Content uploaded by Nicolas Robert Ziebarth
Author content
All content in this area was uploaded by Nicolas Robert Ziebarth on Sep 07, 2018
Content may be subject to copyright.
DISCUSSION PAPER SERIES
IZA DP No. 11631
Michael Kvasnicka
Thomas Siedler
Nicolas R. Ziebarth
The Health Effects of Smoking Bans:
Evidence from German Hospitalization Data
JUNE 2018
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.
Schaumburg-Lippe-Straße 5–9
53113 Bonn, Germany
Phone: +49-228-3894-0
Email: publications@iza.org www.iza.org
IZA – Institute of Labor Economics
DISCUSSION PAPER SERIES
IZA DP No. 11631
The Health Effects of Smoking Bans:
Evidence from German Hospitalization Data
JUNE 2018
Michael Kvasnicka
Otto-von-Guericke-University Magdeburg, RWI and IZA
Thomas Siedler
University of Hamburg and IZA
Nicolas R. Ziebarth
Cornell University, DIW Berlin and IZA
ABSTRACT
IZA DP No. 11631 JUNE 2018
The Health Effects of Smoking Bans:
Evidence from German Hospitalization Data*
This paper studies the short-term impact of public smoking bans on hospitalizations in
Germany. It exploits the staggered implementation of smoking bans over time and across
the 16 federal states along with the universe of hospitalizations from 2000-2008 and
daily county-level weather and pollution data. Smoking bans in bars and restaurants have
been effective in preventing 1.9 hospital admissions (-2.1%) due to cardiovascular diseases
per day, per 1 million population. We also find a decrease by 0.5 admissions (-6.5%)
due to asthma per day, per 1 million population. The health prevention effects are more
pronounced on sunny days and days with higher ambient pollution levels.
JEL Classification: D12, H19, I12, I18
Keywords: smoking bans, health effects, hospital admissions,
second-hand smoke
Corresponding author:
Thomas Siedler
University of Hamburg
Esplanade 36
20354 Hamburg
Germany
E-mail: Thomas.Siedler@wiso.uni-hamburg.de
* We thank seminar participants at the 2015 Meeting of the American Economic Association (AEA) and the 2014
Meeting of the European Society for Population Economics (ESPE). We also thank the research data centre (RDC)
of the Federal Statistical Office and Statistical Offices of the Länder, the German Meteorological Service (Deutscher
Wetterdienst (DWD)) and the German Federal Environmental Office (Umweltbundesamt (UBA)) for providing
administrative data access. We also thank Jörg Blankenback, Maike Schmitt and Martin Karlsson for valuable support
in extrapolating the weather and pollution measures. Excellent research assistance by Peter Eibich, Phillip Susser and
Beining Niu is gratefully acknowledged. We take responsibility for all remaining errors and shortcomings of the article.
The research reported in this paper is not the result of a for-pay consulting relationship. Our employers do not have a
financial interest in the topic of the paper which might constitute a conflict of interest.
1. Introduction
In a recent report on trends in smoking prevalence, the World Health Organization (WHO)
estimates that tobacco use is responsible for the death of around six million people worldwide
each year (World Health Organization, 2015a). This includes about 600,000 people who are
estimated to die each year due to exposure to second-hand smoke. Reducing smoking
prevalence and exposure to environmental tobacco smoke is one of the key public health
priorities of the WHO and many governments around the world.
This paper extends previous work on the causal health effects of smoke-free legislation in
several important ways. First, we exploit variation in smoke-free laws across states and over
time in Germany along with exceptionally rich high-quality register data to investigate the
effects on hospital admissions. Our data cover the universe of more than 160 million
hospitalizations between 2000 and 2008 in Germany. It enable us to separate time effects,
state effects, and smoke-free legislation effects which makes it less likely that unobserved
factors (coinciding with the introduction of public smoking bans) confound the estimates.
Second, we control more comprehensively than most existing studies for potentially
important environmental factors that are likely to affect hospitalizations, such as local weather
and local pollution conditions. To the best of our knowledge, our paper is one of the first to
study potential interaction effects between these environmental factors and the effectiveness
of public smoking bans. Accounting for such environmental interactions is important as the
combination of smoking and exposure to second-hand smoke with, for example, high pollution
levels might result in more hospitalizations due to cardiovascular diseases.
Third, we investigate objective health outcome measures (i.e., cardiovascular hospital
admissions, asthma admissions) using high-frequency register data.1 Most previous studies in
1 A comprehensive literature documents the direct link between smoking, smoking intensity, exposure to second-
hand smoke and cardiovascular events (see, for example, U.S. Department of Health and Human Services, 2006,
2014; Mons et al., 2015, and references therein).
2
the economics literature investigate changes in smoking behavior using survey data.2 The
findings of these studies are mixed. Whereas some studies find a reduction in tobacco
consumption following smoke-free laws (e.g. Evans et al., 1999 for the US and workplace
smoking bans), others report no or only very small changes (Adda and Cornaglia, 2010 for the
US; Anger et al., 2011 for Germany; Carpenter et al., 2011 for Canada; Jones et al. 2015 for the
UK). Kuehnle and Wunder (2017) report improvements in self-assessed health among non-
smokers following the introduction of smoking bans in Germany, but no effects among
smokers. Exploiting variation over time and 40 European countries, Odermatt and Stutzer
(2015) find that smoking bans increase the happiness of smokers who would like to quit. A
similar finding has been reported by Gruber and Mullainathan (2005) who exploit cigarette tax
variation in the US. More generally, our paper contributes to the rich reduced-form literature
evaluating the causal effects of tobacco policies (mostly cigarette taxes) on smoking behavior
(Gruber, 2001; DeCicca et al., 2002, 2008, 2013; Gruber et al., 2003; Gruber and Koszegi, 2004;
Gruber and Frakes, 2006; Avery et al., 2007; Carpenter and Cook, 2008; Lovenheim, 2008;
Courtemache, 2009; Goolsbee et al., 2010; Shetty et al., 2011; Wehby and Courtemanche,
2012; Kvasnicka and Tauchmann, 2012; Lillard et al., 2013; Callison and Kaestner, 2014;
Friedman, 2015; Rozema and Ziebarth, 2017; Hansen et al., 2017).
Fourth, we study the consequences of smoking bans in a high-income country with a high
smoking prevalence. The estimated age-standardized WHO prevalence rate among those aged
15 years and more is 31% in Germany, compared to 18% in the United States, 20% in the United
Kingdom and Northern Ireland, and 13% in Australia (World Health Organization, 2015b).
However, the latter countries have been researched most extensively in the literature (Cutler
and Glaeser, 2009; World Health Organization, 2015b).
Finally, our setting provides a potential explanation for why it is so difficult to identify
(objective) health effects following smoking bans. Even in a country like Germany with 82
million residents, a relatively high smoking prevalence, a high population density, low access
barriers to the hospital infrastructure (universal insurance coverage and no provider networks),
2 Shetty at al. (2011) is one exception and estimates the effects of US smoking bans on hospitalizations.
3
state-year variation in smoking bans, and data on the universe of around 17 million annual
hospital admissions, we operate in a setting where statistical power becomes an issue. For
example, our -0.19 point estimate for the largest disease category cardiovascular admissions
per 100,000 population is still precisely estimated—even though the effect size is only two
percent of the mean. However, trying to assess the effects for a subset of this disease
category—such as heart attacks or deaths after cardiovascular admissions—is technically very
challenging due to power issues. In Germany with about 15 heart attacks per day and 1 million
residents, heart attacks are still a relatively rare event and small changes in these rates are very
hard to identify when using rich sets of time and region fixed effects, even with register data.
Our findings show that, in the short run, cardiovascular admissions decreased by 1.9 cases
per 1 million population (-2.1%) because of state-level smoking bans in Germany between 2007
and 2008. We also find a significant short-run reduction in hospital admissions due to asthma
(-6.5%). Many countries around the world have not yet banned smoking in public. Our results
suggest that these countries could achieve health improvements by banning public smoking.
