NBER WORKING PAPER SERIES
THE VIRTUOUS TAX:
LIFESAVING AND CRIME-PREVENTION EFFECTS OF THE 1991 FEDERAL ALCOHOL-TAX INCREASE
Philip J. Cook
Christine Piette Durrance
Working Paper 17709
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
We thank Liz Ananat, Ashu Handa, Jens Ludwig, Christina Peters, and Seth Sanders for helpful suggestions
on an earlier draft. Erin Hye-Won Kim and Vignesh Nathan provided superb assistance. The views
expressed herein are those of the authors and do not necessarily reflect the views of the National Bureau
of Economic Research.
NBER working papers are circulated for discussion and comment purposes. They have not been peer-
reviewed or been subject to the review by the NBER Board of Directors that accompanies official
© 2011 by Philip J. Cook and Christine Piette Durrance. All rights reserved. Short sections of text,
not to exceed two paragraphs, may be quoted without explicit permission provided that full credit,
including © notice, is given to the source.
The Virtuous Tax: Lifesaving and Crime-Prevention Effects of the 1991 Federal Alcohol-Tax
Philip J. Cook and Christine Piette Durrance
NBER Working Paper No. 17709
JEL No. H2,H23,I12,K42
On January 1, 1991, the federal excise tax on beer doubled, and the tax rates on wine and liquor increased
as well. These changes are larger than the typical state-level changes that have been used to study
the effect of price on alcohol abuse and its consequences. In this paper, we develop a method to estimate
some important effects of those large 1991 changes, exploiting the interstate differences in alcohol
consumption. We demonstrate that the relative importance of drinking in traffic fatalities is closely
tied to per capita alcohol consumption across states. As a result, we expect that the proportional effects
of the federal tax increase on traffic fatalities would be positively correlated with per capita consumption.
We demonstrate that this is indeed the case, and infer estimates of the price elasticity and lives saved
in each state. We repeat this exercise for other injury-fatality rates, and for nine categories of crime.
For each outcome, the estimated effect of the tax increase is negatively related to average consumption,
and that relationship is highly significant for the overall injury death rate, the violent crime rate, and
the property crime rate. A conservative estimate is that the federal tax reduced injury deaths by 4.7%,
or almost 7,000, in 1991.
Philip J. Cook
Sanford School of Public Policy
215 Sanford Building
Durham, NC 27708-0245
Christine Piette Durrance
University of North Carolina
Thirty years of economic research has established that consumer choices about drinking are
responsive to prices in the expected way (Chaloupka et al. 2002; Cook 2007). It is reasonable to
conclude that alcohol excise taxes are “virtuous” in that by raising prices they reduce alcohol
abuse and related consequences for public health and safety. Indeed, a number of studies have
explored the direct effect of alcohol taxes on injury mortality, crime, and other negative
consequences of abuse. Some but by no means all of the reported findings support this
hypothesis. At this point it is difficult to distinguish between two possible explanations for the
inconsistent results: a substantive explanation (that the reduction in drinking that results from
tax-induced price increases in fact does not have much effect on some of these outcomes) and an
explanation based on the lack of statistical power. The latter explanation is plausible: analyses
based on observed changes in state excise-tax rates have been limited by the fact that in recent
decades it has been rather rare for state legislatures to change tax rates, and changes that have
occurred have typically been small. In effect, “nature” has failed to perform the experiment that
scholars need to achieve more confident results.
Actually, there has been one notable instance of a large excise tax increase, but it has been
bypassed as a possible source of evidence. On January 1, 1991, the federal tax on beer doubled
(to about a nickel per drink), and tax rates on wine and spirits were also increased. Alcohol
prices jumped an average of 6 percent (adjusting for overall inflation). The problem for would-
be evaluators is that because this change was nationwide, it did not come paired with any
obvious control group.
In this paper, we propose and implement a simple method for estimating the effects of this
national tax increase that does make use of interstate differences – not in tax changes (which
were uniform), but rather in persistent differences in drinking. Our working hypothesis is that
the proportional effect of an alcohol tax increase on, say, traffic fatalities, is positively related to
a state’s per capita alcohol consumption. Simply put, relatively “wet” states will benefit more
from a given tax increase than “dry” states. We provide evidence in support of that hypothesis
(including the strong positive correlation between per capita consumption and the percent of
traffic fatalities that involve drinking), and then use it to justify a new method for estimating the
effect of the federal tax change on injury deaths and crime.
