DRUNK DRIVING AFTER THE PASSAGE OF SMOKING BANS IN BARS
Department of Economics
University of Wisconsin-Milwaukee
Department of Economics
University of South Carolina
Using geographic variation in local and state smoke-free bar laws in the US, we observe an increase
in fatal accidents involving alcohol following bans on smoking in bars that is not observed in places
without bans. Although an increased accident risk might seem surprising at first, two strands of
literature on consumer behavior suggest potential explanations—smokers driving longer distances to
a bordering jurisdiction that allows smoking in bars and smokers driving longer distances within
their jurisdiction to bars that still allow smoking, perhaps through non-compliance or outdoor
seating. We find evidence consistent with both explanations. The increased miles driven by drivers
wishing to smoke and drink offsets any reduction in driving from smokers choosing to stay home
following a ban, resulting in increased alcohol-related accidents. This result proves durable, as we
subject it to an extensive battery of robustness checks.
JEL Classification: H75, K42
*Author’s address is Department of Economics, Bolton Hall, University of Wisconsin-Milwaukee,
Milwaukee, WI, 53201; phone: 414-229-4212; email: email@example.com.
** Author’s address is Department of Economics, Moore School of Business, University of South
Carolina, Columbia, SC, 29208 phone: 920-203-4660; email: firstname.lastname@example.org.
We thank McKinley Blackburn, Scott Drewianka, John Heywood, Matthew McGinty, and seminar
participants at the University of Wisconsin-Milwaukee, the University of South Carolina, and
Indiana University for helpful suggestions. We are thankful for data assistance provided by Lorenzo
W. Daniels of the US National Highway Traffic Safety Administration. Any errors are ours.
Levitt and Porter (2001) show that drunk drivers impose an externality per mile driven of at
least 30 cents because of their greater likelihood of causing fatal accidents. In this paper, we identify
a source of increased alcohol-related car accidents that has previously gone unnoticed—the
prohibition of smoking in bars. The list of countries that have bans on smoking in public places
continues to grow and now spans six continents. Even within countries without bans, many states,
provinces, counties, cities, and villages have banned smoking.1 Although other aspects of smoking
bans have been studied, we are the first to investigate the effect of bans on drunk driving.
The expected effect of smoking bans on drunk driving is ambiguous. Many would suspect a
decline as smokers go to bars less often. Recent studies (Adams and Cotti 2007; Adda et al. 2007)
indeed show evidence consistent with bar patronage falling as a result of smoking bans being
implemented. A closer look reveals that an increase in alcohol-related accidents might be more
likely, however. First, coverage of smoking bans is not universal. In the US, many communities
with smoking bans border counties without bans. Bars subject to bans have reported losing
customers to nearby communities (McCormick-Jenkins 2007). Moreover, even within communities
or states with smoking bans, anecdotal evidence of rampant non-compliance by bars abound (e.g.
Widenman 2006; Falgoust 2004). Given this non-compliance, along with the fact that most bans
exempt outdoor seating and patios, it would be hardly surprising if the private cost associated with
finding one of these other numerous locations to smoke and drink falls short of the private benefit
some consumers get from continuing to be able to smoke and drink. When one further considers
that, according to the National Institute on Alcohol Abuse and Alcoholism (NIAAA), approximately
46 million adults used both alcohol and tobacco in the past year (roughly 24% of licensed drivers
1 The Americans for Non-Smokers’ Rights (www.no-smoke.org) maintains a frequently updated list
of such bans.
who are of legal drinking age),2 and approximately 6.2 million adults reported both an alcohol-use
disorder (AUD) and dependence on nicotine (NIAAA 2007), it is certainly plausible that there might
be an increase in drunk driving risk even if the average bar’s business declines. If the miles driven
by smokers to avoid a ban are great enough to offset the potential reduction in driving by smokers
who stay home to avoid the ban, then there will be in increase in alcohol-related accident risk.
To test whether we observe a change in alcohol-related accidents following smoking bans,
we use county-level data from the US, a country without a national smoking ban but a large number
of states, counties, and municipalities that have individually passed bans at different points in time,
the vast majority of which were implemented over the past six years. This presents a natural
laboratory to demonstrate the effects of smoking bans. Previous research has used this variation to
test for effects of smoking bans on bar and restaurant employment (Adams and Cotti 2007). In this
paper we use this variation to test for changes in fatal accidents involving alcohol following smoking
ban implementation. Our estimates reveal a significant increase in the danger posed by drunk
drivers following the passage of bans. Specifically, our preferred estimate indicates that fatal
accidents involving a drunk driver increase by about 13%. This is approximately 2.5 fatal accidents
a year for a typical county.3 Although a striking result and one that must be considered by
policymakers, we show later in the paper that the effect is plausible in light of estimates in the
literature regarding the effects of other policies.