Our findings also suggest that health improvements can be expected even if countries are
unable or unwilling to impose 100% smoke-free laws. Public smoking bans in Germany are not
very strict. Exemptions exist (e.g., smoking bars and separate smoking rooms in some premises)
and the bans are not always strictly enforced. On the other hand, in the only published
economics paper that uses hospitalizations as an outcome, Shetty et al. (2011) do not find
evidence for significant decreases in hospitalizations in the US. Bitler et al. (2010) study state-
level clear indoor air laws in the U.S. and find evidence that these laws were strongly enforced
in bars, but not in restaurants, private workplaces, schools, or government buildings.
This paper also enriches the previous literature by showing that the reduction in
cardiovascular admissions is reinforced on sunny days. We discuss several potential
mechanisms for these findings. Our estimates suggest that smoking bans are more effective
under adverse environmental conditions (unrelated to tobacco smoke) that negatively affect
the human body, such as high pollution levels.
4
The next section briefly describes the staggered implementation of the smoking bans in
Germany. Section 3 describes the datasets, Section 4 discusses the empirical strategy, and
Section 5 presents and discusses our findings. Section 6 concludes.
2. Smoking Bans in Germany
At a meeting in March 2007, the 16 state health ministers decided to ban smoking in bars
and restaurants in all 16 German federal states. Shortly after, smoke-free policies were
introduced over a time period of just 13 months (August 2007 to August 2008). Our empirical
analysis exploits this state-time variation across the sixteen federal states. We distinguish states
by the exact month when smoking bans in bars and restaurants were legally implemented. As
Table 1 and Figure 1 show, the south-western state of Baden-Wuerttemberg was the first to
introduce smoke-free legislation in August 2007. By the end of August 2008, all states in
Germany had introduced public smoking bans (see Table 1).
[Insert Table 1 and Figure 1 about here]
The German bans were less comprehensive than in other countries and allowed for
exemptions. All states except Bavaria allowed smoking in separate “smoking rooms” in bars and
restaurants and some states made additional exceptions. In practice, because of the
bureaucratic regulations for separate smoking rooms, one can summarize that the ban was very
effective in banning smoking in restaurants (see, for example, Baxmann and Eckoldt, 2007;
NRauchSchG SH, 2007; LNRSchG, 2007). It was also effective in banning smoking in bigger and
popular (tourist) bars. However, basically each state allowed small pubs to self-declare as
“smoker pubs” where people could still smoke inside. In practice, almost every German city still
had a small number of such “smoker pubs” that were typically attended by a large share of
(local) smokers who also tend to drink heavily. 3 Germany does not have a general closing time
3 Note that there exist no official figures for the number of “smoker pubs” in Germany.
5
for bars and that it was not unusual that such smoker bars were open 24/7. Moreover, smoking
outside of bars and restaurants is not prohibited and still common in Germany.
Although tobacco control measures increased both in number and strictness since the early
2000s in Germany (for a review, see Göhlmann and Schmidt, 2008), their overall effectiveness
remained low in comparison to other European countries (Joossen et al., 2011). Landmark
policy initiatives include a federal law that made the protection of non-smokers in the
workplace mandatory in 2002, and the introduction of a nation-wide smoking ban in federal
public buildings and transportation in September 2007 as well as the concurrent increase of the
minimum smoking age from 16 to 18. Note that our econometric models include month-year
fixed effects which net out common time effects among all German states.
More detailed information on the smoke-free legislation in Germany can be found in Anger
et al. (2011), Brüderl and Ludwig (2011), and Kuehnle and Wunder (2017). In particular, Anger
et al. (2011) examine whether the timing of the implementation of the smoke-free German
legislation is associated with various pre-ban characteristics at the state level. The authors do
not find statistically significant associations with (i) the percentage of smokers in a state’s
population, (ii) whether the state government is conservative, (iii) the average age of the state
residents, (iv) the proportion of university graduates, (v) the proportion of singles in the state’s
population, or (vi) the state’s GDP per capita. Hence, the variation in smoke-free legislation in
Germany over time and states likely provides plausible exogenous variation to study the causal
effects of public smoking bans.
3. Datasets
We make use of high-quality register data. We use the census of all German hospital
admissions from 2000 to 2008 and link these data with weather measures, pollution measures,
as well as socioeconomic background information at the county level. Being able to identify all
health shocks that require inpatient treatment, this dataset allows us to study serious objective
health effects of smoking (bans).
6
3.1 German Hospital Admissions Census (2000-2008)
The German Hospital Admission Census (Krankenhausstatistik – Diagnosedaten, 2000-2008),
includes all hospital admissions from 2000 to 2008. By law, German hospitals are required to
submit depersonalized information on every single hospital admission. The 16 German states
collect this information and the research data centre (RDC) of the Federal Statistical Office and
Statistical Offices of the Länder (FDZ der Statistischen Ämter des Bundes und der Länder)
provides restricted data access for researchers. Germany has about 82 million inhabitants and
registers about 17 million hospital admissions per year. We observe every single hospital
admission from 2000 to 2008, i.e., a total of about 160 million hospitalizations.4 In this data,
hospital admissions include admissions requiring an overnight stay. It is not possible to
separately track emergency room admissions; the data do not contain information on the
admission route (i.e., ambulance, self-admitted). However, we observe the main diagnosis and
the number of nights that the patient spent in the hospital.
To obtain our working dataset, we aggregate the individual-level admission data on the
daily county level and normalize admissions per 100,000 or 1,000,000 population. As seen in
the Appendix, besides others, we have information on age and gender, the day of admission,
the county of residence, as well as the diagnosis in form of the 10th revision of the International
Statistical Classification of Diseases and Related Health Problems (ICD-10) code. The total
number of county-day observations in our main models is 1,429,196.5
Construction of Main Dependent Variables
Using the ICD information on the primary diagnosis, we generate the following dependent
variables: (a) Aggregating over the total admissions on a given day in a given county, we obtain
admissions representing the overall admission rate. (b) By extracting the ICD-10 codes I00-I99—
diseases of the circulatory system—we generate cardiovascular admissions. (c) Extracting the
4 This excludes military hospitals and hospitals in prisons.
5 If we had a stable set of 430 counties over all years, we would have 430*365*9 = 1,412,550 observations.
However, due to county mergers, the number of counties decreased over time and varies from 442 (2000) to 413
(2008).
7
codes I20 and I21, we generate a third dependent variable, heart attacks. (d) A fourth variable
measures hospital admissions due to asthma (ICD-10 code J45). Cardiovascular admissions and
asthma admissions are our main dependent variables because tobacco smoke mainly triggers
diseases of the circulatory system and asthma (U.S. Department of Health and Human Services,
2006, 2014). Admissions due to cardiovascular diseases are also the single most important
subgroup of admissions (16% of all admissions).
Smoking bans change people’s going out behavior to bars and restaurants (Anger et al.,
2011) and might therefore influence (excessive) drinking behavior, hospital admissions due to
alcohol intoxication, and traffic injuries. Adams and Cotti (2008), for instance, find an increase
in fatal accidents involving alcohol following smoking bans in the U.S. because smokers drive
longer distances to bars without smoking restrictions. Therefore, we also study (e) alcohol
poising (code T51) and (f) injuries (codes V01-X59). Finally, we conduct falsification exercises
and report estimates for the placebo outcomes (g) drug overdosing (ICD-10 code T40) and (h)
suicide attempts (code T14).
Summary statistics for all dependent variables are provided in the Appendix. On a given day,
we observe about 57.9 hospital admissions per 100,000 population in Germany.6 On average,
there are 9.1 cardiovascular admissions and 0.9 asthma admissions per 100,000 population.