The method involves estimating cross-section regressions with per capita alcohol consumption as
the independent variable, for which the dependent variables are proportional changes in the
outcome variables around the time of the federal tax increase. We generate these estimates for
all injury deaths (total, and then for suicide, homicide, and motor-vehicle deaths separately), as
well as for nine categories of crime. In every case, the estimated effect is negative, and in most
cases the null hypothesis can be confidently rejected. Our interpretation of these estimates rests
on two linked assumptions: (1) alcohol abuse causes higher proportions of injury deaths and
crime in wet states than dry; and (2) an alcohol tax increase reduces injury deaths and crime that
are caused by alcohol abuse, cet. par.
This approach provides a joint test of the two assumptions, and also provides the basis for
estimating the impact of the tax increase at different points in the spectrum of “wetness.” For
example, we predict that wetter states like California, Florida, and Wisconsin enjoyed greater
proportional reductions in injury deaths and crime than did drier states like Utah, West Virginia,
and Kansas. Thus our approach may be of interest in documenting heterogeneity in the
“treatment effects” of alcohol taxes. Instead of attempting to estimate the average treatment
effect, we estimate the spectrum of treatment effects associated with a particular moderator – per
capita consumption. Our approach contributes to a recent body of literature that seeks to estimate
the structure of heterogeneous policy effects (Ananat & Michaels, 2008; Bitler, Gelbach, &
Hoynes, 2006; Neumark, Schweitzer, & Wascher, 2004; Blank & Schoeni, 2003).
Our substantive findings provide consistent support for a conclusion that the alcohol-price
elasticity for several outcomes is closely related to average alcohol consumption. For injury
fatalities, we find strong and consistent results for traffic and overall. For our crime categories,
we find strong and consistent results for overall violent crime, aggravated assault, robbery,
overall property crime, burglary, and motor-vehicle theft. The gradient relating per capita
consumption to price elasticity is about twice as steep for violent crime as for property crime.
We proceed as follows. The next section summarizes relevant knowledge linking drinking to
injury and crime, and the effects of alcohol control measures on drinking and related outcomes.
We then characterize the federal tax increase of 1991 more fully, documenting its effects on
alcohol prices. Section 4 develops our regression specification and documents the relationship
between per capita consumption and the proportion of traffic fatalities that involve alcohol.
Section 5 presents results for injury mortality and then crime. Section 6 reports the results of
alternative specifications, concluding that the results are highly robust. A final section concludes.
Drinking and consequences
Intoxication has long been considered an important cause of injury and crime. This connection
has figured most prominently in shaping policies to prevent traffic fatalities, but drinking is also
an important factor in homicide, other types of violent crime, suicide, and other injury.
The evidence for a causal effect of intoxication begins with what is known about the
pharmacological and social effects. The pharmacological effects generally depend on the
amount and duration of drinking relative to body mass, and a variety of other factors that make
the effects quite heterogeneous. Generally speaking, when a drinking session is initiated, the
increasing alcohol level is a stimulant, and may engender loss of inhibition and increased
sociability. At higher levels of consumption, impairment of judgment, other cognitive processes,
and coordination will occur. As alcohol in the blood is metabolized, the stimulant effect gives
way to a depressant effect. Very high levels of intoxication may lead to coma or death due to
overdose. Also relevant is that alcohol is an anesthetic (and in fact was used as such in medical
procedures before modern anesthetics were introduced), which may change individual
assessment of getting into a bar fight, say. In sum, the pharmacological effects are diverse. In
some people under some circumstances, alcohol intoxication can lead to reckless driving,
suicide, other injuries, or involvement in crime either as a perpetrator or a victim.
The social effects of drinking depend on the cultural context in which it occurs. In Western
cultures, intoxication often serves as an excuse for bad behavior – the man who abuses his wife
or gets in a fight or sexually assaults an acquaintance may have a better chance of forgiveness if
he gets drunk first. Furthermore, drinking is linked to the nature of the social context, since
much drinking occurs in bars and other public places. Crowds of impaired people provide
opportunities for conflict and victimization.
The statistical characterization of these associations begins with the rather high prevalence of
intoxication in injury deaths and crimes. Table 1 provides a consistent set of estimates for 2001
injury deaths, in which deaths are “alcohol-attributable” if “the decedent (or, as in the case of
motor-vehicle traffic, a driver or non-occupant) had a BAC of>0.10 g/dL” (Midanik et al. 2004,
p. 866). This standard is conservative in that it exceeds the nationwide per se limit for driving
under the influence (0.08%). About 25% of the motor vehicle deaths meet that standard.