As we will demonstrate below, these estimates are robust to the inclusion of controls for area
and time fixed effects, changes in population, changes in other policies that may impact drunk
driving behavior (e.g. beer taxes, blood alcohol content regulation), as well as changes in factors that
2 Number of licensed drivers obtained from U.S. Department of Transportation, Federal Highway
Administration, Highway Statistics 2003.
3 Details on this calculation will be provided in the paper.
may influence overall driving risk separate from drinking behavior (e.g. construction, weather, etc.).
Furthermore, these estimates are also robust to several alternative definitions of the control group,
the dependent variable, the policy variable, the level of analysis, and to the estimation method
selected (e.g., weighted least squares, negative binomial, etc).
We test the underlying mechanism of the smoking ban-drunk driving relationship by
investigating specific cases where smokers might reasonably be expected to cross into another
jurisdiction where smoking and drinking are allowed, and consistent with the expectations, we find
pronounced evidence of increased fatal accidents in border jurisdictions as well. We also engage in
additional case studies where crossing a border is not possible due to the geographic isolation of the
jurisdiction passing the laws. In many of these cases, we find increases in alcohol-related accidents,
suggesting that non-compliance within a jurisdiction is potentially causing some accidents as well.
Ultimately, we view the findings in this paper as a warning that the heightened risk posed by
drunk drivers must be addressed by local and state governments when they ban smoking in their
bars. If not, they face an unanticipated social cost of an ordinance that is intended to improve
people’s health. Alternatively, the results are suggestive that a well-enforced ban that covers the
entire nation, thereby eliminating the attractiveness of crossing a border to smoke and drink, might
avoid the heightened risk posed by drunk drivers.
2. Background on US smoking bans and theoretical considerations
Nearly one-third of the US population lives in communities that have banned smoking in bars
according to the Americans for Nonsmokers’ Rights. More bans in the US are inevitable.
Ordinances were initially enacted at the local level, but as of January 2007, California, Colorado,
Connecticut, Delaware, Hawaii, Maine, Massachusetts, Ohio, New Jersey, New York, Rhode Island,
Vermont, and Washington had prohibited smoking in bars. Although the prevalence of these laws is
higher in urban areas, statewide bans have resulted in many rural bars being smoke-free.
Most existing research has focused on the impact of smoking bans on bar business. Because
bans are a relatively new phenomenon in the US, the initial evidence was mixed and often depended
on subjective opinions of bar owners (Hyland et al. 1999; Dunham and Marlow 2000). The only
fixed effects estimation on a panel of US jurisdictions passing smoking bans on bars observed a
decrease in bar staffing following bans that is not observed in a control group of counties without
bans (Adams and Cotti 2007). Outside the US, one recent study of Scottish public houses found
evidence consistent with a reduction in bar patronage following the Scottish smoking ban. This
reduction was relative to a comparison group of pubs in northern England (Adda et al. 2007). The
long run impact on bar business will likely take more years of data and study to resolve.4
As mentioned in the introduction, at first glance one might be inclined to think that the
danger posed by drunk drivers might decrease following bans if smokers choose not to go out as
often or not stay at their favorite bar as long. It may also be likely, however, that many consumers
will choose to find alternative locations to smoke and drink. Even if the number of bar patrons falls,
the patrons choosing to find new locations to drink and smoke may still increase the total number of
miles driven to and from bars. This will result in a heightened accident risk.
Two strands of existing literature on consumer behavior support this contention. First, the
cross border shopping literature informs that people will consume what they desire in another
4 Another strand of evidence investigates the relationship between smoking bans and heart attacks.
The initial work on this shows a decrease in heart attacks following passage of an indoor smoking
ban (Sargent et al. 2004). We recognize that the increase in drunk driving deaths following smoking
bans must be weighed against other health gains from bans, both potentially short term in the case of
a reduction in heart attacks, and long term in terms of less exposure to second hand smoke by
patrons and employees. For this reason, we do not think our findings should be considered an
indictment of smoking bans. Rather, potential drunk driving risks must be addressed by
communities passing bans.