3.2 Merging Hospital Data with Weather, Pollution and Socioeconomic Data
We merge the German Hospital Admission Census with official daily weather and pollution
data to exploit additional exogenous variation in weather and pollution conditions. This allows
us to study the effectiveness of smoking bans under specific environmental conditions.
Weather Data. The weather data are provided by the German Meteorological Service
(Deutscher Wetterdienst (DWD)). The DWD is a publicly funded federal institution and collects
information from hundreds of ambient weather stations which are distributed across Germany.
We have information on the minimum, average, and maximum temperature, as well as rainfall
6 Note that German data protection laws prohibit us from reporting minimum and maximum values.
8
and hours of sunshine from up to 1,044 monitors and the years 2000 to 2008 (see Appendix for
summary statistics). We extrapolate the point measures of the ambient weather stations into
space using inverse distance weighting. This means that we use the measures for every county
and day as the inverse distance weighted average of all ambient monitors within a radius of 60
km (37.5 miles) of the county centroid (Hanigan et al., 2006).
Pollution Data. The pollution data are provided by the GERMAN FEDERAL ENVIRONMENTAL OFFICE
(Umweltbundesamt (UBA)). The UBA is a publicly funded federal agency that collects daily
information on ambient air pollution. As for the weather data, we use data for 2000 to 2008
from up to 1,314 ambient monitors. As with our weather measures, we extrapolate the point
measures into the county space on a daily basis. The Appendix shows all summary statistics.
Socioeconomic Background Data. Because the hospital data only contain gender and age
information, we collect administrative data on the unemployment rate, the number of hospitals
in a county, as well as the number of hospital beds per 10,000 population (see Appendix). Our
empirical models control for these county-level characteristics. For example, Dehejia and
Lleras-Muny (2004) find that smoking changes with the business cycle, and Ruhm (2007)
reports a negative relationship between unemployment and deaths from coronary heart
disease.7
4. Empirical Strategy
4.1 Empirical Model
To estimate the causal effect of smoking bans on hospitalizations, we estimate several
variants of the following econometric difference-in-differences (DD) model:
(1)
where
cd
y
stands for one of our dependent variables (normalized hospital admissions) and
7 Similarly, Ruhm and Black (2002) and Ruhm (2005) report that smoking, drinking and excess weight decline in
times of economic downturns.
cd
cd
ststmwsmtcd
ZXBany
εθγνδφϕββ
++++++++=
''
10
9
varies at the daily (d) county (c) level. The binary
smt
Ban
indicator is our main variable of
interest. It indicates whether a bar/restaurant smoking ban was in effect in state s in month m
of year t. To net out persistent differences across states, we incorporate a set of state dummy
variables
s
ν
. To control for seasonal and other time shocks, we additionally include sets of year
(
t
δ
), month (
m
φ
), and week-of-year (
w
ϕ
) fixed effects. In the most saturated models employed,
we replace the state, year and month fixed effects with month-year (
tm
δφ
×
) and state-year (
ts
δν
×
) fixed effects. Note that one could easily rewrite the variable of interest in this DD
model,
smt
SmokingBan
, as an interaction term between binary time-invariant state indicators
and time variables that indicate when the ban became effective in each state.
cd
Z
stands for a vector of controls that we generated from the individual-level hospital
admission data; we aggregated those controls at the daily county level. For example, in the
German Hospital Admission Census, we have information on patients’ gender and age (reported
in 17 different age groups).
Finally, we include a set of annual county-level covariates Xst that control for differences in
the unemployment rate, the number of hospitals in the county, and the number of hospital beds
per 10,000 population (see Appendix). The standard errors are clustered at the state level
(Bertrand et al., 2004).
4.2. Identification
It is plausible to assume that the implementation of the German smoking bans across the 16
German states over time was exogenous to individual smoking behavior. Particularly helpful
features of our setting are (i) that all German states eventually introduced smoking bans, (ii)
that it all happened between August 2007 and August 2008, i.e., within only 13 months, and (iii)
that we can exploit temporal variation in the implementation that ranges across all seasons of a
year. Finally, Anger et al., 2011 have shown (iv) that the timing of the implementation across
states shows no systematic relationship with a rich set of pre-ban state, among them the
percentage of smokers in a state.
10
Our preferred specification in equation (1) does not only control for socio-demographic
characteristics and the health care infrastructure at the county level, but also nets out week
and month-year fixed effects and controls for persistent differences across states by including
state-year effects. Such a rich specification does not leave much space for unobservables that
could affect changes in admission rates which could be systematically correlated with the
introduction of a smoking ban, but were not caused by it.
A general identification issue—but common to virtually all smoking ban papers (Adda and
Cornaglia, 2006, 2010 being notable exceptions)—is that individual-level exposure to tobacco
smoke is unknown. This means that smoking ban papers typically estimate a reduced-form
effect of a smoking ban on the outcome variables of interest. Even if actual individual-level
cigarette consumption could be measured without error, it is unclear whether (and if so, by
how much) the actual consumption intensity changed (cf. Adda and Cornaglia, 2006).
Thus, the smoking ban-related change in individual-level exposure to environmental
tobacco smoke—for smokers and non-smokers—is determined by (a) the population share of
smokers and (their smoking intensity), (b) the (cultural) habits of smokers in terms of where
they preferably smoke, (c) the details of the law banning smoking in certain locations, as well as
(d) the specific implementation and enforcement of the law.
With respect to (a) to (d) above, one can summarize that (a) in the German population, the
share of smokers is roughly one third across all cohorts, which is considerably larger than in the
US (Cutler and Glaeser, 2009). (b) At least before the smoking ban, smoking was still to a large
degree a social activity and not necessarily associated with a stigma as in the US. (c+d) The
smoking bans mostly apply to indoor smoking in pubs and restaurants. Almost no exceptions
exist for restaurants. However, as discussed in Section 2, in almost all states during the
observation period, it was still possible to smoke in dive bars that self-declared as “smoker
pubs.” On the other hand, the majority of pubs, particularly popular and touristic pubs, entirely
banned indoor smoking.
11
5. Results
5.1 Main Estimates
Table 2 shows the results from regression models as in equation (1). Every column
represents one model. The dependent variable in the first two columns is all-cause admissions
per 100,000 population. Columns (3) and (4) use cardiovascular admissions per 100,000
population, columns (5) and (6) heart attacks per 100,000 population, and the final two
columns report estimates for the dependent variable asthma admissions per 100,000
population. All models in the odd-numbered columns control for state, week, month, and year
effects. All models in the even-numbered columns control for week, month-year, and state-year
effects and control for socio-demographics.
[Insert Table 2 about here]
We learn the following from Table 2: First, following the implementation of a smoking ban,
the all-cause admission rate decreased significantly by about 10 admissions per 1 million
population, or 1.6% (column (2)). According to our preferred specification in column (4), the
cardiovascular admission rate decreased significantly by 1.9 per 1 million population or by 2.1%.
This is a small, but highly significant, effect. For the entire German population with its 82 million
residents, 1.9 fewer people admitted due to heart problems per 1 million population translates
into 156 fewer hospital admissions per day or about 56,867 fewer cardiovascular admissions
per year. Applying the average health care costs of one hospital day of about €500, just these
avoided cardiovascular admissions would yield a resource savings estimate worth €78 thousand
per day (Gesundheitsberichterstattung des Bundes, 2012). For a comprehensive welfare loss
estimate, one would need to add the patients’ quality of life lost during hospital stays and the
welfare loss of lost working days.
Could it be that the reduction in cardiovascular admissions is mainly driven by fewer
hospitalization of employees in establishments that are newly smoke free (e.g., bartenders and
waitresses)? According to the Federal Statistical Office, around 1.1 million worked in cafes,
12
pubs, and bars in 2016 (Destatis, 2017). A decline by 1.9 cardiovascular hospital admissions per
1 million population per day among employees in these smoke-free venues would result in
around 730 fewer admissions per year. This simple back-of-the-envelope calculation suggests
that the reduction in cardiovascular admissions is very likely due to lower population exposure
rather than a decline only among employees in venues directly affected by the smoke free laws.