Overall, 19% of injury deaths of all types were deemed alcohol-attributable based on the
Table 1 about here
Statistics on the prevalence of alcohol involvement in criminal acts come from several sources.
Occasional surveys of inmates include questions on whether they had been drinking at the time
of their crime. The 1997 Survey of Inmates in State Correctional Facilities (conducted by the US
Department of Justice), for example, found that 42 percent of inmates convicted of a violent
crime had been drinking, compared with 34.5 percent of those convicted of a property crime
(Table 2). Another source of information is the National Crime Victimization Survey (NCVS),
which routinely asks respondents who report a violent crime whether they thought the
perpetrator had been drinking or using drugs at the time of the crime. Excluding those who said
that the perpetrator was using drugs but not drinking, 29 percent of the violent crimes involved
or possibly involved drinking. The percentage so reported for rape is 41.5%, similar to the
percentage from the inmate survey. (Note that neither the NCVS nor the National Prisoner
Survey inquires whether the victim was drinking at the time of the crime.)
Table 2 here
Of course this association does not demonstrate the extent to which alcohol is a cause of injury
or crime. People who drink to excess tend to have other personality traits that may explain why
they engage in risky and criminal behavior: in the case of drunk driving, these traits include
emotional instability, impulsiveness, hostility, and depression (Donovan, Marlett, and Salzberg
1983). Indeed, people who drive drunk from time to time are more likely to get into a crash even
when sober (Levitt and Porter 2001). In principle this sort of confounding can be overcome
through laboratory experiments, and there has been some research of this sort. Laboratory
experiments with humans have demonstrated that drinking degrades driving ability, while
experiments with both humans and animals have provided strong evidence that drinking leads to
more aggressive behavior in some people (and monkeys). But the scope for such experiments is
A quite different approach to assessing the causal effect of drinking on injury and crime is to
evaluate policies and events that affect the price and availability of alcoholic beverages.1 The
alcohol-control measures that have been most thoroughly evaluated are the minimum legal
drinking age and alcohol excise taxes. State-level changes in the minimum drinking age were
common in the 1970s and 1980s, and a number of studies (beginning with Cook and Tauchen
1984) utilized difference-in-difference panel-regression methods on state data to assess the
effects on various outcomes including teen traffic fatalities and suicide rates. There have been
1 Cook (2007) provides a review of this evidence for both injury and crime. Carpenter and Dobkin
(2011b) review the evidence on drinking and crime.
no changes in minimum age since 1987, but a series of regression discontinuity studies focused
on age 21 have given persuasive evidence of its effects on crime and other outcomes (Carpenter
& Dobkin 2009; Carpenter & Dobkin 2011a).
The economic literature on the effects of alcohol-excise taxes has also made extensive use of
panel regression studies of state-level data (beginning with Cook and Tauchen 1982). In this
analysis, the use of the tax rate as the regressor is justified from a “reduced form” interpretation:
the presumed mechanism is that a tax increase is passed on to consumers in the form of higher
prices (confirmed by Young & Bielinska-Kwapisz 2002), and that the price increase reduces
alcohol abuse related to the outcome in question. This literature helps establish the effects of
alcohol sales on alcohol-related outcomes such as crime, traffic fatalities, and sexually
transmitted diseases (Chesson et al 2004). Note that a finding that an increase in an excise tax
affects outcome Y (crime, injury) is directly relevant as part of evaluation of such taxes, and
also, indirectly, a confirmation of the causal effect of drinking on outcome Y. One possible
challenge to this causal interpretation is that the decision by state legislatures to change taxes
may be influenced by near-term patterns in alcohol use or abuse, and hence cannot be interpreted
as completely exogenous (Young & Bielinska-Dwapisz 2006).
To illustrate the results from panel-regressions on state excise tax changes, Table 3 provides a
consistent set of estimates for alcohol sales (measured as gallons of ethanol sales per capita) and
four types of injury deaths – those resulting from crashes, falls, homicide, and suicide. In each
case, the regression is based on annual state-level data for the period 1981-2000. The second
column provides evidence that changes in state per capita alcohol sales are closely associated
with the injury-death outcomes (with the exception of homicide). The last column reports that
the effect of tax increases on sales is highly significant (p<.01), while the “reduced form” effect
of excise taxes on motor-vehicle fatalities is marginally significant. The estimated effects on
falls, homicide, and suicide are not significant, and the effect for homicide has the “wrong” sign.
