ArticlePDF Available

Do Casinos Contribute to Violent Crime? A Panel Data Analysis of Michigan Counties


Abstract and Figures

A key part of the debate over the current rapid increase in the number of casinos in the U.S. concerns the impact on crime rates resulting from the presence of a casino. In this analysis we use panel data covering all 83 Michigan counties for each year 1994-2010 to investigate whether the existence and size of a casino in a host county or in a nearby county affect the rates of four violent crimes: murder, rape, aggravated assault, and arson. We include a number of variables to control for factors that affect crime more generally. We find that county violent crime rates in Michigan do not rise in the presence of a casino, and in the case of aggravated assault, may actually fall. Previous studies of the impact of casinos on a variety of crimes in a variety of locations have shown mixed results.
Content may be subject to copyright.
The Journal of Gambling Business and Economics 2014 Vol 8 No 2 pp 35-55
Gregory A. Falls
Department of Economics
Central Michigan University
Mt. Pleasant, MI, 48859
Philip B. Thompson
Department of Economics MS-9074
Western Washington University
516 High St.
Bellingham, WA 98225
A key part of the debate over the current rapid increase in the number of
casinos in the U.S. concerns the impact on crime rates resulting from the
presence of a casino. In this analysis we use panel data covering all 83
Michigan counties for each year 1994-2010 to investigate whether the
existence and size of a casino in a host county or in a nearby county affect the
rates of four violent crimes: murder, rape, aggravated assault, and arson. We
include a number of variables to control for factors that affect crime more
generally. We find that county violent crime rates in Michigan do not rise in
the presence of a casino, and in the case of aggravated assault, may actually
fall. Previous studies of the impact of casinos on a variety of crimes in a
variety of locations have shown mixed results.
Keywords: casinos, casinos and crime, panel data, crime, violent crime
The primary factors driving recent increases in legal gaming1 in the U.S.
are state and local governments' desires for additional revenue sources and the
passage of the Indian Gaming Regulatory Act of 1988 (IGRA), which
Corresponding author: Philip B. Thompson, Department of Economics MS-9074,
Western Washington University, 516 High St., Bellingham, WA 98225, USA. Email:
1 In 2011, commercial (non-tribal) gaming revenue was $35.6 billion in the U.S.
(AGA, 2012). We will generally use the term gaming as opposed to gambling; the
latter would include state lotteries, which have also grown during the same period, but
we wish to confine our attention to the subset of gambling activities that take place in
casino-like venues. It is also the case that the industry prefers the term gaming,
presumably because it sounds better than gambling, but we are not attempting to
make any sort of value judgments by our choice of terms.
2014, 8 2
permitted Native American tribes to establish casinos offering what is known
as Class III gaming (slot machines, blackjack, craps, etc.). Proposals to allow
casino gambling have sparked debate over the social and economic impacts of
casino development, including the incidence of crime. Proponents of casinos
typically stress their government revenue and potential economic
development aspects while opponents are chiefly concerned about increases in
crime and problem gambling.
In the past 25 years over 20 casinos have opened in Michigan: nineteen
Native American casinos scattered around the state and three commercial
(non-tribal) casinos in Detroit. This paper investigates the impact of these
Michigan casinos on crime rates for murder, aggravated assault, rape, and
arson in casino host counties and in nearby counties. We use a panel data set
with annual observations on all 83 Michigan counties for the 17-year period
1994-2010. The dataset, described at greater length below, includes numbers
of crimes taken from the FBI’s Uniform Crime Report Crime by County data
set, variables for the presence of a casino in a county or in a nearby county,
the scale of a casino’s operations (here based on measures of gaming-related
revenues), the age of each casino, and a variety of demographic and economic
control variables suggested by the broader literature on the factors that affect
crime rates generally.
The casino-crime link is important for two reasons. First, state and local
officials need information to address the desirability of having a casino in
their area.2 Second, these officials would be interested in how a casino would
affect the deployment of local government resources, particularly police and
social services. For example, Calhoun County, Michigan, commissioned a
study by the Upjohn Institute (Erickcek, Timmeney, and Watts, 2008) to
explore the potential impacts of the pending establishment of a casino in the
county. Resource deployment is especially important in Michigan, since all of
the IGRA compacts between the state of Michigan and the tribes require that
tribes make annual payments (in lieu of taxes) to local governments equal to
2% of a tribe’s net casino revenues (gross amounts wagered less winnings
paid to patrons) from electronic gaming machinesslot machines, video
poker, and the like. These so-called 2% payments to local governments are
significant amounts, annually averaging $15.4 million in the aggregate over
the 1994-2010 period, and totaling $26.2 million in 2010 alone (Michigan
Gaming Control Board, 2012a). If the presence of a casino does in fact result
in higher crime rates, local governments would want to increase policing
2 Since Indian casinos are established on tribal lands pursuant to compacts between
tribes and states, local governments' ability to influence the location of such casinos
is, strictly speaking, negligible. States are required by the IGRA to negotiate compacts
in good faith if the state permits such [Class III] gaming for any purpose by any
person, organization, or entity. (25 U.S.C. 2710) But establishing a commercial
casino often requires local approval, perhaps including the passage of a local ballot
activities, but if not, funds could be more wisely directed to other uses,
including increases in social services designed to deal with the effects of
problem gambling or to meet road construction and maintenance costs
associated with increased traffic levels.
There is a substantial and still-growing body of research on the casino-
crime link. Not surprisingly, some of it has been undertaken on behalf of
either pro- or anti-gambling interests or by researchers with ties to such
groups (e.g., Margolis 1997). In a review of earlier research, Miller and
Schwartz (1998) conclude that much of it was characterized more by its
rhetoric in defense of a particular position than by its probative value, and
point out that there were many contradictory results. Others (e.g., Walker,
2007) also criticize some studies of the casino-crime link on the basis of
potential investigator bias, but there are also many investigations outside that
Walker (2010) provides a comprehensive and detailed review of the
casino-crime literature, critically examining 21 papers from the period 1985-
2010. The studies he examines have mixed results: some show that the
presence of casinos increases crime while others do not. We see no need to
duplicate Walker's thorough survey here. In the following abbreviated
literature review we concentrate on a subset of papers that are either more
widely cited or have some commonality with the model we present and test
Researchers at the National Opinion Research Center at the University of
Chicago, in their report (NORC 1999) to the National Gambling Impact Study
Commission, conclude that the presence of a casino is not a statistically
significant predictor of the crime measures they examined: larcenies,
burglaries, robberies, motor vehicle thefts, and assaults. They add that the
effects are either too small to be identified or that such crime problems are
offset by other casino effects. The U.S. General Accounting Office (2000)
cites the NORC (1999) study and its own research on Atlantic City in
concluding that, for various reasons, it is difficult to tie crime rates directly to
the incidence of gambling. Reece's (2010) study of crime and Indiana casinos
suggests that the presence of casinos may actually reduce many crime rates. In
a panel data study of Michigan casino counties and crimes rates for the period
1994-2010, Falls and Thompson (2014) find that the existence, size, and age
of casinos have little or no impact on county crime rates for robbery, burglary,
larceny; motor vehicle thefts appear to decline when a casino is present.