Because our data contain information on whether people died in the hospital after having
been admitted, we experiment with an additional outcome death after cardiovascular
admission but do not find significant effects. The reason is likely that only 0.45 people per
100,000 population die after being hospitalized due to heart problems. This number only
represents five percent of all cardiovascular admissions. Even with our high-frequency register
data counting 17 million admissions per year, we operate in an underpowered setting. This
illustrates a general structural issue for researchers trying to identify specific objective
population health effects of anti-tobacco policies. If the death after cardiovascular admission
effect was symmetric to the general cardiovascular admission effect, one would need to
identify a 0.007 point estimate which is hardly possible even with register data.
Our next two models with the outcome heart attack admission rate (a subset of the overall
cardiovascular admission rate) illustrates this point as well. Although we find consistently
reductions in the heart attack rate in our specifications, only the estimate of the
“parsimonious” DD model, with only state, week, month and year fixed effects in column (5) of
Table 2 yields a marginally significant -0.05 point estimate, translating into a reduction in the
heart attack rate of -3.1%. Our preferred saturated model estimate in column (6) is just -0.02
and not statistically significant any more. Relative to the mean of 1.48 heart attack admissions
per 100,000 population and day, the estimate translates into a decrease of only 0.1%. This
illustrates why many studies that intend to identify health effect estimates following smoking
bans cannot provide precise estimates: even in a densely populated country with a very good
health care infrastructure, short distances to the next hospital as well as a relatively high
smoking prevalence, heart attacks are still a relatively rare event and small changes in the rates
are very hard to identify, even with register data. It is also important to point out that the
13
reduction in heart attack admissions in Germany is considerably smaller than what has been
found in the study by Pell et al. (2008) following smoke-free legislation in Scotland. The authors
report a reduction in acute coronary syndromes by 17%. However, methodologically Pell et al.
(2008) compare hospitalizations before and after the smoke-free legislation was implemented
in Scotland. Consequently, their findings are not directly comparable to our differences-in-
differences model, which considers rich sets of time and regional fixed effects which net out
seasonal as well spatial confounding factors.
The last two columns of Table 2 present the results for asthma admissions. In our preferred
specification in column (8), we find a statistically significant decrease by around 0.06 fewer
asthma admissions per 100,000 population and day. This is a considerable reduction as it
implies a decline of nearly seven percent at the mean.
In unreported regressions (available upon request), we also estimate the smoking ban
effects on cardiovascular admissions and asthma admissions separately by age group.
Specifically, we ran our preferred model in columns (4) and column (8) of Table 2 for nine
different age groups.8 Furthermore, we explored potential heterogeneity in the treatment
effect by gender. Overall, we did not find any significant differential impacts of public smoking
bans on cardiovascular or asthma admissions by age group or gender.
[Insert Table 3 about here]
Table 3 presents the results for alcohol poisoning, injuries, and for the placebo outcomes
drug overdosing and suicide attempts. Smoking bans might alter where people smoke and
drink. They may drive longer distances by car to venues without smoking restrictions, increasing
the risk of accidents and fatalities (Adams and Cotti, 2008). However, our results in Table 3 do
not suggest a significant decline of alcohol poisonings or an increase in the number of injuries as
a result of more car accidents. However, keep in mind that low means per 1 million population
and that we may not have enough power to identify small changes, e.g., the point estimates for
alcohol poisonings are consistently negative and between 2 and 7% of the mean.
8 Ages 0-10, 11-20, 21-30, 31-40, 41-50, 51-60, 61-65, 66-75, and 75+.
14
Finally, columns (5) to (8) show the results for the falsification outcomes drug overdosing
and suicide attempts. The estimates do not show consistent signs, are relatively close to zero
and not statistically significantly different at conventional levels.
5.2 Robustness Checks
Table 4 provides a series of robustness checks for cardiovascular admissions. The first
column is our preferred estimate from column (4) of Table 2 and serves as comparison. Column
(2) clusters standard errors at the county level, which barely changes the standard errors.
Column (3) includes 467 county fixed effects. As a result, the size of the coefficient decreases
to -0.08 but remains significant. Excluding the three early adopting states that implemented the
ban in the second half of 2007 increases the estimate only slightly (column 4). In column (5), we
run a placebo test and assume that the ban was implemented exactly one calendar year earlier
than it actually was implemented. As seen, the point estimate is very small, positive, and not
statistically significant. Finally, the regression in the last column in Table 4 excludes all counties
that border Germany’s neighboring countries Switzerland, Austria, Czech Republic, Poland, the
Netherlands, Belgium, Denmark, Luxembourg or France. The point estimate of -0.183 is very
close to the standard estimate suggesting that potential cross-border effects are unlikely to play
a major role.
[Insert Table 4 about here]
5.3 Event Studies
Next, we present event study graphs for our main outcome variables cardiovascular
admissions and asthma admissions. We estimate our preferred saturated version of equation
(1) with state-year, week, and month-year fixed effects, but interact Bansmt with an indicator
that counts the 12 months before and after the smoking ban implementation. Figures 2 and 3
plot the coefficient estimates of this indicator along with their 95% confidence intervals. The
event studies illustrate how changes in admission rates evolve in the months before and after
the smoking bans have been implemented.
[Insert Figure 2 about here]
15
Figure 2 displays the event study for cardiovascular admissions. The pre-ban estimates are
relatively flat and fluctuate around the zero line (8 positive, 3 negative point estimates); except
for one, all confidence intervals include the zero line. In contrast, in post-ban months, we
observe a smooth slight decrease in cardiovascular admission rates. Although most monthly
point estimates are imprecisely estimated, all post-reform point estimates are negative. This
pattern is suggestive of a persistent, albeit hard to quantify negative effect of smoking bans on
cardiovascular admissions. In terms of relative size, the post-ban estimates lie between 0.7%
and 3.8% of the mean with most estimates lying between -1% and -2%, which explains why it is
hard to estimates these coefficients precisely.
[Insert Figure 3 about here]
Figure 3 shows the event study for asthma admissions. Similar to the results for
cardiovascular admissions, the pre-ban point estimates fluctuate around zero (6 positive and 5
negative) and only one is statistically significant. In contrast, almost all post-ban estimates are
negative (but imprecisely estimated). Although imprecisely estimated, Figure 3 suggests a
short-term reduction in asthma admissions particularly in the first months after the ban. When
estimating the event study using the basic model in equation (1) with separate week, state,
month and year fixed effects, this suggestive evidence of a short-term effect is reinforced: In
this specification, the four post-ban months estimates t=0 to t=3 all carry relatively large
negative signs with effect sizes of almost 10% and all are statistically different from zero.
However, for asthma admissions, there appears to be a rebound to the zero line after
significant reductions in the first post-ban months. Further research on the longer-term health
effects of smoking bans would be highly warranted.
5.4 Effectiveness of Bans by Weather Conditions
We now exploit exogenous variation in our weather data and use continuous measures of
rainfall quantities, hours of sunshine, and temperature to stratify the estimates by weather
conditions. Methodologically, we interact Bansmt with the weather variable of interest (and its
16
square), and also add the weather variable in levels and squares to the model. Table 5 reports
our effect heterogeneity estimates by weather conditions.
[Insert Table 5 about here]
Whereas we do not find evidence that smoking ban laws are more or less effective with
respect to the average temperature, there is evidence that for every additional hour of sunshine
on a given day, admission rates additionally decrease by 0.068 (-0.7%) per 100,000 population.
Similarly, with each additional hour of rain on a given day, cardiovascular admission rates
increase by around 0.02 (+0.2%) per 100,000 population (ignoring the squared coefficient). This
decrease (increase) in admissions on sunny (rainy) days could be due to several factors. For
example, one could hypothesize that people spend more time outside when the sun shines.