These rather weak results may well reflect the lack of statistical power in the quasi-experiment –
nominal excise-tax rates are rarely changed, and the changes that occurred during this period
were in most cases quite small.2
Table 3 about here
Several studies have assessed the effects of beer-excise taxes on crime rates (Carpenter and
Dobkin 2011b). Cook and Moore (1993) used state-level panel data for the period 1979-1987
with two-way fixed effects, finding that increases in beer tax rates reduced rape and robbery
rates. (See also, Chaloupka et al., 2002; Sloan et al., 1994). DeSimone (2001) conducted a
similar study of 29 large cities for the period 1981-1995, finding a negative effect of beer taxes
on rape, assault, larceny, and motor vehicle theft. A study of injury-producing violence in
2 It should be noted that research evaluating the effect of tax changes on mortality is by no means limited
to the economics literature. Much of this research utilizes time-series methods to evaluate changes in a
single state. For example, Wagenaar et al. (2009) finds that two rather large changes in alcohol-excise
taxes in Alaska at different times were associated with immediate and sustained reductions in alcohol-
related disease mortality. A similar analysis for Florida had a similar result (Maldonado-Molina and
Wagenaar 2010). There is also a related literature based on British and European experience (e.g.,
Purshouse et al., 2010; Babor et al., 2003).
England and Wales also found a negative effect for beer taxes (Matthews et al., 2006).
Markowitz (2005) utilized panel data on individuals from three consecutive years of the NCVS,
finding marginal evidence that the beer tax reduced physical assault victimization, but with null
results for rape or robbery. She has also analyzed the effect of beer taxes on domestic violence
in a series of studies, including several that utilize repeated cross sections (Markowitz and
Grossman, 2000; Markowitz 2000) with mixed results.
One consistent finding of the “crime” literature is a null result for the effect of alcohol taxes on
homicide rates. The most recent contribution is Durrance et al. (forthcoming), which utilizes a
panel of state-level data, reporting a null effect of state alcohol excise tax rates on female
The Federal tax increase of 1991
In 1990, federal legislation was signed by President George H.W. Bush that increased excise
taxation on tobacco, gasoline, and alcohol. Taxes on beer increased from 29 cents to 58 cents per
gallon, taxes on wine increased from 17 cents to $1.07 per gallon, and taxes on liquor increased
from $12.50 to $13.50 per proof gallon (which is the volume of liquid that contains 64 ounces of
ethanol). Using appropriate conversion figures, these tax increases amounted to approximately a
5.0 cent per ounce of ethanol increase in beer, a 5.5 cent per ounce of ethanol increase in wine,
and a 1.5 cent per ounce of ethanol increase in liquor.3
It is of interest that Congress has not changed the alcohol tax rates since 1991, and the price
effect of these increases has been substantially eroded by general inflation. At the time,
however, the tax increases were associated with an abrupt departure from the trend in alcohol
prices. Figure 1 depicts the ratio of the Consumer Price Index for alcoholic beverages to the
overall CPI for the years around 1991. The increase of 6 percent (in real terms) between 1990
and 1991 began fading after 1992.
Figure 1 about here
Our window of analysis is the two-year period that brackets the federal tax increase. During that
time there were four states that increased alcohol excise tax rates, three of them by more than a
trivial amount. We return to this matter below.
One approach to justifying our regression specification builds on the assumption that injury
deaths and crimes can be divided between those that are in some sense caused by drinking and
those that are not. Epidemiologists estimate the “attributable portion” of deaths that are caused
by drinking, as noted above. We do not attempt to specify an operational definition of the
attributable portion in what follows, but only suppose that there is a subset of fatalities that are
attributable to drinking, and that the proportion of such fatalities is closely related to how “wet”
or “dry” the state is, as indicated by average consumption of ethanol. Thus we posit that the
effect of the tax increase on each of our outcome variables (injury deaths, crimes) is moderated
3 There is approximately 5.76 ounces of ethanol in a gallon of beer, 16.51 ounces of ethanol in a gallon of
wine. A standard drink of beer, wine, or spirits includes about 0.5 ounces of ethanol.
by the proportion of the outcome that is attributable to alcohol – and that that proportion is
directly linked to per capita consumption in the state.
In what follows, variables can be subscripted by state s and by period t.