Other researchers have demonstrated a positive relationship between
crime and the presence of a casino. Gazel, Rickman, and Thompson (2001)
employ a county level panel data set for Wisconsin and find some increases in
crime related to the presence of casinos. In perhaps the most widely cited
paper on this topic, including in the popular media (see, e.g., Dissell, 2011),
2014, 8 2
Grinols and Mustard (2006) examined all counties in the U.S. over the 1977-
1996 period and concluded that the presence of a casino does affect some
crime rates and does so with a lag. They find that casinos increase a variety of
crimes and that this impact strengthens in the years following the
establishment of a casino. Hyclak (2011) finds higher rates of car thefts and
robberies on college campuses (in four Midwestern states) within 10 miles of
a casino.
The mixed nature of prior results seems to hold for the more specific case
of violent crimes. (In section 3 below we discuss possible theoretical
connections between casinos and violent crimes and suggest that the
connection is less clear than for property crimes.) Gazel, Rickman, and
Thompson (2001) find in some model specifications that an aggregated
measure of murder, rape, and assault is increased by the presence of a casino
in a county, although they do not look at these crimes separately. Grinols and
Mustard (2006) find that assaults and rapes increase with somewhat of a lag
after a casino opens, while murder rates do not rise. Wheeler, et al. (2008),
find that Australian ―non income-generating‖ crimes (essentially all crimes
other than property or drug crimes) are not affected by gambling on electronic
gaming machines, which is as the authors expect. Reece (2010) finds that
assaults actually decrease in the years on either side of a casino opening.
Walker has authored a series of papers (2007, 2008a, 2008b, 2008c, 2010)
analyzing casino-crime research, two of which (Walker 2008b and 2008c) are
part of a critical exchange with Grinols and Mustard (2008a and 2008b) in
which these authors address three criticisms of casino-crime research. These
criticisms are 1) not including a variable related to the scale of casino
operations in addition to a dummy variable for the mere presence of a casino;
2) the quality of U.S. crime data generally; and 3) the correct measure of
population to use in calculating crime rates. We consider each of these in turn.
If casinos are in fact associated with higher crime levels, the larger the
scale of operations of a casino, either in absolute terms or relative to the size
of the host community, the greater its impact would be on various crime rates.
One might also argue that larger casinos represent greater employment
opportunities (i.e., legal opportunities for earning income) and would
therefore lead to reduced crime. Most econometric studies of the casino-crime
link use only a dummy variable to merely indicate the presence of a casino,
without accounting for the amount of gambling activity in an area. Two
exceptions are Wheeler et al. (2008) and Reece (2010). Wheeler, et al. (2008),
in a cross-section study of South Australia, found that rates of "income
generating" crimes are positively related to the level of expenditures (wagers)
on gaming machines in hotels and clubs (i.e., not in casinos). Reece (2010)
finds that the turnstile count essentially the number of patrons visiting a
casino is negatively related to the six crimes studied, although the effect is
small. In this paper we include a revenue-based measure of the scale of
gambling activity as well as the more typical dummy variables associated
with the mere existence of a casino in a given area; Section 3 below has more
detail on the data we use here.
There are many reasons to question the accuracy of U.S. crime data, but
we are not convinced that such inaccuracies have different effects on counties
with casinos and those without. Nevertheless, these concerns warrant some
discussion. First it is important to understand the difference between the "raw"
crime data provided by the Federal Bureau of Investigation (FBI) in its annual
Uniform Crime Reports, which is reported by the FBI as stated by reporting
police agencies, and the data available through the University of Michigan's
Inter-University Consortium for Political and Social Research (ICPSR). Many
researchers of the casino-crime link in the U.S. use the ICPSR data, including
Reece (2010), who provides a good discussion of its potential problems. An
important feature of this data is that ICPSR makes revisions to the data to
adjust crime numbers to full year values when agencies report less than a full
year of crime data. ICPSR cautions against comparing data prior to 1994,
when it changed its adjustment method, to earlier data. We avoid this problem
altogether by starting with the FBI's raw data files rather than ICPSR's data3,
and by beginning our study in 1994.
A third issue raised by Walker concerns the calculation of crime rates,
typically expressed as crimes per 100,000 population, from the number of
crimes: how important is it to include visitors to a given county in the
denominator of the crime rate? One version of the story underlying the casino-
crime link is that casinos attract out-of-county visitors who are victims of (and
perhaps committers of) crimes that would not have occurred in the absence of
the casino. Thus, it is reasoned, the true crime rate calculation should include
visitors in the rate's denominator, and the failure to do so makes the casino
counties' crime rates appear larger than they should. (The reader is referred to
the exchange between Walker 2008b and 2008c and Grinols and Mustard
2008a and 2008b for a lengthy discussion of this issue.) Miller and Scwhartz
(1998) discuss this issue in the context of Atlantic City, which did in fact
experience a large influx of visitors upon its legalization of gambling.
While we agree in principle that crime rates ought to be calculated with
visitors somehow included in a county’s population, we do not believe the
failure to do so causes serious problems. Even if one could obtain data on the
number of visitors and their average length of stay by county-year, it must be
the case that said visitors are visiting from somewhere else. This would
presumably require some sort of downward population adjustment in the
calculation of crime rates in those ―source of visitors‖ counties.
But visitors to a casino host county make up only part of the storywhat
about casino county residents visiting non-casino counties? In order to have a
completely accurate population plus visitors figure for the denominator of a
3 As described below in Section 3, however, we adopt portions of the ICPSR method
of creating full year data out of partial year crime reports contained in the raw FBI
UCR files.