Because smoking bans only apply to places outside individuals’ homes, this could explain the
effect.
Note that the sunshine effect cannot be explained by the correlation between hours of
sunshine and temperature—i.e., this is a true sunshine effect—because the estimate remains
unchanged when we add temperature to the model. However, when creating an indicator of
hot days with maximum temperatures of more than 30°C (86°F) instead of using the continuous
temperature indicator (not shown), we also find a reinforcement of the health promotion
effect, in line with the arguments above.
5.5 Effectiveness of Bans by Ambient Air Pollution
Table 6 analyzes whether high pollution levels reinforce or attenuate the smoking ban
health effects. Again we add the pollutant of interest and its square to our baseline model in
levels and in interaction with the Bansmt indicator. Like hot days (cf. Karlsson and Ziebarth 2017;
Deschênes and Moretti 2009), elevated ambient air pollution potentially induces stress for the
human body and can trigger negative health effects. Tobacco smoke could reinforce this effect.
Thus one could hypothesize that the health promoting effect of smoking bans should be
reinforced under adverse environmental conditions.
17
[Insert Table 6 about here]
The findings for O3 and NO2 are in line with this conjecture. First, the highly significant Ban*
O3 indicator in the first two columns implies that, on average, an additional 5 cardiovascular
admissions per 10 million population are avoided by the German smoking bans when ozone
levels are 10μ/m3 higher (relative to a mean O3 concentration of 46μ/m3, see Appendix).
Second, the sign of the estimates for NO2 also suggest that the health promotion effects of
smoke-free legislation are reinforced under adverse environmental conditions. However, even
though the point estimates are relatively large, they are not statistically significant. In an
alternative specification (available upon request), we generate pollution alert indicator
variables that are one when the NO2 concentration exceeds EU alert threshold levels. Replacing
the continuous NO2 measures with this dummy yields a highly significant -0.0088*** interaction
term, suggestion that health effects of smoking bans are larger when pollution levels are
critical. Finally, the main interaction terms for PM10 are not statistically significant.
6. Conclusion
Reducing smoking prevalence and exposure to environmental tobacco smoke remains a key
public health priority of international health organizations and governments worldwide. Public
smoking bans have become particularly popular in many countries in the last decade. Evidence
on the effects of such bans on health outcomes, however, is still inconclusive. A lack of natural
tempo-spatial variation and suitable treatment-control group designs—in combination with
limited data availability to achieve enough statistical power and to control for potential
confounders—additionally complicates identification.
This paper studies the short-run effects of the staggered implementation of state-level
public smoking bans on hospital admissions in Germany. We contribute to the literature (i) by
exploiting both time and regional variation in smoke free legislation in a high smoking
prevalence country, (ii) by focusing on objective health measures and (iii) by using high-
frequency administrative data in combination with auxiliary weather, pollution and socio-
economic county data. This setting allows us to control more comprehensively than many
18
existing studies for potentially important confounders. For example, in addition to being able to
net out important seasonal confounders by employing rich sets of week, month-year, and state-
year fixed effects, we also study the role of environmental pollution and weather effects.
We find that daily cardiovascular admissions decreased by about 1.9 per 1 million
population (or by 2.1%) after the introduction of state-level smoking bans in Germany between
2007 and 2008. This translates into 57 thousand fewer cardiovascular admissions per year for
the whole of Germany. Our findings hence suggests that sizable health gains can be achieved
from such an anti-smoking policy—even if the law allows for exemptions and enforcement is
imperfect.
19
Literature
Adams, S.; Cotti, C. (2008): “Drunk driving after the passage of smoking bans in bars”, Journal of
Public Economics, 92(5), 1288-1305.
Adda, J.; Cornaglia, F. (2006): “Taxes, cigarette consumption, and smoking intensity”, American
Economic Review, 96(4): 1013-1028.
Adda, J.; Cornaglia, F. (2010): “The effect of bans and taxes on passive smoking”, American
Economic Journal: Applied Economics, 2(1): 1-32.
Anger, S.; Kvasnicka, M.; Siedler, T. (2011): “One last puff? Public smoking bans and smoking
behavior”, Journal of Health Economics, 30(3): 591-601.
Avery, R.; Kenkel, D.; Lillard, D. R.; Mathios, A. (2007): “Private profits and public health: Does
advertising of smoking cessation products encourage smokers to quit?” Journal of Political
Economy, 115(3): 447-481.
Baxmann, M.; Eckoldt, M. (2007): “Typisch deutsch? Nicht-Rauchen in Deutschland?“,
Deutschlandfunk Kultur. Beitrag vom 23.02.2017.
Bertrand, M.; Duflo, E.; Mullainathan, S. (2004): “How much should we trust differences-in-
differences estimates?” Quarterly Journal of Economics, 119(1): 249-275.
Bitler, M. P.; Christopher S. C.; Zavodny, M. (2010): “Effects of venue-specific state clean indoor
air laws on smoking-related outcomes“, Health Economics, 19(12): 1425–40.
Brüderl, J.; Ludwig, V. (2011): “Does a smoking ban reduce smoking? Evidence from Germany”,
Schmollers Jahrbuch, 131(2): 419-429.
Callison, K.; Kaestner, R. (2014): “Do higher tobacco taxes reduce adult smoking? New evidence
of the effect of recent cigarette tax increases on adult smoking”, Economic Inquiry, 52(1): 155-
172.
Carpenter, C.; Cook, P. J. (2008): “Cigarette taxes and youth smoking: New evidence from
national, state, and local Youth Risk Behavior Surveys”, Journal of Health Economics, 27(2): 287-
299.
Carpenter, C.; Postolek, S.; Warman, C. (2011): “Public-place smoking laws and exposure to
environmental tobacco smoke (ETS)”, American Economic Journal: Economic Policy, 3(3): 35-61.
Courtemanche, C. (2009): “Rising cigarette prices and rising obesity: Coincidence or unintended
consequence?” Journal of Health Economics, 28(4): 781-798.
Cutler, D. M.; Glaeser, E. L. (2009): “Why do Europeans smoke more than Americans?” NBER
Chapters, in: Developments in the Economics of Aging: 255-282.
20
DeCicca, P.; Kenkel, D.; Liu, F. (2013): “Excise tax avoidance: The case of state cigarette taxes”,
Journal of Health Economics, 32(6): 1130–1141.
DeCicca, P.; Kenkel, D.; Mathios, A. (2002): “Putting out the fires: Will higher taxes reduce the
onset of youth smoking?” Journal of Political Economy, 110(1): 144–169.
DeCicca, P.; Kenkel, D.; Mathios, A. (2008): “Cigarette taxes and the transition from youth to
adult smoking: Smoking initiation, cessation, and participation”, Journal of Health Economics,
27(4): 904–917.
Dehejia, R.; Lleras-Muny, A. (2004): “Booms, busts, and babies’ health”, Quarterly Journal of
Economics, 119(3): 1091-1130.
DEHOGA (2008): Nichtraucherschutzgesetze in den Bundesländern. Synopse zu den
Landesgesetzen, Deutscher Hotel- und Gaststättenverband (DEHOGA), Berlin, Germany.
Deschênes, O.; Moretti, E. (2009): “Extreme weather events, mortality, and migration”, The
Review of Economics and Statistics, 91(4): 659-681.
Destatis (2017): “Unternehmen, Beschäftigte, Umsatz, Warenbezüge, Wareneinsatz, Rohertrag
im Gastgewerbe 1995 - 2008”, GENESIS-Online.
European Environment Agency (2012): “Air quality in Europe—2011 report”,
http://www.eea.europa.eu/publications/air-quality-in-europe-2011, last accessed on
November 24, 2012.
Evans, W. N.; Farrelly, M. C.; Montgomery, E. (1999): “Do workplace smoking bans reduce
smoking?” American Economic Review, 89(4): 728-747.