Dts = death rate in state s during a specified period t (t=0 for baseline period; t=1 for period
following tax increase)
Θs = fraction of deaths due to alcohol in baseline
1-k = fraction of alcohol-related deaths prevented by the 1991 tax increase
Cs = per capita ethanol consumption
D1s = [k θ s + (1- θ s)] D0s
ΔDs/D0 s = (-1+k) θ s
Now suppose there is a uniform proportional change p in the state death rates between the two
periods due to factors other than the tax increase:
ΔDs/D0s = p + (-1+k) θ s
Finally, we postulate state-specific random variation in the proportional change between the two
periods that is additive:
ΔDs/D0s = p + (-1+k) θ s + εs
We assume a linear relationship between θ and C:
θ s = a + b Cs
Substituting, we have
ΔDs/D0s = p + (-1+k)(a + bC s) + εs = p + a(- 1+ k) + (-1+k)b Cs + εs
Thus the proportional change in the outcome is linear in C, as follows:
(1) ΔDs/D0s = α + β C s + εs where α = p + a(-1 + k)] and β = (-1 + k) b
An estimate of β then quantifies the range of effects of the 1991 tax increase on the proportional
change in the outcome variable (crime or death rate). In particular, we can estimate how much
more the tax increase affected alcohol-related outcomes in, say, Wisconsin, than in a relatively
dry state such as Utah. However, without an estimate of p, it is not possible to estimate the
absolute effect of the tax increase (as opposed to the effect relative to other states). It is logically
possible to estimate p from knowledge of the parameters a, b, and β (which then provides an
estimate of k), all of which we will estimate. However, in what follows we do not attempt that
estimate, but rather report the results of a more conservative approach that sets a lower bound on
the magnitudes of the effects.
The model assumes that in the absence of the federal tax increase, the states would have
followed the same relative trajectory from 1990 to 1992 with respect to outcome variables, so
that ps = p for all states. If that assumption is not correct, and in particular if the trajectories are
correlated with C, then the OLS estimate of equation (1) will produce a biased estimate of β.
There is some evidence, discussed below, that that assumption is incorrect for some outcome
variables. For that reason, we experiment with a “momentum” model that assumes that ps is
linearly related to the recent trend:
ps = γ + φ (ΔDs/D0s)-1
where the independent variable is the proportional change over a previous period (such as 1989
to 1990, or 1985 to 1990).
We then estimate the following equation:
(2) ΔDs/D0s = α’ + β’ C s+ φ (ΔDs/D0s)-1 + εs
The relationship between alcohol consumption and alcohol involvement in traffic
The explicit assumption for specifications 1 and 2 is that C is linearly related to the proportion of
deaths or crimes that are alcohol-related. While there are some data on alcohol involvement in
crime and injury (summarized in the previous section), a systematic state-by-state tabulation is
lacking. Fortunately, more comprehensive data are available for traffic fatalities.
As it turns out, the relative importance of drinking in fatal accidents is indeed highly correlated
with per capita alcohol consumption across states. To demonstrate this relationship, we use data
from the Fatality Analysis Reporting System (FARS). FARS publishes data by state and by year
on fatal crashes according to various characteristics. These compilations are based on
administrative reports from state agencies. The year 1994 is the first one for which the data are
reasonably complete, and we analyze data from that year and also from 2005 (which is generally
considered of higher quality). Our measure of alcohol involvement is Alc%, the percentage of
drivers killed in fatal crashes in which at least one of the drivers in the crash had blood alcohol
content exceeding the legal limit of 0.08%.4
Per capita ethanol consumption data are from the National Institute of Alcohol Abuse and
Alcoholism (NIAAA) and identified as “apparent consumption” by that agency. The data are
based on alcohol sales measured (as part of the excise-tax collection system) as withdrawals
from the distributors’ warehouses. The unit of measurement for consumption is gallons of
ethanol per year, C. Ethanol, the “active ingredient” of alcoholic beverages, is measured and
taxed directly for spirits; for beer and wine, it can be estimated from market averages as 4.5% of
4 In other models (not reported), we use the percentage of drivers involved in fatal crashes where alcohol
was involved. The results are similar.
gallons of beer, and 12.9% of gallons of wine. In most states the assumption that sales are about
equal to consumption is reasonably accurate.5
As shown in Figure 2, the cross-section scatterplot between the two variables appears to have a
linear axis over the observed range, despite the fact that the dependent variable is bounded.
Figure 2 about here
For the two years that we consider (1994 and 2005), the slope coefficients are almost identical.
The correlation coefficients are 0.56 in 1994 and 0.70 in 2005. Table 4 provides additional
Table 4 about here
While there is no guarantee that per capita consumption is also highly correlated across states
with alcohol involvement in other types of injury deaths, or with crime, we believe that the
“proxy” interpretation is plausible.