2014, 8 2
crime rate, it would be necessary to have information on net visitors the
number of visitors to county A from outside the county less the number of
county A residents visiting outside their home county, on average over a
given year. Furthermore, it is difficult in the absence of good tourism data to
distinguish visitors whose main purpose in visiting is to gamble at a local
casino from those whose visit would have occurred even in the absence of a
casino a key point when attempting to determine the impact of a casino on
crime rates. The case of Michigan is illustrative. Many of Michigan’s tribal
casinos are located in the Upper Peninsula, an area attractive for its summer
and winter recreation opportunities well before the advent of legal Class III
gambling. We also wonder if the spread of casino gambling has made these
businesses less dependent on large numbers of outsiders who come strictly to
gamble (e.g., Atlantic City) and more so on local residents, which would
make the omission of visitors from the denominator of a crime rates less
Reece (2010) approaches the visiting population issue by first noting that
visitors often stay overnight and then examining the role of the number of
hotel rooms in the casino-crime link. It may be that visitors using hotel rooms
cause or attract crime, not casinos per se any source of increased visitors
would lead to increased crime. Reece, using county-level data for Indiana,
first finds that the opening of a casino in a county causes (with a lag) an
increase in the number of hotel rooms. Then, in Reece's crime equations, hotel
rooms have negative and often significant (though relatively small)
coefficients, implying that increased hotel rooms (visitors) slightly decreases
crime rates. We would argue, however, that casinos are the true cause of the
change in crime rates, either positive or negative, if in fact it is the existence
of the casinos that lead to additional hotel rooms and the visitors who stay in
A mention of the more general literature on crime determinants is in order
since we have used it to guide our choice of some control variables to include
in our regressions. Becker’s (1968) work on the economics of crime suggests
one set of the other (i.e., not casino related) factors that determine the extent
of crimes of various types. Controlling for these other influences on crime is
important, and Becker’s insights suggest some control variables to include.
We should expect the number of crimes committed to depend on, among other
things, the expected value of the crime to the criminal in comparison to that of
legal ways of achieving the same end. In the specific case of property crimes,
the criminal’s goal is to gain income, and the expected value of that income
depends on the availability of valuable items to steal as well as on the
probability of being caught and the associated punishment. For violent crimes,
the topic of the present work, economic incentives may be less clear.
Empirical studies of crime should include some measure(s) of the
deterrence activities of society (such as the number of police), demographic
factors such as population density and the age profile of the population. While
they may be less likely to influence the crimes under consideration here than
they would affect property crimes, we will account in our model for the
opportunities for and the potential payoffs to successful crimes (e.g., assaults
committed in conjunction with robberies which are in turn affected by the
presence of gaming-generated cash) along with the legal alternative sources of
income available to potential criminals (economic conditions in the area).
Section 3 below contains a discussion of the economic, demographic, and
other control variables used herein.
We note that casinos and related activities are also sources of jobs,
particularly in areas such as the rural Upper Peninsula of Michigan that
have few other bases for economic activity. To the extent that jobs in casinos
provide legal alternatives to crime, casinos could arguably help reduce crime.
Anderson (2013) finds that American Indians on reservations with large or
medium-sized casinos had increased income and lower poverty rates during
the 1990s relative to tribal members on reservations with smaller or no
casinos. Since all but the Detroit casinos in Michigan are owned by Native
American tribes, this income effect could result in a negative relationship
between crime rates and a casino’s presence for the counties and casinos in
The empirical model examined here rests on the theoretical relationships
developed above. We estimated a random effects model with the following
general form:
rit = α + Citβ + Xitγ + ei + vit (1)
where the rit are the crime rates (crimes per 100,000 population) for county i
in year t; the Cit are variables having to do with the presence of a casino in
county i in year t; the Xit are control variables for county i in year t; and the α,
β, and γ are coefficients to be estimated. In (1) ei + vit is a residual in which
ei is a county specific residual that differs across counties but is constant over
time for any specific county, while vit is a residual with the usual properties.
Our dataset, which contains annual values for the variables, covers all 83
Michigan counties and the years 1994-2010. One reason the random effects
model was used is that it is reasonable to assume that the county specific
component of the error term is distributed independently of the explanatory
The first group of variables in Table 1 includes the four different crimes
for which we estimated equations. Murder, rape, aggravated assault, and arson
are all included as Index Crimes for FBI crime rate calculations. We obtained
our data directly from the FBI in the form of annual text files (Uniform Crime
Report Crime by County data series) consisting of tables containing the actual
2014, 8 2
reported number of crimes in each of the eight Index crime categories, listed
by law enforcement agency within each county4 of each state.
Table 1: Variable Descriptions
Variable Name
Murders per 100,000 population
Rapes per 100,000 population
Aggravated Assaults per 100,000 population
Arsons per 100,000 population
Dummy; = 1 if a county hosts a casino, 0 otherwise
Dummy; = 1 if a county is within 50 miles of a casino, 0 otherwise
A measure of the scale of gambling activity in a county-year; see the
text for complete description
REV per capita
Age of the casino’s class III gaming activities
Per capita personal income
Unemployment rate
Population density, persons per square mile
Number of total (sworn officers and civilian) police employees per
100,000 population
Percentage of the county population in each of the following groups:
white, black, male, 15 to 19 years old, 20 to 24 years old, 25 to 29
years old, 30 to 34 years old, 35 to 85 years old
These files consist of the raw collected data, unadjusted by the FBI, along
with the portion of the county population associated with each agency and the
number of months in that specific year that each agency reported crimes. For
agencies reporting fewer than 12 months of data, we imputed crimes largely
following the Interuniversity Consortium for Political and Social Research
(ICPSR) method, as described in Reece (2010, 148-9). On average over the
sample period, 88% of Michigan agencies (accounting for 96% of Michigan’s
population) reported 12 months of data. If from three to eleven months of data
were reported, we increased the actual numbers reported proportionally to 12-
4 An agency is an individual law enforcement organization, of one of five general
types: city or township police, county sheriff, state police reporting in that county,
tribal police, and university police forces. On average there were between 7 and 8
agencies reporting per county (an average of 615 agencies per year across the 83
counties in Michigan), with smaller (population) counties having fewer and larger
counties having more.
month equivalent figures5. That is, we doubled crime numbers reported for 6
months, multiplied by 2.4 numbers covering only five months, etc. If an
agency reported only one or two months of crime data, we set the numbers of
crimes equal to that year’s average number across agency jurisdictions in the
same population size class that reported the full 12 months during that year. If
an agency did not appear in the data for a given year, we did not impute any
crimes. This is different from the ICPSR approach, which does impute crime
for an agency that makes no report.6 Once we calculated adjusted agency
numbers, we summed crimes for agencies in a county to get the county total
for the relevant year. We then used each year’s county population figures as
reported by the U.S. Census Bureau County to calculate crime rates per
hundred thousand population.