Friedman, A. S. (2015): “How does electronic cigarette access affect adolescent
smoking?”Journal of Health Economics, 44: 300-308.
Gesundheitsberichterstattung des Bundes (2012), www.gbe-bund.de, last accessed on June 20,
2012.
Goolsbee, A.; Lovenheim, M.; Slemrod, J. (2010): “Playing with fire: Cigarettes, taxes, and
competition from the internet”, American Economic Journal: Economic Policy, 2(1): 131–154.
Göhlmann, S.; Schmidt, C. M. (2008):“Smoking in Germany: Stylized Facts, behavioral models,
and health policy”, Ruhr Economic Papers, #64.
Gruber, J. (2001): “Tobacco at the crossroads: The past and future of smoking regulation in the
United States”, Journal of Economic Perspectives, 15(2): 193-212.
Gruber, J.; Anindya, S.; Stabile, M. (2003): “Estimating price elasticities when there is smuggling:
the sensitivity of smoking to price in Canada”, Journal of Health Economics, 22(5): 821-842.
21
Gruber, J.; Frakes, M. (2006): “Does falling smoking lead to rising obesity?” Journal of Health
Economics, 25(2): 183-197.
Gruber, J.; Koszegi, B. (2004): “Tax incidence when individuals are time-inconsistent: the case of
cigarette excise taxes”, Journal of Public Economics, 88(9-10): 1959-1987.
Gruber, J. H; Mullainathan, S. (2005): “Do cigarette taxes make smokers happier”, The B.E.
Journal of Economic Analysis & Policy, 5(1): 1-45.
Hanigan, I.; Hall, G.; Dear, K. B.G. (2006): “A comparison of methods for calculating population
exposure estimates of daily weather for health research”, International Journal of Health
Geographics, 5(38): 1-6.
Hansen, B.; Sabia, J.; Rees, D. (2017): “Have cigarette taxes lost their bite: New estimates of the
relationship between cigarette taxes and youth smoking”, American Journal of Health
Economics, 3(1): 60-75.
Jones, A. M.; Laporte, A.; Rice, N.; Zucchelli, E. (2015): “Do public smoking bans have an impact
on active smoking? Evidence from the UK”, Health Economics, 24(2): 175-192.
Joossens L.; Raw M. (2011): The Tobacco Control Scale 2010 in Europe. Association of European
Cancer Leagues; Brussels, http://www.europeancancerleagues.org/tobacco-control/ectoh/219-
tobacco-control-scale-2010.html, retrieved September 15, 2017.
Karlsson, M.; Ziebarth, N.R. (2017): “Population health effects and health-related costs of
health events: Comprehensive evidence from Germany”, http://www.nicolasziebarth.com,
retrieved August 14, 2017.
Kuehnle, D.; Wunder, C. (2017): “The effects of smoking bans on self-assessed health: Evidence
from Germany”, Health Economics, 26(3): 321-337.
Kvasnicka, M.; Tauchmann, H. (2012): “Much ado about nothing? Smoking bans and Germany's
hospitality industry”, Applied Economics, 44(35): 4539-4551.
Lillard, D. R.; Molloy, E.; Sfekas, A. (2013): “Smoking initiation and the iron law of demand”,
Journal of Health Economics, 32(1): 114–127.
LNRSchG (2007): “Landesnichtraucherschutzgesetz (LNRSchG)” vom 25. Juli 2007. Landesrecht
Baden-Württemberg.
Lovenheim, M. (2008): “How far to the border? The extent and impact of cross-border casual
cigarette smuggling”, National Tax Journal, 61(1): 7–33.
Mons, U.; Müezzinler,A.; Gellert, C.; Schöttker, B.; Abnet, C. C.; Bobak, M., de Groot, L. u. a.
(2015): “Impact of smoking and smoking cessation on cardiovascular events and mortality
22
among older adults: meta-analysis of individual participant data from prospective cohort
studies of the CHANCES consortium“, British Medical Journal, 350(April): 1-12.
NRauchSchG SH (2007): „Gesetz zum Schutz vor den Gefahren des Passivrauchens“. 10.
Dezember 2007. Gesetz- und Verordnungsblatt 2007, 20, 473-560. Schleswig-Holsteinischer
Landtag.
Odermatt, R.; Stutzer, A. (2015): “Smoking bans, cigarette prices and life satisfaction”, Journal
of Health Economics, 44: 176-194.
Pell, J. P.; Haw, S.; Cobbe, S.; Newby, D. E..; Pell, A. C. H.; Fischbacher, C.; McConnachie, A.;
Pringle, S.; Murdoch, D.; Dunn, F.; Oldroyd, K.; MacIntyre, P.; O’Rourke, B.; Borland, W. (2008):
“Smoke-free legislation and hospitalizations for acute coronary syndrome”, The New England
Journal of Medicine, 359(5): 482-491.
Rozema, K.; Ziebarth, N. R. (2017): “Taxing consumption and the take-up of public assistance:
The case of cigarette taxes and food stamps," Journal of Law and Economics, 60(1): 1-27.
Ruhm, C. J. (2005): “Healthy living in hard times”, Journal of Health Economics, 24(2): 341-363.
Ruhm, C. J. (2007): “A healthy economy can break your heart”, Demography, 44(4): 829-848.
Ruhm, C. J.; Black, W. E. (2002): “Does drinking really decrease in bad times”, Journal of Health
Economics, 21(4): 659-678.
Shetty, K. D.; DeLeire, T.; White, C.; Bhattacharya, J. (2011): “Changes in U.S. hospitalization and
mortality rates following smoking bans”, Journal of Policy Analysis and Management, 30(1): 6-
28.
U.S. Department of Health and Human Services (2006). The health consequences of involuntary
exposure to tobacco smoke: a report of the surgeon general. Atlanta, GA: Centers for Disease
Control and Prevention, Coordinating Center for Health Promotion, National Center for Chronic
Disease Prevention and Health Promotion, Office on Smoking and Health,
https://www.surgeongeneral.gov/library/reports/secondhandsmoke/fullreport.pdf, retrieved
September 15, 2017.
U.S. Department of Health and Human Services (2014). The health consequences of smoking: 50
years of progress. A report of the surgeon general. Atlanta, GA, Centers for Disease Control and
Prevention, National Center for Chronic Disease Prevention and Health Promotion, Office on
Smoking and Health. https://www.surgeongeneral.gov/library/reports/50-years-of-
progress/index.html, retrieved September 15, 2017.
Wehby, G. L.; Courtemanche, C. J. (2012): “The heterogeneity of the cigarette price effect on
body mass index”, Journal of Health Economics, 31(5): 719-729.
23
World Health Organization (2015a): WHO global report on trends in prevalence of tobacco
smoking 2015, Geneva, http://apps.who.int/iris/handle/10665/156262, retrieved September
15, 2017.
World Health Organization (2015b): WHO report on the global tobacco epidemic, 2015: Raising
taxes on tobacco, Geneva, http://www.who.int/tobacco/global_report/2015/report/en/,
retrieved September 15, 2017.
24
Figures and Tables
Figure 1: Dates of Enforcement of State Smoking Bans in Germany
Notes: For further information, see notes to Table 1.
25
Figure 2: Event Study—Smoking Ban Effect on Cardiovascular Admissions per 100K pop.
Note: German Hospital Admission Census; Y-axis displays the mean change in ppt and x-
axis the months before and after the smoking ban implementation. The solid black line
represents the point estimates and the dotted line 95% confidence intervals. Regression
is based on the saturated version of equation (1) and includes week, state-year and
month-year fixed effects. Cardiovascular admissions are generated by extracting the
ICD-10 codes I00-I99.