The effect of the 1991 tax increase on injury deaths and crime
Here we present and discuss the regression results using the specifications (1) and (2). In every
case, the dependent variable is the percent change in the outcome variable between the year
following the federal tax increase (1991) and the year preceding the tax increase (1990). The
regressions are estimated from 47 states. Excluded are the District of Columbia, Nevada, and
New Hampshire. For these jurisdictions a large proportion of sales are to non-residents in these
jurisdictions, so that the sales figure is a poor representation of consumption (Cook 2007). We
also exclude Alaska, due to problems with the crime data. The independent variable in all these
regressions is the per capita ethanol consumption (C) for 1989. We did not use 1990 data since
they would be affected by inventory adjustments in anticipation of the federal tax increase. The
regressions are weighted by state population in 1990.
First results for injury deaths
Fatal injury data are drawn from the Web-Based Injury Statistics Query and Reporting System
(WISQARS), compiled by the National Center for Health Statistics as part of the Vital Statistics
program. We use non-age-adjusted crude death rates (CDRs), where the population data used to
create the rates are available through WISQARS but originally drawn from the US Census
Bureau. These rates are calculated per 100,000 state residents. Specifically, we use CDRs for
motor-vehicle traffic deaths. Further information on these, and other variables used in the
subsequent analysis, is presented in Tables 5 and 6.
Tables 5 and 6
The data columns of Table 7 report the regression estimates for model (1) on all injury,
homicide, suicide, and traffic fatalities. The coefficients on 1989 ethanol consumption are
negative in every case. The coefficients for all injuries and motor vehicle traffic fatalities are
5 Wastage, inventory changes, interstate sales, and home production are among the sources of difference.
statistically significantly different from zero by the usual standards. Interestingly, the null result
with respect to homicide is a recurrent finding in the literature (Durrance, et al.2011).
The magnitudes of the point estimates are of interest. Combined with the 1989 values of C (per
capita ethanol sales), they provide the basis for computing the relative effects of the tax increase
across states. C ranges from 1.39 gallons in Utah, up to 3.07 in Wisconsin (Table A1), with a
median of 2.42 gallons (Michigan). With respect to the reductions in “all injury” death rate, the
difference between Utah and Michigan is -4.598 (2.42 –1.39) = - 4.7 %. Thus if we make the
conservative assumption that the tax increase had no effect on outcomes in Utah, the 6% increase
in alcohol prices induced by the federal tax increase resulted in a reduction of 4.7% in injury
deaths for the median state. Under this assumption of zero effect in Utah, the elasticity of injury
deaths with respect to alcohol prices is then -0.8 for the median, ranging up to -1.3 for
Results for crime
We now consider the effect of the drinking on the effect of the alcohol tax increase on crime.
Crime data are taken from the Uniform Crime Reports (UCR) compiled from crimes known to
law enforcement agencies and submitted to the Federal Bureau of Investigation (FBI). These data
are available from the Bureau of Justice Statistics.6Rates are calculated by using population data
available from the Census and are per 100,000 people. We employ violent crime rates (murder
and non-negligent manslaughter, aggravated assault, forcible rape, robbery) and property crime
rates (burglary, larceny theft, motor vehicle theft). It should be noted that the “murder and non-
negligent manslaughter” rates are similar but not identical to the “homicide” rates compiled by
the NCHS Vital Statistics program; the two variables are compiled by different agencies from
The regression estimates indicate that both violent and property crime were affected by an
increase in the federal tax rates. As expected (see the discussion above), the effect on violent
crime appears to be more sensitive to state “wetness” than is property crime. Within the
category of violent crime, robbery, aggravated assault, and rape are the most sensitive, and once
again murder appears largely immune (although the sign is negative). Within the category of
property crime, the effect of the tax increase on burglary and motor vehicle theft rates are the
most sensitive to state “wetness,” although the coefficient for larceny is also statistically
significant. Employing the same assumptions as before, the price elasticities for the median state
are -1.3, violent crime and -0.7 (property crime).
The findings from specification (1) could be biased if consumption in the baseline year (1989) is
correlated with state trends in mortality and crime near the time of the 1991 tax increase. To test
for this possibility we first conduct a simple graphical analysis of the data.7 We divide the states
into three groups according to per capita sales in 1989: “low” (the 10 states with sales of 2.0
gallons or less), “medium” (the 27 states with sales between 2.0 to 2.6 gallons), and “high” (the
10 states with sales above 2.6). Our theory predicts that the trajectory of the “high” states will
drop in 1991 by more than the “medium,” and the “medium” by more than the “low.” Figure 6
depicts the relevant trends for traffic fatalities. Panel A demonstrates that the three groups of
states differ widely in traffic safety. To adjust for those permanent differences, Panel B indexes
all three groups to their level in 1990. What then becomes clear is that the post-tax trends are as
predicted (with dryer states having a post-tax trajectory above the wetter states). This picture is
not entirely reassuring, however, since pre-tax trends are somewhat different across the three
groups, as is also clear from Panel B – although the difference is more constrained.
