
mandatory testing can mitigate the spread of sars-cov-2. Proceedings of the National
Academy of Sciences, 119(26):e2201724119.
Dimeglio, C., Miedougé, M., Loubes, J.-M., Mansuy, J.-M., and Izopet, J. (2021). Side
effect of a 6 pm curfew for preventing the spread of sars-cov-2: A modeling study from
toulouse, france. Journal of Infection, 82(5):186–230.
Federal Statistical Office of Germany (2021). Genesis-Online: Database of the Fed-
eral Statistical Office of Germany. https: // www-genesis. destatis. de/ genesis/
online? operation= sprachwechsel& language= en .
Flaxman, S., Mishra, S., Gandy, A., Unwin, H. J. T., Mellan, T. A., Coupland, H., Whit-
taker, C., Zhu, H., Berah, T., Eaton, J. W., Monod, M., Perez-Guzman, P. N., Schmit,
N., Cilloni, L., Ainslie, K. E. C., Baguelin, M., Boonyasiri, A., Boyd, O., Cattarino,
L., Cooper, L. V., Cucunubá, Z., Cuomo-Dannenburg, G., Dighe, A., Djaafara, B.,
Dorigatti, I., van Elsland, S. L., FitzJohn, R. G., Gaythorpe, K. A. M., Geidelberg, L.,
Grassly, N. C., Green, W. D., Hallett, T., Hamlet, A., Hinsley, W., Jeffrey, B., Knock,
E., Laydon, D. J., Nedjati-Gilani, G., Nouvellet, P., Parag, K. V., Siveroni, I., Thomp-
son, H. A., Verity, R., Volz, E., Walters, C. E., Wang, H., Wang, Y., Watson, O. J.,
Winskill, P., Xi, X., Walker, P. G. T., Ghani, A. C., Donnelly, C. A., Riley, S., Vollmer,
M. A. C., Ferguson, N. M., Okell, L. C., and Bhatt, S. (2020). Estimating the effects of
non-pharmaceutical interventions on COVID-19 in Europe. Nature, 584(7820):257–261.
Friedson, A. I., McNichols, D., Sabia, J. J., and Dave, D. (2021). Shelter-in-place orders
and public health: evidence from california during the covid-19 pandemic. Journal of
Policy Analysis and Management, 40(1):258–283.
Ghasemi, A., Daneman, N., Berry, I., Buchan, S. A., Soucy, J.-P., Sturrock, S., and
Brown, K. A. (2021). Impact of a nighttime curfew on overnight mobility. medRxiv
preprint doi: https: // doi. org/ 10. 1101/ 2021. 04. 04. 21254906 .
Goodman-Bacon, A. and Marcus, J. (2020). Using difference-in-differences to identify
causal effects of covid-19 policies. Survey Research Methods, 14(2):153–158.
Hsiang, S., Allen, D., Annan-Phan, S., Bell, K., Bolliger, I., Chong, T., Druckenmiller,
H., Huang, L. Y., Hultgren, A., Krasovich, E., Lau, P., Lee, J., Rolf, E., Tseng, J.,
22
and Wu, T. (2020). The effect of large-scale anti-contagion policies on the COVID-19
pandemic. Nature, 584(7820):262–267.
Illenberger, N. A., Small, D. S., and Shaw, P. A. (2020). Impact of regression to the
mean on the synthetic control method: bias and sensitivity analysis. Epidemiology,
31(6):815–822.
Isphording, I. E., Lipfert, M., and Pestel, N. (2021). Does re-opening schools contribute to
the spread of sars-cov-2? evidence from staggered summer breaks in germany. Journal
of Public Economics, 198:104426.
Ledford, H. (2022). The next variant: Three key questions about what’s after Omicron.
Nature, 603(7900):212–213.
Mitze, T., Kosfeld, R., Rode, J., and Wälde, K. (2020). Face masks considerably re-
duce COVID-19 cases in Germany. Proceedings of the National Academy of Sciences,
117(51):32293–32301.
Pailañir, D. and Clarke, D. (2022). Sdid: Stata module to perform synthetic difference-
in-differences estimation, inference, and visualization.
Robert Koch Institute (2021). Täglicher Lagebericht des RKI zur Coronavirus-Krankheit-
2019 (COVID-19) 31.03.2021. Technical report.