-2
-1,5
-1
-0,5
0
0,5
1
1,5
2
Month t-12
Month t-11
Month t-10
Month t-9
Month t-8
Month t-7
Month t-6
Month t-5
Month t-4
Month t-3
Month t-2
Month t-1
Month t=0
Month t=1
Month t=2
Month t=3
Month t=4
Month t=5
Month t=6
Month t=7
Month t=8
Month t=9
Month t=10
Month t=11
Month t=12
26
Figure 3: Event Study—Smoking Ban Effect on Asthma Admissions
Note: German Hospital Admission Census; Y-axis displays the mean change in ppt and x-
axis the months before and after the smoking ban implementation. The solid black line
represents the point estimates and the dotted line 95% confidence intervals. Regression
is based on the saturated version of equation (1) and includes week, state-year and
month-year fixed effects. Asthma admissions are defined according to ICD-10 code J45.
-0,5
-0,4
-0,3
-0,2
-0,1
0
0,1
0,2
0,3
0,4
0,5
Month t-12
Month t-11
Month t-10
Month t-9
Month t-8
Month t-7
Month t-6
Month t-5
Month t-4
Month t-3
Month t-2
Month t-1
Month t=0
Month t=1
Month t=2
Month t=3
Month t=4
Month t=5
Month t=6
Month t=7
Month t=8
Month t=9
Month t=10
Month t=11
Month t=12
27
Table 1:
Dates of Enforcement of State Smoking Bans in Germany
Federal State Enforcement of State Smoking Ban
Baden-Wuerttemberg
August 2007
Bavaria
January 2008
Berlin
July 2008
Brandenburg
July 2008
Bremen
July 2008
Hamburg
January 2008
Hesse
October 2007
Lower Saxony
November 2007
Mecklenburg-West Pomerania
August 2008
North Rhine-Westphalia
July 2008
Rhineland-Palatinate
February 2008
Saarland
June 2008
Saxony
February 2008
Saxony-Anhalt
July 2008
Schleswig-Holstein
January 2008
Thuringia
July 2008
Notes: All smoking bans were enforced at the start of the month with the
exception of Rhineland-
Palatinate, which introduced the smoking ban on
February 15, 2008. Information on individual states was
compiled from original
law texts and from a survey of state-
level smoking ban legislation by the
German Hotels and Restaurants Federation (DEHOGA, 2008).
28
Table 2: Main Estimates
Bar and Restaurant Smoking Ban Effect on Hospital Admission Rates
All-Cause Admissions
Cardiovascular
Admissions
a Heart Attacksb Asthma Admissionsc
(per 100,000 pop.) (per 100,000 pop.) (per 100,000 pop.) (per 100,000 pop.)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Ban
-1.0499**
-0.9565***
-0.0954
-0.1870***
-0.0453*
-0.0190
-0.0078
-0.0574***
(0.4850)
(0.2248)
(0.1137)
(0.0434)
(0.0258)
(0.0203)
(0.0460)
(0.0183)
Mean of outcome variable
57.99
57.99
9.11
9.11
1.48
1.48
0.88
0.88
Change in %
-1.8
-1.6
-1.0
-2.1
-3.1
-1.3
-0.9
-6.5
State Fixed Effects
Yes
No
Yes
No
Yes
No
Yes
No
Week Fixed Effects
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Month Fixed Effects
Yes
No
Yes
No
Yes
No
Yes
No
Year Fixed Effects
Yes
No
Yes
No
Yes
No
Yes
No
Month-Year Fixed Effects
No
Yes
No
Yes
No
Yes
No
Yes
State-Year Fixed Effects
No
Yes
No
Yes
No
Yes
No
Yes
Socio-demographic daily controls
No
Yes
No
Yes
No
Yes
No
Yes
Annual county-level controls
No
Yes
No
Yes
No
Yes
No
Yes
R²
0.1132
0.4132
0.1150
0.3834
0.0684
0.1532
0.0085
0.0123
Notes: a ICD-10 codes I00-I99. b ICD-10 code I20 and I21. c ICD-10 code J45. * p<0.1, ** p<0.05, *** p<0.01; standard errors are in parentheses and
clustered at the state level. All models are variants of equation (1). All models have 1,429,196 county-day observations. The column headers
indicate the dependent variable. The even and odd-numbered columns only differ by the sets of covariates included. Ban is the main variable of
interest and varies at the month-year-state level. This dummy is one for states and months with a smoking ban in bars and restaurant and zero else.
Sources: RDC of the Federal Statistical Office and Statistical Offices of the Länder, [Krankenhausstatistik – Diagnosedaten 2000-2008], own
calculations. The “German Hospital Admission Census” is administrative county-level data, daily weather data from the German Meteorological
Service, daily pollution data from the German Federal Environmental Office, unbalanced panel at daily county level; see Section 3 for more details.
29
Table 3: Main Estimates II
Bar and Restaurant Smoking Ban Effect on Selected Further Hospital Admissions Rates
Alcohol Poisoning
a
Injuries
b
Drug Overdosing
c
Suicide Attempts
d
(per 1 million pop.)
(per 1 million pop.)
(per 1 million pop.)
(per 1 million pop.)
(1) (2) (3) (4) (5) (6) (7) (8)
Ban
-0.0127
-0.0036
1.0493**
-0.3787
0.0045
-0.0016
-0.0126
-0.0039
(0.0152)
(0.0145)
(0.4527)
(0.3209)
(0.0047)
(0.0055)
(0.0207)
(0.0161)
Mean of outcome variable
0.18
0.18
58.94
58.94
0.09
0.09
0.31
0.31
Change in %
-7.0
-2.0
1.8
-0.6
5.0
-1.8
-4.1
-1.3
State Fixed Effects
Yes
No
Yes
No
Yes
No
Yes
No
Week Fixed Effects
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Month Fixed Effects
Yes
No
Yes
No
Yes
No
Yes
No
Year Fixed Effects
Yes
No
Yes
No
Yes
No
Yes
No
Month-Year Fixed Effects
No
Yes
No
Yes
No
Yes
No
Yes
State-Year Fixed Effects
No
Yes
No
Yes
No
Yes
No
Yes
Socio-demographic daily controls
No
Yes
No
Yes
No
Yes
No
Yes
Annual county-level controls
No
Yes
No
Yes
No
Yes
No
Yes
R²
0.0042
0.0093
0.0835
0.1477
0.0008
0.0017
0.0196
0.022
Notes: a ICD-10 code T51. b ICD-10 codes V01-X59. c ICD-10 code T40. d ICD-10 code T14. * p<0.1, ** p<0.05, *** p<0.01; standard errors are in
parentheses and clustered at the state level. All models are variants of equation (1). All models have 1,429,196 county-day observations. The
column headers indicate the dependent variable used. The even and odd-numbered columns only differ by the sets of covariates included. Ban is
the main variable of interest and varies at the month-year-state level. This dummy is one for states and months with a smoking ban in bars and
restaurant and zero else.
Sources: RDC of the Federal Statistical Office and Statistical Offices of the Länder, [Krankenhausstatistik – Diagnosedaten 2000-2008], own
calculations. The “German Hospital Admission Census” is administrative county-level data, daily weather data from the German Meteorological
Service, daily pollution data from the German Federal Environmental Office, unbalanced panel at daily county level; see Section 3 and Appendix
for more details.