We made similar plots for the other outcomes. Of all the outcomes, robbery comes closest to
having similar pre-tax trends for the three groups of states, as shown in Figure 7. Note that the
robbery results confirm our basic results. Other outcomes follow somewhat different trends both
pre- and post-tax.
One way to account for differences in pre-tax trends more systematically is to expand our basic
regression specification to include a covariate that measures the previous trend in the outcome
variable; in one specification check, the trend is measured as the change in the outcome variable
between 1989 and 1990, and in a second check as the change in the outcome variable between
1985 and 1990.
Table 9 presents the results of these two related specification tests. The results using either
covariate suggest little evidence of so-called momentum (or anti-momentum) effects. In fact, the
coefficients on consumption in 1989 are little changed, even in those cases for which the trend
measure is statistically significant. The one exception appears to be suicide, where inclusion of
the pre-tax trends generates stronger, statistically significant results than our original model.
Changes in state excise taxes and other concerns
While the 1991 tax increase occurred at the federal level, four of the 47 states in our analysis
sample opted to change their own alcohol tax levels during the 1990-1991period. Table 10
reports these state tax changes (in cents per ounce of ethanol) and averages them using as
weights the proportion of ethanol consumed (in 1990) for each type of beverage. The change in
Colorado was trivial. When we drop the other three states from the analysis, we obtain the
results presented in Tables 11 and 12. The results are quite similar to those of all 47 states.
7 We are indebted to Jens Ludwig for suggesting this test.
Table 10, 11, and 12
We also conducted our analysis using two-year periods both pre- and post-change in tax. In
other words, we estimated the effects of consumption per capita in 1989 on changes in mortality
or crime rates for 1991-1992 relative to 1989-1990. The empirical results were quantitatively
similar to those presented here. We chose ultimately to present the results using 1990-1991 data
to avoid the handful of additional state tax changes that occurred in 1989 and 1992.
The federal increase in alcohol excise tax rates of 1991 provides a relatively high-powered
intervention, exogenous to the individual states, that helps establish the causal effect of higher
prices on alcohol abuse and a variety of costly consequences.
Our analysis offers several innovations:
Establish that the state-level prevalence of DUI in fatal crashes is directly related to
average ethanol consumption;
Demonstrate that ethanol consumption levels moderate the effect of alcohol prices on the
rates of injury fatality and crime;
Develop and implement a statistical method for utilizing this heterogeneity to estimate
the aggregate effects of a federal tax increase on rates of injury fatality and both property
and violent crime.
The empirical findings provide consistent support for a conclusion that the alcohol-price
elasticity for several outcomes is closely related to average alcohol consumption. For injury, we
find strong and consistent results for the predicted gradient between consumption and the price
elasticity of traffic fatalities and overall fatalities. Under the conservative consumption that the
driest state (Utah) experienced no benefit from the tax increase, our national point estimate is
that it reduced the number of injury deaths by 6,824, or 4.7%, in the first year (1991). For our
crime categories, we find strong and consistent results for overall violent crime, aggravated
assault, robbery, overall property crime, burglary, and motor-vehicle theft. The gradient relating
drinking to price elasticity is about twice as steep for violent crime and for property crime.
Alcohol tax policy is a powerful tool for reducing alcohol-attributable deaths and crime. But in
the two decades since this tax increase, Congress has not changed the nominal rates, with the
result that the inflation-adjusted rates have been substantially eroded.
Ananat, Elizabeth O. & Guy Michaels (2008). The Effect of Marital Breakup on the Income
Distribution of Women with Children, Journal of Human Resources, 43(3), 611-629.
Babor, Thomas, Raul Caetano, Sally Casswell, Grigffith Edwards, Norman Giesbrecht, Kathryn
Graham, Joel Grube, Paul Gruenewald, Linda Hill, Harold Holder, Ross Homel, Esa Osterberg,
Jurgen Rehm, Robin Room, & Ingeborg Rossow (2003). Alcohol: No ordinary commodity New
York: Oxford University Press.
Blank, Rebecca M. & Schoeni, R.F. (2003). Changes in the Distribution of Children’s Family
Income over the 1990s, American Economic Review, 93(2), 304-308.