Missing data therefore occurs in the FBI’s UCR files for one of two
reasonsagencies report fewer than 12 months of crime data in a given year,
and agencies that for some reason are not even listed in the data for a given
county-year, in effect reporting no crimes. In order to examine the impact of
these missing data on our results, we run our regressions on both the full set of
county years (1,411, or 83 counties times 17 years, or the "full sample") and
on a subset of 648 county-years (the "reduced sample"). We derive the
smaller sample by including only those county-years in which all agencies
reported 12 months of data and for which no agency was missing that was
present in other years in that county. We report results below in Section 4 for
both samples.
The theoretical basis for connecting casinos to crime is clearer for
property crimes than for the violent crimes we consider here. Property crime
rates would be higher in casino host counties if the associated increase in
visitors and items to steal provides additional targets for criminals, or because
those who steal to get money to gamble live in some proximity to a casino.
One might also (alternatively) argue that a casino’s presence would reduce
crime by increasing legal employment opportunities, making property crime
relatively less attractive. But while similar stories might be told for some
violent crimes (e.g., casinos attract additional potential rape victims, or
provide the conditions conducive to alcohol-fueled aggravated assaults, or
prompt problem gamblers to commit arson to gain insurance proceeds), the a
priori case for casino-related increases (or decreases) in violent crimes is less
compelling. Grinols and Mustard (2006, 32) present some anecdotal evidence
5 Unfortunately, this approach does not allow for seasonal variations in crime over a
year because the raw FBI UCR files state how many months of data were reported by
each law enforcement agency in each county but do not identify which months are
included. Lacking that information, there is no reason to believe that missing months
were predominately either high-crime or low-crime months.
6It is clear from a close examination of the raw UCR data files that in many cases,
other agencies in the county picked up the reporting for the missing agency for one or
more years. Imputing crimes for agencies missing in a given year may therefore
double count those crimes.
2014, 8 2
associating gambling activity with violent crime, including an account of
murders done in the course of robberies whose objective was to get gambling
money. They also argue that casinos cause increases in adult entertainment
types of services, and from that imply a likely increase in rapes. But they also
refer (p. 33) to claimed casino-related increases in prostitution activity, which
could arguably reduce the number of rapes. In our view, therefore, a
theoretical case could be made crediting the presence of casinos with
increases, decreases, or no effect on the violent crimes considered here; we
thus have no a priori expectation in this regard. More surprising than evidence
that there exists either a positive or negative relationship between casinos and
these four crimes would be a finding that such an impact is substantial.
The second group of variables in Table 1 includes variables about the
presence, size, and age of casinos in Michigan. The CAS variable takes the
value of one in a particular year if a county had one or more casinos in that
year and zero otherwise. NEAR equals one if there is a casino located within
50 miles of any portion of a county and zero otherwise7. NEAR is zero for all
casino host counties, so that it is not possible for both CAS and NEAR to
equal one in the same county in the same year.
Our measure of the scale of casino operations is REV, based on data from
the Michigan Gaming Control Board (MGCB; 2012a, b). MGCB reports two
types of revenue-based data, one for tribal casinos and another for the
commercial casinos in Detroit. Tribal casino data is the total payments made
each year by each tribe to local governments in lieu of taxes; these are the 2%
payments described above that are based on electronic gaming machine net
revenues. These amounts are reported by tribe rather than by county. If a tribe
operates only one casino, we assign the 2% payments to that casino's host
county. If a tribe operates casinos in multiple counties, we allocate total tribal
2% monies to the host counties in proportion to the number of electronic
gaming machines at each casino.
Raw revenue data reported by MGCB for the three Detroit (Wayne
County) casinos consists of the total net gaming revenues for the Wayne
County casinos attributable to gaming machines. Because this amount
apparently includes more than the revenues from the electronic games on
which the tribal 2% payments are based (―electronic games‖ appear to be a
subset of ―gaming machines‖), it is not possible to obtain precisely the same
calculation for the Detroit casinos, but we believe that the two types of
7 Although the 50-mile distance is somewhat arbitrary, the National Gambling Impact
Study Commission stated that (1999, p. 4-4) the presence of a gambling facility
within 50 miles roughly doubles the prevalence of problem and pathological
gamblers and that (p. 7-17) 85% of riverboat gamblers in Illinois lived within 50
miles of the casino in which they were gambling. In an ideal world we would be able
to determine the distance between a casino and the point of commission of a given
crime, which would allow the use of spatial regression techniques, but the only
geographic information in the UCR data is the county in which crimes are committed.
revenue related quantities are reasonable comparable figures. We multiplied
the Detroit figures by 2% to put them on the same scale as the tribal 2%
Just as we use the CAS and NEAR dummy variables described earlier to
assess the crime impacts of a casino's presence in or near a county, we also
derive a casino scale variable to reflect the level of gambling activity in or
near a given county. The REV variable for a given county therefore depends
on the revenues of casinos in that county and the revenues of casinos within
50 miles of that county’s borders. We then classified a county based on
whether 1) it was a casino host county; or 2) if there was a casino in another
county within 50 miles of the borders of the county in question; or 3) if
neither of these conditions was met. Once we made those determinations,
REV was calculated as follows:
for a host county that had no other casinos within 50 miles of its
borders, REV is equal to the actual or allocated dollars assigned to the
host county as described above;
for a host county that did have one or more other casinos within 50
miles of its borders, REV is equal to the figure assigned to that host
county plus the assigned figures for the counties with casinos that are
within 50 miles of the host county’s borders;
for a county that did not itself have a casino, REV is equal to the sum
of the assigned figures for the counties that have casinos that are
within 50 miles of the borders of the county in question. Therefore, if
a county without a casino also had no casinos within 50 miles of its
borders, REV is zero; if it had one casino within 50 miles of its
borders, REV is equal to the assigned figure from step one for that
nearby host county; and if it had more than one casino within 50 miles
of its borders, REV is equal to the sum of the assigned figures for the
counties that have casinos that are within 50 miles of the borders of
the county in question.
This approach to the calculation of REV yields many counties in the
dataset that have no casino but nevertheless have a nonzero value for REV; it
also explains why the mean of REV is higher in non-casino counties than in
host counties (see Table 2 below), as some non-host counties are within 50
miles of multiple casinos.