30
Table 4: Robustness Checks for the Outcome Variable Cardiovascular Admissions
Standard
Estimate
Cluster at
County Level
County-Level
FE
No Early
Implementing
States
Placebo
Treatment One
Year Before
No Border
Counties
(1)
(2)
(3)
(4)
(5)
(6)
Ban
-0.1870***
-0.1870***
-0.0765**
-0.2135**
0.0087
-0.1831***
(0.0434)
(0.0442)
(0.0353)
(0.0967)
(0.0815)
(0.0489)
State-Year Fixed Effects
Yes
Yes
Yes
Yes
Yes
Yes
Week Fixed Effects
Yes
Yes
Yes
Yes
Yes
Yes
Month-Year Fixed Effects
Yes
Yes
Yes
Yes
Yes
Yes
Daily and annual county controls
Yes
Yes
Yes
Yes
Yes
Yes
R²
0.3834
0.3828
0.3460
0.3666
0.3804
N
1,429,196
1,429,196
1,429,196
1,048,268
1,429,196
1,249,505
Notes: * p<0.1, ** p<0.05, *** p<0.01; standard errors are in parentheses and clustered at the state level (except for column (2)). All models are
models as in equation (1) but using state-year and month-year instead of separate state, year, and month fixed effects. Ban varies at the month-
year-state level and is one for states and months with a smoking ban in bars and restaurant and zero else. Column (1) equals column (4) in Table
2. Column (2) clusters the standard estimate at the county instead of the state level. Column (3) uses 467 county-level fixed effects. Column (4)
excludes the three states that implemented the ban in 2007 (Baden-
Wurttemberg, Lower Saxony, Hesse). Column (5) is a placebo check and
assumes that the bans were implemented exactly one year earlier than they were actually implemented. Column (6) excludes the 67 counties
that border a the neighboring country of Germany.
Sources: RDC of the Federal Statistical Office and Statistical Offices of the Länder, [Krankenhausstatistik – Diagnosedaten 2000-2008], own
calculations. The “German Hospital Admission Census” is administrative county-level data, daily weather data from the German Meteorological
Service, daily pollution data from the German Federal Environmental Office, unbalanced panel at daily county level; see Section 3 and Appendix
for more details.
31
Table 5: Effect Heterogeneity, Weather Conditions
(1)
(2)
(3)
(4)
(5)
(6)
Ban*Rain
0.0205**
0.0199**
(0.0093)
(0.0092)
Ban*Rain2
-0.0011**
-0.0010**
(0.0005)
(0.0004)
Ban*Sunshine
-0.0684**
-0.0685**
(0.0247)
(0.0243)
Ban*Sunshine2
0.0028
0.0029
(0.0023)
(0.0023)
Ban*Temperature
0.0035
0.0012
(0.0121)
(0.0114)
Ban*Temperature2
0.0000
0.0001
(0.0004)
(0.0004)
State-Year Fixed Effects
Yes
Yes
Yes
Yes
Yes
Yes
Week Fixed Effects
Yes
Yes
Yes
Yes
Yes
Yes
Month-Year Fixed Effects
Yes
Yes
Yes
Yes
Yes
Yes
Daily and annual county controls
Yes
Yes
Yes
Yes
Yes
Yes
Weather and pollution controls
No
Yes
No
Yes
No
Yes
R²
0.3828
0.3831
0.3829
0.3832
0.3831
0.3833
N
1,429,196
1,429,196
1,429,196
1,429,196
1,429,196
1,429,196
Notes: * p<0.1, ** p<0.05, *** p<0.01; standard errors are in parentheses and clustered at the state level. Each cell represents one model
similar to Column (4) of Table 2. The dependent variable is always cardiovascular admissions per 100,000 population. The only difference
to Column (4) of Table 2 is that we add interaction terms of the Ban indicator and our variable of interest (in levels and in squares) as
indicated in the rows. In addition, we add the variable of interest (and its square) in levels. We do not report the level estimates for the
sake of saving space.
Sources: RDC of the Federal Statistical Office and Statistical Offices of the Länder, [Krankenhausstatistik – Diagnosedaten 2000-2008], own
calculations. The “German Hospital Admission Census” is administrative county-level data, daily weather data
from the German
Meteorological Service, daily pollution data from the German Federal Environmental Office, unbalanced panel at daily county level; see
Section 3 for more details.
32
Table 6: Effect Heterogeneity, Ambient Air Pollution
(1)
(2)
(3)
(4)
(5)
(6)
Ban*O3
-0.0047***
-0.0046***
(0.0011)
(0.0012)
Ban*O32
0.0000***
0.0000***
(0.0000)
(0.0000)
Ban*NO2
-0.0107
-0.0163
(0.0119)
(0.0134)
Ban*NO22
0.0000
0.0001
(0.0001)
(0.0001)
Ban*PM10
0.0308*
0.0185
(0.0172)
(0.0149)
Ban*PM102
-0.0008*
-0.0006*
(0.0004)
(0.0003)
State-Year Fixe Effects
Yes
Yes
Yes
Yes
Yes
Yes
Week Fixed Effects
Yes
Yes
Yes
Yes
Yes
Yes
Month-Year Fixed Effects
Yes
Yes
Yes
Yes
Yes
Yes
Daily and annual county controls
Yes
Yes
Yes
Yes
Yes
Yes
Weather and pollution controls
No
Yes
No
Yes
No
Yes
R²
0.3836
0.3929
0.3913
0.3941
0.3833
0.3919
Notes: * p<0.1, ** p<0.05, *** p<0.01; standard errors are in parentheses and clustered at the state level. Each cell represents
one model similar to Column (4) of Table 2. All models have 1,429,196 county-day observations. The dependent variable is
always cardiovascular admissions per 100,000 population. Each specification adds interaction terms of the Ban indicator and
the pollution variable of interest (in levels and in squares) to the model in column (4) of Table 2, as indicated by the rows. The
pollution variable of interest and its square is also added in levels. We do not report the level estimates to save space.
Sources: RDC of the Federal Statistical Office and Statistical Offices of the Länder, [Krankenhausstatistik – Diagnosedaten 2000-
2008], own calculations. The “German Hospital Admission Census” is administrative county-level data, daily weather data from
the German Meteorological Service, daily pollution data from the German Federal Environmental Office, unbalanced panel at
daily county level; see Section 3 for more details.
33
Appendix: Hospital Admission Data, Merged with Weather & Pollution Data
Mean
Std. Dev.
Min.
Max.
Obs.
Outcome Variables
All cause admissions per 100K pop.
57.987
25.709
N/A
N/A
1,429,196
Cardiovascular admissions per 100K pop.
9.112
4.922
N/A
N/A
1,429,196
Heart attack admissions per 100K pop.
1.481
1.362
N/A
N/A
1,429,196
Asthma admissions per 100K pop.
0.877
2.916
N/A
N/A
1,429,196
Alcohol poisoning per 1M pop.
0.1800
1.283
N/A
N/A
1,429,196
Injuries per 1M pop.
58.943
28.073
N/A
N/A
1,429,196
Drug overdosing per 1M pop.
0.0921
0.8743
N/A
N/A
1,429,196
Suicide Attempts per 1M pop.
0.3121
1.650
N/A
N/A
1,429,196
Daily County-Level Controls
Female
0.539
0.0714
0
1
1,429,196
Age Group 0-2 years
0.062
0.043
0
1
1,429,196
….
1,429,196
Age Group above 74 years
0.003
0.008
0
1
1,429,196
Annual County-Level Controls
Unemployment rate
10.465
5.277
1.6
29.3
1,429,196
Hospital beds per 10,000 pop.
1,208
1,586
0
24,170
1,429,196
Hospitals per county
4.828
5.472
0
79
1,429,196
Daily Weather & Pollution Controls
Rainfall
2.225
4.215
0
144.98
1,429,196
Hours of sunshine
4.625
4.237
0
16.7
1,429,196
Temperature
9.557
7.305
-19
30.6
1,429,196
O3 in μ/m3
45.98
22.04
0.86
135.79
1,429,196
NO2 in μ/m3
26.89
10.63
0.31
80.31
1,429,196
PM10 in μ/m3
24.31
11.46
2.06
64.63
1,429,196
Sources: RDC of the Federal Statistical Office and Statistical Offices of the Länder, [Krankenhausstatistik –
Diagnosedaten 2000-2008], own calculations. The “German Hospital Admission Census” is administrative county-
level data, daily weather data from the German Meteorological Service, daily pollution data from the German
Federal Environmental Office, unbalanced panel at daily county level; see Section 3 for more details. N/A means
not available due to legal restrictions and German data protection laws.
34