location in the presence of limits or relatively high costs on consumption in their own jurisdiction
(e.g., Asplund et al. 2007 and Ferris 2000). When this is legal, establishment enclaves bordering
high cost jurisdictions can flourish, as with shopping malls and outlet stores in Pennsylvania near the
border of New Jersey (Pennsylvania has no sales tax on clothing). When cross border shopping is
not legal, smuggling occurs, as is the case with cigarettes (e.g., Chaloupka and Warner 2000 and
Gruber et al. 2003). Some products cannot be smuggled, however, so Canadians cross the border to
consume health services in the US and, in the case of smoking bans in bars, people may want to
consume cigarettes and alcohol across the border of their jurisdiction. For example, the state of
Delaware banned smoking in bars and their neighbor Pennsylvania did not. It is possible that some
people in Wilmington, Delaware might wish to cross the border to smoke and drink. This might
impose an increased danger to drivers in border counties of both states. In fact, some bar owners
have blamed their loss of business on cross border shopping (McCormick-Jenkins 2007).
Second, although cross border shopping would cause greater distances driven by intoxicated
motorists, drunk driving might increase even if consumers remain within their jurisdiction. If
sufficient demand exists, bars will risk non-compliance with a law or set up special outdoor areas
where smoking and drinking are allowed. Ample anecdotal evidence suggests that some bars indeed
risk non-compliance (e.g. Widenman 2006; Falgoust 2004). Others build outdoor patios following
bans (e.g., Rolland 2006). Therefore, after a ban, there is more product differentiation between
bars. Smokers will drive to another bar if their additional costs do not cause their total costs to
exceed their benefits from finding a bar where smoking is permitted. In fact, Lee (1997) applied a
Loschian (1954) location model to describe the hexagonal market areas created by bar service
differentiation. He posits that bar differentiation leads to more drunk driving. Non-compliance and
outdoor seating represent sources of differentiation in terms of ability to smoke, so a heightened
drunk driving risk is anticipated as consumers drive farther to find the bar characteristic they desire.
Both the cross border and product differentiation hypotheses are bolstered by the fact that
cigarettes and alcohol are highly complementary (Dee 1999a). Moreover, smokers are perhaps more
likely to pose a danger on the road than typical bar patrons. According to NIAAA, between 80 and
95 percent of alcoholics smoke cigarettes; a rate that is three times higher than among the population
as a whole. Approximately 70 percent of alcoholics are heavy smokers (i.e., smoke more than one
pack of cigarettes per day), compared with 10 percent of the general population (NIAAA 1998).
Moreover, Di Franza and Guerrera (1990) finds individuals who smoke are about ten times more
likely to be alcoholics. Since smokers are the individuals we would most expect to increase their
driving to evade a smoking ban and a disproportionate number of alcoholics are smokers (heavy
smokers at that), then it is logical to suggest that a smoking ban encourages travel by the individuals
who are most likely to drive while exceedingly intoxicated.
Looking beyond the economic literature, there is growing evidence from laboratory
experiments in neuroscience that nicotine inhibits the intoxicating effects of alcohol in the brain
(Bachtell and Ryabinin 2001; Rohrbaugh et al. 2006). So, if smoking typically mitigates a smoker’s
level of intoxication, then the removal of the nicotine from the bar could increase the level of
intoxication of consumers leaving the bar. Psychological experiments by Palfai et al. (2000) find
that nicotine deprivation increases the urge to drink and the volume of alcohol consumed. This
would suggest that smokers would be inclined to drink more in a smoke-free environment then
would otherwise be the case and, therefore, an increased prevalence of impaired driving results. If
smokers drink more in the absence of smoking or are influenced by the alcohol in a stronger way, or
both, the passage of a smoking ban could increase the number of alcohol-related accidents by
increasing the number of intoxicated drivers. Although none of the aforementioned studies test
driving ability specifically, the interactive effect of alcohol consumption and nicotine are
pronounced enough that a connection between smoking and drunk driving is reasonable speculation.
In the remainder of the paper, we investigate whether these theoretical predictions are indeed
verified by observing the US experience of banning smoking in bars. We find substantial evidence
that the number of fatal accidents involving alcohol increases after smoking bans. We also provide
additional evidence consistent with both the cross border shopping and product differentiation
Data and Methods
Data sources on smoking bans and fatal accidents
Smoking bans have been passed in every region of the US at the state, county, and city level.