Sharma, M., Mindermann, S., Rogers-Smith, C., Leech, G., Snodin, B., Ahuja, J., Sand-
brink, J. B., Monrad, J. T., Altman, G., Dhaliwal, G., Finnveden, L., Norman, A. J.,
Oehm, S. B., Sandkühler, J. F., Aitchison, L., Gavenčiak, T., Mellan, T., Kulveit, J.,
Chindelevitch, L., Flaxman, S., Gal, Y., Mishra, S., Bhatt, S., and Brauner, J. M.
(2021). Understanding the effectiveness of government interventions against the resur-
gence of COVID-19 in Europe. Nature Communications, 12(1):5820.
Sprengholz, P., Siegers, R., Goldhahn, L., Eitze, S., and Betsch, C. (2021). Good night:
Experimental evidence that nighttime curfews may fuel disease dynamics by increasing
contact density. Social Science & Medicine, 286:114324.
Velias, A., Georganas, S., and Vandoros, S. (2022). COVID-19: Early evening curfews
and mobility. Social Science & Medicine, 292:114538.
23
Appendix
Table A.1: City weights in the construction of the synthetic Hamburg
City Weight City Weight
Berlin 0.008 Herne 0.062
Bielefeld 0.053 Jena 0.006
Bochum 0 Kiel 0
Bonn 0.116 Darmstadt 0
Bottrop 0 Kassel 0
Braunschweig 0 Wiesbaden 0
Bremen 0 Lübeck 0
Bremerhaven 0.137 Magdeburg 0
Dortmund 0 Mönchengladbach 0
Duisburg 0 Münster 0
Düsseldorf 0 Oberhausen 0.056
Erfurt 0 Oldenburg 0
Essen 0 Pforzheim 0
Frankfurt 0.383 Rostock 0
Freiburg 0 Solingen 0
Gelsenkirchen 0.077 Trier 0
Hamm 0 Ulm 0.085
Heidelberg 0.019
24
Table A.2: Sensitivity analyses
Alternative time periods Alternative control groups Estimation
-1 +1 excl. w/o 4 more more daily
Main month month Easter states cities cities 2 weights log(y) infect.
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Curfew1−10 5.7** 17.4*** 0.0 9.4*** 14.9*** 0.6 12.1*** -0.0 0.4
(2.7) (3.3) (.) (3.3) (3.0) (3.8) (2.5) (0.0) (0.4)
Curfew11−20 -34.1*** -22.3*** -25.2*** -33.9*** -25.7*** -49.6*** -27.1*** -0.3*** -6.8***
(5.5) (5.6) (6.0) (5.9) (5.5) (5.7) (4.9) (0.0) (1.0)
Curfew20+ -75.9*** -64.2*** -67.0*** -81.6*** -67.6*** -83.8*** -62.9*** -0.6*** -10.8***
(8.6) (8.3) (9.0) (9.1) (7.9) (6.3) (10.0) (0.1) (1.4)
N 2196 4320 2700 2581 3916 6052 3204 3204 3204
Note: The table displays various robustness tests for the effect of the nighttime curfew on the reported
Covid-19 incidence rate. Columns (1) and (2) change the length of the pre-treatment period by minus
and plus one month, while column (3) disregards the week before and the week after Easter. Column
(4) excludes all cities from the northern German states (Schleswig-Holstein, Bremen, Lower Saxony,
Mecklenburg-West Pomerania) from the control group. Column (5) also considers the eight cities with a
population of more than 100,000 that are not independent counties, while column (6) also includes cities
from Bavaria and Saxony, the two states implemented similar regulations before the nationwide regula-
tions. Column (7) weights all cities by their population, while column (8) takes the natural logarithm
of the incidence rate. Column (9) uses as outcome daily new cases. Standard errors in parentheses (*
p < 0.1; ** p < 0.05; *** p < 0.01).
25
Figure A.1: In-space-placebo results
200
Difference in incidences
100
0
-100
Feb 1 Mar 1 Apr 1 May 1
date
Note: The graph displays the differences in incidences between every city and their synthetic control.
The thick black line represents Hamburg and the thin gray lines represent all 35 other cities. The dashed
vertical line indicates the date of the curfew implementation (April 2, 2021), while the solid vertical line
indicates the date one week after the curfew implementation.
26
File and source
- File
- the-effect-of-a-nighttime-curfew-on-the-spread-of-covid-19.pdf
- Size
- 642,609 bytes
- SHA-256
- 364d97d1ed29a18e9002b6c3de54e8aeb02a20e5631ad016f07e3248c575cd60
- Original
- doi.org