The size variable used in the regressions is REVPER, equal to the
calculated REV figures divided by population for the associated county-year.
And as is the case for the casino presence dummies CAS and NEAR, we are
unable to predict a priori the sign of the coefficient on REVPER. A positive
8 It should be noted that Wayne County is the only county so affected, and that there
are neither Indian casinos in Wayne County nor any tribal casino host counties within
50 miles of the Detroit casinos.
2014, 8 2
relationship between casino size and the violent crimes would result if larger
casinos attract a larger criminal element, or lead to more problem gambling.
But finding a negative relationship would not be surprising, since larger
casinos might also represent greater legal income earning opportunities for
potential criminals, thereby increasing the criminal's opportunity costs of
committing these crimes.
We also consider the age of the casino’s class III gaming activities. Four
of the tribes had relatively small bingo hall type casinos prior to having a
compact with the state of Michigan, but they did not offer class III activities.
As a casino ages its reputation grows. If this reputation is desirable from a
gambler’s viewpoint (e.g. bigger payouts) it may attract additional clientele
and potential criminals possibly resulting in a higher crime rate. This
suggests a positive coefficient. There is also the possibility that as a casino
ages and its ―newness‖ wears off, fewer customers and potential criminals are
attracted to it, suggesting a negative coefficient.
Table 1 also includes control variables that represent non-casino related
crime determinants. These include two economic indicators: per capita
personal income (U.S. Bureau of Economic Analysis) and county annual
unemployment rates (U.S. Bureau of Labor Statistics). Population density, the
total number of law enforcement employees, both sworn officers and civilians
(FBI 2012) per 100,000 population, and variables describing aspects of the
race and age structure of the population (U.S. Bureau of the Census) are other
potential influences on crime, which we have included as controls.
The economic variables (per capita personal income and the
unemployment rate) may have little if any impacts on violent crime rates,
either up or down. Generally speaking one might argue that improved
economic conditions (higher income or lower unemployment rate) increases
the potential opportunity cost associated with being caught, thereby lowering
crime. Raphael and Winter-Ebmer (2001) find that lower incomes and higher
unemployment lead to increased crime.9 Similarly, Wheeler et al. (2008) find
a positive crime-unemployment relationship. Models we tested used each of
these measures individually and together.
Population density might be expected to be positively related to crime:
urban areas are perceived to be more crime-prone than rural areas, and it is
easier for criminals to remain anonymous in heavily-populated areas. Our
population density variable is simply county population divided by county
area in square miles. We also included variables controlling for the age, race,
and sex composition of the population, as these demographic factors can also
affect crime rates. Age composition may be especially important: Levitt
(1999, 583) says ―The relationship between age and criminal
involvement….is one of the most well-known and robust relationships in all
9 Interestingly, these authors (p. 266) expected a positive relationship, hypothesizing
that higher incomes lead to purchases of more criminogenic commodities such as
drugs and guns.
of criminology….‖ Raphael and Winter-Ebmer (2001, p.266) refer to the
―well-documented age-crime profile.‖ Simply put, younger persons are more
likely to commit crimes, and our a priori expectation is that higher
percentages of the population in younger age brackets increases crime rates.
Table Two: Summary Statistics
All Counties
Casino Counties
Non Casino Counties
Std. Dev.
Std. Dev.
Std. Dev.
per 100,000
per 100,000
per 100,000
per 100,000
1 = yes, 0 =
1 = yes, 0 =
dollars per
percent (0-
persons /
per 100,000
percent (0-
percent (0-
percent (0-
percent (0-
percent (0-
percent (0-
percent (0-
percent (0-
83 total counties; number of casino counties (CAS=1) increases from 8 in 1994 to 17 in 2010
County years with CAS=1m 223 county years of 1411 county years
One might expect that increased numbers of police would tend to increase the
probability of arrest, thus increasing the expected cost of crime to a potential
criminal; the standard prediction is that increased numbers of police would
2014, 8 2
negatively affect crime rates. Many studies, however, show a positive
coefficient on measures of police presence. Marvell and Moody (1996) cite
several earlier works, many of which include a positive police-crime
relationship. Cornwell and Trumbull (1994) find a positive relationship, but it
becomes statistically insignificant when they use a 2SLS approach. More
recently, Wheeler et al. (2008) found a positive relationship between police
"charge rates" (similar to arrest rates) in a study examining the impact of
gambling on crime in South Australia. Simultaneity is likely, with higher
crime rates leading to increases in the number of police on the streets, as
governments respond to public pressure as crime increases, and in higher
arrest rates, even with no increase in police numbers. This effect may be
stronger for some crimesmore visible, violent crimes, such as those under
consideration in this studythan for other crimes, such as property crimes.
The U.S. General Accounting Office (2000, 37) suggests that a positive
relationship can be explained as follows: ―reported crime increased because
more police employees were available to uncover crimes.‖ While it would not
be surprising to see the number of arrests increase by such logic, it is more
difficult to see how additional police would cause more crimes to be reported.
As explained below, we used a 2SLS model in the present work to anticipate
this potential endogeneity.
Table 2 contains the means and standard deviations of the variables in
Table 1, for the entire 83-county sample, for the casino county segment (223
out of 1,411 county-years), and for the non-casino segment (1,188 county-
years). An examination of the raw means of the crime rates tells us little about
those rates in casino vs. non-casino counties; murders are slightly lower on
average in non-casino county-years, while the average number of the other
three crimes is higher in county-years without casinos.10
We estimated Equation (1) in log linear form, primarily because the
resulting coefficients are elasticities, which allows for a clearer understanding
of the importance of individual crime determinants beyond their statistical
significance.11 There is a slightly different interpretation of the coefficients on
dummy variables; multiplying the coefficients by 100 yields the percentage
change in the crime measure as the dummy changes from zero to one.
We used an instrument for the police variable, police employment per
hundred thousand persons, to address the possible endogeneity (discussed
above) between police numbers and crime levels. We ultimately chose total
10 Although not shown in Tables 1 or 2, we also included in the regressions dummy
variables for the years 1995-2010, with 1994 serving as the base year, to pick up
statewide crime trends affecting all counties.
11 Results from equations based on the actual values of the variables (i.e., not natural
logs) were similar.
county land area and total county population as instruments after exploring
several possible combinations.
Table 3 : Full Sample
Dependent Variable
Aggravated Assault
Wald Chi Sq
*, **, *** indicates coefficient is significant at the 10%, 5% and 1% levels, respectively. Numbers in
parentheses are bootstrap standard errors based on 50 replications.