Bitler, M.P., J. B. Gelbach &Hilary W. Hoynes, (2006). What Mean Impacts Miss: Distributional
Effects of Welfare Reform Experiments, American Economic Review, 96, 988-1012.
Carpenter, Christopher, & Dobkin, Carlos (2009). The Effect of Alcohol Consumption on
Mortality: Regression Discontinuity Evidence from the Minimum Drinking Age, American
Economic Journal Applied Economics, 1(1): 164-182.
Carpenter, Christopher &Carlos Dobkin (2011a). The minimum legal drinking age and public
health. Journal of Economic Perspectives,25(2), 133-156.
Carpenter, Christopher & Carlos Dobkin (2011b). Alcohol regulation and crime. In Philip J
Cook, Jens Ludwig, and Justin McCrary, eds. Controlling Crime: Strategies and Tradeoffs,
Chicago: University of Chicago Press.
Chaloupka, Frank J., Michael Grossman, , &Henry Saffer, (2002). The effects of price on
alcohol consumption and alcohol-related problems. Alcohol Research & Health, 26(1), 22-35.
Chaloupka, Frank J., Henry Saffer, &Michael Grossman, (1993). Alcohol-control policies and
motor-vehicle fatalities. Journal of Legal Studies, 22, 161-186.
Chesson, Harrell, Paul Harrison, & William J. Kassler (2000). Sex Under the Influence: The
Effect of Alcohol Policy on Sexually Transmitted Disease Rates in the United States. Journal of
Law and Economics, 43, 215-238.
Cook, Philip J. (2007). Paying the Tab: The Costs and Benefits of Alcohol Control, NJ: Princeton
Cook, Philip J. &Michael J. Moore (1993). Violence reduction through restrictions on alcohol
availability. Alcohol Health and Research World, 17, 151-156.
Cook, Philip J., & George Tauchen (1982). The effect of liquor taxes on heavy drinking. The
Bell Journal of Economics, 379-390.
Cook, Philip J. & George Tauchen (1984) “The effect of minimum drinking age legislation on
youthful auto fatalities, 1970-77.” Journal of Legal Studies,13 (January): 169-190.
DeSimone, Jeffrey (2001). The effect of cocaine prices on crime. Economic Inquiry,39: 627-43.
Donovan, Dennis M., G. Alan Marlett, & Philip M. Salzberg (1983). Drinking behavior
personality factors and high risk driving: A review and theoretical formulation. Journal of
Studies on Alcohol,44(3): 395-428.
Durrance, Christine Piette, Shelley Golden, Krista Perreira, & Philip J. Cook (2011) Taxing sin
and saving lives: Can Alcohol Taxation Reduce Female Homicides? forthcoming in Social
Science & Medicine.
Levitt, Steven D. and Jack Porter (2001).How dangerous are drinking drivers? Journal of
Political Economy,109(6): 1198-1237.
Maldonado-Molina, Mildred M & Alexander C. Wagenaar (2010). Effects of alcohol taxes on
alcohol-related mortality in Florida: Time-series analyses from 1969 to 2004.Alcoholism:
Clinical and Experimental Research,34(11): 1915-1921.
Markowitz, Sara (2005). Alcohol, Drugs and Violent Crime. International Review of Law and
Markowitz, Sara & Michael Grossman (2000). The effects of beer taxes on physical child abuse.
Journal of Health Economics, 19(2), 271-282.
Markowitz, Sara (2000). The price of alcohol, wife abuse, and husband abuse. Southern
Economic Journal, 67, 279-303.
Matthews, Kent, Jonathan Shepherd, & Vaseekaran Sivarajasingham (2006). Violence-related
injury and the price of beer in England and Wales. Applied Economics,38: 661-70.
Midanik, Lorraine T, Frank J Chaloupka, R Saitz, TL Toomey, JL Fellows, M Dufour, M
Landen, PJ Brounstein, MA Stahre, RD Brewer, TS Naimi, JW Miller (2004). Alcohol-
Attributable deaths and years of potential life lost – United States, 2001. MMWR Weekly,
53(37), Sept 24: 866-870.
Maldonado-Molina, Mildred M. and Alexander Wagenaar C. (2010) Effects of alcohol taxes on
alcohol-related mortality in Florida: Time-series analyses from 1969 to 2004. Alcoholism:
Clinical and Experimental Research 34(11), Nov: 1-7.
Neumark, D., Schweitzer, M., Wascher, W. (2004). Minimum Wage Effects throughout the
Wage Distribution, Journal of Human Resources, 39(2), 425-450.