For this study, we identify the set jurisdictions that enacted smoking bans from 2000 to 2005, a
period in which many of the bans in existence in the US were passed.5 Information on the dates and
coverage of smoking bans was obtained from the Americans for Nonsmokers’ Rights. Table 1 lists
the jurisdictions with smoking bans that we study, which compose the treatment group for our
We link these data on state, county and city smoking bans to data on fatal vehicle crashes
obtained through the Fatality Analysis Reporting System (FARS) of the National Highway Traffic
Safety Administration (NHTSA). Our primary variable of interest is the annual number of fatal
accidents in a county for which a driver’s imputed blood alcohol content (BAC) exceeds 0.08.
Federal law requires BAC levels be obtained from every fatal crash, but this frequently is not done.
5 The cutoff for 2005 was the enactment of a ban by mid-year. Locations enacting bans later in 2005
are excluded from the sample due to insufficient information to analyze effects of the ban.
Therefore, using only cases with measured BAC drastically understates the number of crashes where
alcohol was a factor and this understatement will vary across jurisdictions, causing substantial bias
in any estimation. The NHTSA is aware that this poses a problem to empirical research and public
discourse. As a result, the FARS data allow for imputation of the BAC for all drivers who were not
tested via a multiple imputation procedure yielding ten different simulated BAC measures for each
driver in every accident. These values are obtained using the multitude of characteristics in each
case, including factors such as time of day, day of week, contents of the police report, position of car
in the road, etc. (NHTSA 2002). This follows suggestions from Rubin et al. (1998) and improves on
the former procedure based on discriminant analysis (Klein 1986; NHTSA 2002). While previous
studies using counts generated from older FARS data used imputed values based on discriminant
analysis or relied on counts generated from accidents that were more likely to be alcohol-related (eg,
crashes on weekend evenings), more recent studies use data generated by the new procedure (e.g.,
Vallaveces et al 2003; Hingson et al. 2005; Cummings et al. 2006). Although we are certain that our
counts of fatal accidents involving alcohol by county-year are the most accurate possible, one must
still be aware that estimated effects of policies may be biased if the rate of the need to impute
information is systematically related to the policy in question. There is little concern that this is the
case here, as smoking bans should not affect how officers investigate a crash scene, as opposed to
changing BAC requirements for example.6
6 As noted above, other studies restrict attention to certain types of accidents, such as whether an
accident took place on the weekend, during the evening, age of drivers, etc. to estimate whether
policies affect alcohol-related crashes or fatalities. We do not need to do this because our multiple
imputation procedure uses a combination of all of this information to generate highly accurate counts
of alcohol-related accidents in a county-year directly. Moreover, to the extent that some heavy
drinkers who smoke may choose to evade smoking bans during off peak times where enforcement or
bar compliance might be more lax, we do not want to restrict our attention to any one characteristic
of an accident.
Following NHTSA procedures that are used to generate their official statistics, we aggregate
our counts of fatal accidents involving a driver with a BAC content exceeding 0.08 by county. We
can link annual fatal accident counts to other data available by county annually (e.g., population data
from the US Census Bureau). Moreover, annual counts provide us with a sufficient number of
accidents for each county upon which to base the analyses.
Our policy variable is a dummy variable indicating that a county has a smoke-free bar law in
place for a given year. Since counties are subsets of states, the only problem may emerge with city
bans, where the city is a subset of the county population. In each case, we attempted to only identify
those counties that we judged to be predominantly smoke-free because of a ban. The cutoff we use
is whether half of the county’s population fell under the provisions of the law.7 Since we expect
going smoke-free to be a process for some bar owners, especially for those wishing to build patio
seating, we suspect that some adjustments could be made after passage of a ban but before actual
enactment of the ban. For this reason, we code bans enacted by the summer to be effective for a
given year. This seems most appropriate to detect effects of the policy, but later in the paper we
check whether alternative measures of the timing of laws influence our results. We also test for the
presence of lead and lagged effects. Our results will prove robust.
For our main estimates, we include only counties that have greater than zero fatal accidents
for the six years of our study to facilitate our use of logs of the dependent variable in the basic
estimation. This excludes some small counties with zero estimated accidents. We verify their
exclusion causes no meaningful change in the results. We also exclude counties that are difficult to
7 For example, Columbus, Ohio went smoke-free and we identified Franklin County as smoke-free,
but we did not code Pitkin County, Colorado smoke-free when Snowmass Village banned smoking
in bars because it is a much smaller percentage of the county’s overall population. We checked to
see whether coding only counties with 100% bans changed the results. It actually makes the results
stronger in the rest of the paper. Thus, we are reporting more conservative estimates.