In Table 3 we present results based on the full sample of 1,411 county-
years, consisting of all 83 Michigan counties for the years 1994 through 2010.
The figures in parentheses are bootstrap standard errors which were estimated
using 50 replications. One pair of models for each crime is shown, one of
which uses PERSINC (personal income per capita) as the independent
2014, 8 2
variable capturing economic conditions while the other adds UNEMRT (the
unemployment rate). We first discuss the results crime by crime and then we
offer a comparison of the impacts of the independent variables across crime
types. Our primary interest is in the effects of the casino related variables
CAS, NEAR, REVPER, and CASAGE, but we also discuss the influence of
our control variables. After addressing results from the full sample, we will
provide a similar discussion of regressions run on the reduced sample of
The murder rate is unaffected by any of the casino related variables. It is
reduced by higher personal income per capita, higher population density, and
a larger percentage of the population aged 20 to 24, with the latter two
findings being contrary to our expectation. We also see in Table 3 a positive
relationship (with an elasticity of about 1.7) between the murder rate and the
number of police, although we doubt that having more police does in fact
cause more murders. As discussed earlier, it is more likely that an increase in
crimes leads to public pressure to increase police numbers, especially for a
crime like murder that receives extensive coverage in the news media. Murder
is the only crime we examine that shows a positive and statistically significant
relationship with per capita police employment.
The age of the casino is the only casino related variable affecting the rate
of rapes, with the rate falling slightly (the elasticity is less than 0.1) as a
casino ages. Only two of our control variables have an impact, with increases
in either the unemployment rate or the proportion of the population aged 15 to
19 causing rapes to increase.
Aggravated assaults by far account for the largest proportion (nearly 80%)
of the four crimes considered here. In each assault specification of Table 3,
no coefficients for the casino related variables are statistically significant.
Assaults increase with increases in population density, and having a larger
white percentage of the population reduces the assault rate, with a relatively
large elasticity of about 2.2. No other variables are significant in either assault
Arson, the only one of our crimes that could be thought of as a property
crime committed in order to gain financially, falls as a casino ages, but none
of the other casino related variables has a statistically significant impact.
Arson decreases with increases in per capita personal income and does so
dramatically (the elasticity exceeds 3.0) with increases in the white proportion
of the population. Increases in the proportion of the population in the 15 to 19
and 25 to 29 age groups cause reasonably substantial increases in arson, with
elasticities in the 0.64 to 0.94 range.
We also performed the same regressions in models based on the reduced
sample12 of 648 county-years. The results are in Table 4. We do see some
differences when we drop county-years that might have substantial amounts
12 Recall that the reduced sample contains only those county years in which all
agencies were present and all of them reported a full 12 months of data.
of missing crime reports; we discuss these differences here. In general we find
more casino-related variables having a statistically significant impact, and
there are some minor changes in the significance of some of the control
In the case of murder, while the income variable loses its statistical
significance, the NEAR variable, which indicates the presence of a casino
within 50 miles of the county border, becomes statistically significant. The
coefficient is approximately -.38, indicating a drop of 38% in the murder rate
when there is a casino nearby. Once again, the (instrumented) police variable
is positive and statistically significant; the elasticity is over 1.6.
In the rape equations, both casino age and the unemployment rate lose
their statistical significance in the smaller sample. The positive coefficient on
the 15 to 19 age proportion loses its statistical significance, while the
coefficient on the proportion of 20 to 24 year-olds is negative and significant.
Moving to the smaller sample leads to the largest changes for aggravated
assault. Both CAS and NEAR become statistically significant and are
relatively large, with the assault rate in a casino host county dropping over
40% relative to non-casino counties. Having a casino within 50 miles
decreases the assault rate by about 27%. The coefficient on the casino SIZE
variable also becomes statistically significant, but it is small, showing an
elasticity of less than 0.1. We would suggest that this combination of
coefficients means that the presence of casinos reduces the rate of aggravated
assault, but the effect is reduced slightly as a casino becomes larger. The
reduced sample equations for assault also suggest that a larger proportion of
the population aged 15 to 19 tends to increase this crime rate.
The arson results also change somewhat when we move to the reduced
sample. The casino age variable is no longer statistically significant, but
population density becomes significant and is negative. The personal income
and the two age group variables that were significant in the full sample no
longer are in the reduced sample. The proportion of the population that is
white remains significant and increases in absolute value.
Generally speaking, we find no evidence that the presence of a casino
increases violent crime rates in the host county or in nearby counties. Indeed,
we find that the rates of aggravated assault may actually decline with the
presence of a casino in a county, as may the murder and assault rates when a
casino is present in a nearby county. Why these rates would decline is by no
means clear. One possible explanation is that, given the higher risk level to
the criminal in terms of the potential for a long prison sentence associated
with crime, the additional employment opportunities offered by a casino
provides a much more attractive legal alternative to a potential criminal. Only
in the case of assault does the size of a casino increase the crime rate and even
then by only a very small amount.
2014, 8 2
Table 4 : Reduced Sample
Dependent Variable
Aggravated Assault
Wald Chi Sq
*, **, *** indicates coefficient is significant at the 10%, 5% and 1% levels, respectively. Numbers in parentheses are
bootstrap standard errors based on 50 replications.
The results for the more general control variables for crime are mixed.
The two economic variables have little to no impact these crime rates.
Increasing population density generally (and perhaps unexpectedly) decreases
these violent crime rates, except for aggravated assault, the rate for which
decreases. The results for the instrumented police variable suggest that
increased numbers of police seem to have no impact on any of these crimes
except for murder. We find that murder rates and police per capita are
positively related, but we think this is simply evidence of a persistent
simultaneity problem: more police get hired when murder rates are higher.
Results for the age, race, and sex variables, included mainly for control
purposes, could reasonably be characterized as unsurprising.
The regression results reported here indicate generally that the presence or
size of a casino in or near a given Michigan county does not affect violent
crime rates. The largest such impact, in which crime rates actually fall in the
presence of a casino, is aggravated assault. The size of the casino has a
positive but quite small coefficient in only the aggravated assault equation for
the reduced sample case.
Our finding that in Michigan the presence of a casino has no impact on or
perhaps even slightly reduces these crime rates is largely consistent with
NORC (1999) and Reece (2010), while we fail to find evidence confirming
the results of other studies (Gazel, Rickman, and Thompson 2001; Grinols
and Mustard 2006) showing that casinos increase crime rates. Reece’s (2010)
study, which focused on Michigan's neighboring state of Indiana13, found that
the presence of a casino decreases most crime rates (with burglaries being the
one exception), and that an increase in the number of patrons, measured by
casino turnstile counts, reduces larceny and motor vehicle theft rates. In
contrast, our measure of size based on casino revenues was found to have a
small but statistically significant positive relationship with aggravated assault
in the reduced sample case only, with the three other crimes we examined
being unaffected by casino size. Taken together, Reece (2010) and the present
study indicate that, at least in these two Midwestern states, concerns about
casinos leading to increased crime rates are very likely misplaced.
An obvious and potentially useful extension of this work would be to add
models for other crimes, including embezzlement and fraud, which might be
more closely related to problem gambling by individuals (i.e., crime as a
source of money with which to gamble), and morals offenses (e.g.,
prostitution or alcohol-related crimes), that may be complementary to gaming
activities. The results of such analyses would help local communities
receiving 2% payments (or, in Wayne County, casino related taxes) determine
how to best spend the funds. That is, are these communities properly directing
the funds into addressing any problems that arise from the presence of casino
gambling? Should more police be added because of additional crime? (The
answer appears to be ―no,‖ at least for the four violent crimes examined here.)
Should more social workers be added because of the negative effects of
problem gambling on families? Should more roads be built to cope with
increased traffic levels? And while we suspect the broad political debate over
the desirability of casino gambling will never be resolved, the results of this
and future analyses can help gain some insights into the connection between
casinos and crime and other social impacts of gambling. This is important
because of the recent growth of casino gambling across the U.S., not to
13 There is even some competition among casinos in the two states near their common
2014, 8 2
mention its likely future growth as states increasingly turn to legalized
gambling in their continuing efforts to find new revenue sources.
American Gaming Association (AGA), 2012. Taken from
Anderson, Robin J. 2013. ―Tribal Casino Impacts on American Indians Well-Being:
Evidence From Reservation-level Census Data.‖ Contemporary Economic
Policy, v. 31 no. 2, April 2013, 291-300.
Becker, G. S. 1968. ―Crime and Punishment: an Economic Approach.‖ Journal of
Political Economy, v. 76, 169-217
Cornwell, C., and W. N. Trumbull. 1994. ―Estimating the Economic Model of Crime
with Panel Data,‖ Review of Economics and Statistics, May, 76:2, 360-366.
Dissell, Rachel. 2011. "It's no sure bet that casino will bring an increase in crime to
Cleveland," online Cleveland Plain Dealer, May 29, 2011.
Erickcek, G., B. Timmeney, and B. Watts, W.E. Upjohn Institute for Employment
Research. 2008. ―Calhoun County Casino Baseline Study Committee
Recommended Indicators and Baseline Data Report.‖
Falls, Gregory A., and Philip B. Thompson, 2014. ―Casinos, casino size, and crime: A
panel data analysis of Michigan counties.‖ Quarterly Review of Economics and
Finance, Vol 4 Iss. 1, 123-132. Available online at:
Federal Bureau of Investigation Crime Statistics, various years, taken from and from CD-based files obtained
directly from the FBI.
Gazel, R. C., D. S. Rickman, and W. N. Thompson. 2001. ―Casino Gambling and
Crime: A Panel Study of Wisconsin Counties.‖ Managerial and Decision
Economics, v. 22, 65-75.
Grinols, E. L., and D. B. Mustard. 2006. ―Casinos, Crime, and Community Costs.‖
Review of Economics and Statistics, Feb, v. 88 no. 1, 28-45.
_______. 2008a. "Correctly Critiquing Casino-Crime Causality." Econ Journal
Watch, Vol. 5 No. 1, January 2008, 21-31.
_______. 2008b. "Connecting Casinos and Crime: More Corrections of Walker."
Econ Journal Watch, Vol. 5 No. 2, May 2008, 156-162.
Hyclak, T. 2011 ―Casinos and Campus Crime,‖ Economics Letters, v. 112, 31-33.
Levitt, S. 1999. ―The limited role of changing age structure in explaining aggregate
crime rates,‖ Criminology, 37:3, 581-597.
Margolis, J. 1997. ―Casinos and Crime: An Analysis of the Evidence.‖ Report
prepared for the American Gaming Association.
Marvell, T. B., and C. E. Moody. 1996. ―Specification Problems, Police Levels, and
Crime Rates,‖ Criminology, 34:4, 609-646.
Michigan Gaming Control Board. 2012a. "2% Payments to Local Governments,
7/24/2012." Spreadsheet in *.pdf format, taken from
Michigan Gaming Control Board. 2012b. ―Detroit Casino Revenues‖. Taken from,4620,7-120-1380_57134_57590---,00.html
Miller, W. J., and M. D. Schwartz. 1998. ―Casino Gambling and Street Crime.‖
Annals of the American Academy of Political Science, March, v. 556, 124-137.
National Opinion Research Center, U. of Chicago. 1999. ―Gambling Impact and
Behavior Study: Report to the National Gambling Impact Study Commission.‖
Raphael, S., and R. Winter-Ebmer. 2001. ―Identifying the Effect of Unemployment on
Crime,‖ Journal of Law and Economics, April, 44:1, 259-283
Reece, William S. 2010. "Casinos, Hotels, and Crime." Contemporary Economic
Policy, Vol. 28 No. 2, April 2010, 145-161.
U.S. Bureau of Economic Analysis, Survey of Current Business, URL and similar
for various years.
U.S. Bureau of Labor Statistics, Current Population Survey, URL , downloaded various dates 2008-2010.
U.S. Bureau of the Census, various county data reports, URL,
downloaded various dates 2008-2010.
U.S. General Accounting Office. 2000. ―Impact of Gambling: Economic Effects More
Measurable Than Social Effects.‖ Report to the Honorable Frank R. Wolf, House
of Representatives.
Walker, D. M. 2007. ―Problems in Quantifying the Social Costs and Benefits of
Gambling,‖ American Journal of Economics and Sociology, July, 66:3, 609-645.
_______. 2008a. "Evaluating Crime Attributable to Casinos in the U.S.: A Closer
Look at Grinols and Mustard's 'Casinos, Crime, and Community Costs'," Journal
of Gambling Business and Economics, Vol. 2 No. 3, 23-52.
_______. 2008b. "Do Casinos Really Cause Crime?" Econ Journal Watch, Vol. 5 No.
1, January 2008, 4-20.
_______. 2008c. "The Diluted Economics of Casinos and Crime: A Rejoinder to
Grinols and Mustard's Reply." Econ Journal Watch, Vol. 5 No. 2, May 2008,
_______. 2010. "Casinos and Crime in the USA." Chapter 19 of Benson, Bruce L.
and Paul Zimmerman, eds., Handbook on the Economics of Crime, Northampton,
MA: Edward Elgar, November, 2010.
Wheeler, S.A., D. K. Round, R. Sarre, and M. O'Neil. 2008. ―The Influence of
Gaming Expenditure on Crime Rates in South Australia: A Local Area Empirical
Investigation.‖ Journal of Gambling Studies, v. 24, 1-12.
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
Although legalized gambling, and in particular casino gambling, has become an increasingly important American leisure activity, it has not escaped extensive controversy. Among the many evils forecast for communities that open casinos is a major increase in street crime. This article will review what we know about the relationship between street crime and casino gambling.
Full-text available
In this paper, we analyze the relationship between unemployment and crime. Using U.S. state data, we estimate the effect of unemployment on the rates of seven felony offenses. We control extensively for state-level demographic and economic factors and estimate specifications that include state-specific time trends, state effects, and year effects. In addition, we use prime defense contracts and a state-specific measure of exposure to oil shocks as instruments for unemployment rates. We find significantly positive effects of unemployment on property crime rates that are stable across model specifications. Our estimates suggest that a substantial portion of the decline in property crime rates during the 1990s is attributable to the decline in the unemployment rate. The evidence for violent crime is considerably weaker. However, a closer analysis of the violent crime of rape yields some evidence that the employment prospects of males are weakly related to state rape rates. Copyright 2001 by the University of Chicago.
Growth in legal gaming in the United States over the past quarter century or so is well-documented. One important factor fueling this growth was the passage of the Indian Gaming Regulatory Act of 1988, which permitted Native American tribes to establish, under agreements or “compacts” with the states in which they are located, casinos offering what is known as Class III gaming: slot machines, blackjack, roulette, and other games. Since the passage of the Act, there have been 21 Native American casinos established in Michigan. Also, three non-Native American casinos opened in Detroit in 1999 and 2000. This growth in the number of casinos has sparked a wide-ranging debate over the social and economic impacts of casino development. The purpose of this research is to focus on the crime issue in the broader casino debate. We investigate the impact of these Michigan casinos on the rates of burglary, robbery, larceny and motor vehicle theft (property crimes) in casino host counties as well as in nearby counties. We employ a panel data set with annual observations on all 83 Michigan counties for the period 1994–2010. The dataset includes crime rates taken from the FBI crime data series, variables for the presence of a casino in a county or in a nearby county, the scale of a casino's operations as measured by revenues, and a variety of control variables suggested by the broader literature investigating the factors that determine crime rates generally. Our results suggest that in most cases the property crime rates studied are not affected by the presence or size of a casino in a county or in a nearby county. The largest such impact, which is negative, is for motor vehicle theft. The size of a casino does have a small positive effect on the motor vehicle theft rate.
After the passage of the Indian Gaming Regulatory Act in 1988, tribal gaming revenues increased dramatically. Using a differences‐in‐differences methodology with 1990 and 2000 census data, this study finds that American Indians (AI) on gaming reservations experience a 7.4% increase in per capita income and reductions in both family and child poverty rates relative to AI on non‐gaming reservations. Large and medium casinos are associated with changes in well‐being while smaller casinos are not. These results are sensitive to the inclusion of the Navajo reservation, a large non‐gaming reservation with increased income during the 1990s.
Changes in the age structure are shown to have a limited impact on aggregate crime rates. Even the dramatic transformation of the age distribution accompanying the baby boom shifted crime rates by no more than 1 % per year. Projected changes in the age distribution between 1995 and 2010 will lead to slight declines in per capita crime rates. These results are at odds with recent predictions of an impending demographically driven crime wave. Such predictions have focused exclusively on a rise in juvenile crime and ignored the offsetting decreases among adults.
This work examines the hypothesized positive relationship between casinos and local crime rates. Analysis of reported crime data for 173 residential colleges and universities in four Midwestern states suggests that robberies and motor vehicle thefts, but not burglaries, are significantly higher in number for campuses located within 10 miles of a casino.
"This paper examines the links among casinos, hotels, and crime using Indiana's counties for 1994-2004. In estimating casinos' impacts, I introduce a measure of casino activity in addition to variables related to the timing of casino opening. I test whether or not the number of hotel rooms affects crime rates. Increased casino activity reduces crime rates except for burglary, where crime rates rise after a lag. Leaving out a measure of casino activity appears to create a serious specification error. Finally, including problem crime data plagued by incomplete reporting affects the estimated impact of casinos on crime." ("JEL" R11, L83) Copyright (c) 2009 Western Economic Association International.
I. Introduction Since the turn of the century, legislation in Western countries has expanded rapidly to reverse the brief dominance of laissez faire during the nineteenth century. The state no longer merely protects against violations of person and property through murder, rape, or burglary but also restricts "dis­ crimination" against certain minorities, collusive business arrangements, "jaywalking," travel, the materials used in construction, and thousands of other activities. The activities restricted not only are numerous but also range widely, affecting persons in very different pursuits and of diverse social backgrounds, education levels, ages, races, etc. Moreover, the likeli­ hood that an offender will be discovered and convicted and the nature and extent of punishments differ greatly from person to person and activity to activity. Yet, in spite of such diversity, some common properties are shared by practically all legislation, and these properties form the subject matter of this essay. In the first place, obedience to law is not taken for granted, and public and private resources are generally spent in order both to prevent offenses and to apprehend offenders. In the second place, conviction is not generally considered sufficient punishment in itself; additional and sometimes severe punishments are meted out to those convicted. What determines the amount and type of resources and punishments used to enforce a piece of legislation? In particular, why does enforcement differ so greatly among different kinds of legislation?
Previous attempts at estimating the economic model of crime with aggregate data relied heavily on cross-section econometric techniques and, therefore, do not control for unobserved heterogeneity. This is even true of studies that estimated simultaneous equations models. Using a new panel data set of North Carolina counties, the authors exploit both single and simultaneous equations panel data estimators to address two sources of endogeneity: unobserved heterogeneity and conventional simultaneity. Their results suggest that both labor market and criminal justice strategies are important in deterring crime but that the effectiveness of law enforcement incentives has been greatly overstated. Copyright 1994 by MIT Press.