I. INTRODUCTION
In recent decades, casinos have spread to many states in which they
had previously been illegal. Nevada legalized casino gambling in 1931.
In having legal casinos, it remained alone among the states until New
Jersey legalized Atlantic City casinos in 1976. Recently, a total of 12
states have allowed commercial casinos (American Gaming Association,
2009). Other states have allowed gaming devices at racetracks. Following
action by the U.S. Supreme Court in 1987 and the Congress in 1988,
Native American casinos have opened in 28 states (American Gaming
Association, 2009). This spread of casino gaming has been very
controversial. The public, regulators, and researchers in health and
social sciences have been interested, among other things, in the effects
of casino gambling on crime.
A few researchers have empirically investigated the connection
between the opening of new casinos and local area crime rates. Two of
the most important of these are Evans and Topoleski (2002) (E&T) and
Grinols and Mustard (2006) (G&M). (1) Using a large panel of U.S.
county data on casino openings and crime rates, G&M test the
hypothesis that casinos cause crime. They estimate the parameters of a
fixed-effects panel model using all 3,165 U.S. counties for the period
1977 through 1996. Their dependent variables are numbers of reported
offenses in each of the following seven categories: aggravated assault,
rape, robbery, murder, larceny, burglary, and motor vehicle theft. The
source for these crime rates is the Federal Bureau of
Investigation's (FBI) Uniform Crime Reporting (UCR) program, which
the Interuniversity Consortium for Political and Social Research (ICPSR)
makes available to researchers. (2) G&M find that casino openings
cause county crime rates in all categories except murder to rise after a
lag of a few years. E&T also use UCR data to estimate a similar
model to test the effects of opening Native American casinos on county
crime rates. They find statistically significant positive lagged effects
of new Native American casinos on crime.
In this paper, I reexamine the link between new casinos and crime
rates. Here, I use the 92 counties of Indiana over the period 1994
through 2004. During this period, Indiana opened its first commercial
casinos, including 10 casinos in seven counties. (Indiana has no Native
American casinos.) Also, Indiana's casinos have a readily available
measure of casino activity, the turnstile count of patrons, which we can
use in addition to the often used dates of casino opening. Finally, I
have created a database of the number of hotel rooms for each Indiana
county for each of these years. I add this variable to the crime
equations. The annual series of data on hotel rooms by county allows a
test of whether or not the opening of other public accommodations
besides casinos, specifically hotels, affects crime rates.
II. THEORY OF CASINOS AND CRIME
The economic model of crime proposed by Becker (1968), and
investigated by Ehrlich (1973), Sjoquist (1973), Cornwell and Trumbull
(1994), Raphael and Winter-Ebmer (2001), Gould, Weinberg, and Mustard
(2002), Levitt (1998, 2004a, 2004b), Ihlanfeldt (2006), and many others
asserts that potential criminals are utility maximizing agents who
allocate time between legal work and crime based on the potential
benefits and costs of each. (3) To the extent that this is true,
increasing the opportunity cost of criminal behavior would reduce crime.
Increasing the probability of arrest and conviction along with more
severe penalties would increase the opportunity cost of crime. Also,
increasing legitimate work opportunities for potential criminals or the
wages for that work would increase the opportunity cost of crime. As
explained below, this theory has testable implications for the link
between new casinos and crime.
Following the economic model of crime, new casinos could either
increase or decrease local crime rates. If introducing new casinos
increases job opportunities in the area, new casinos could, by
increasing the opportunity cost of crime, reduce crime rates. Also,
increasing activity within existing casinos could have the same effect.
Many studies have supported the idea that better local labor market
conditions reduce local crime rates and have found this result using
various levels of aggregation. Raphael and Winter-Ebmer (2001) find that
higher state unemployment rates lead to higher property crime rates.
Gould, Weinberg, and Mustard (2002) find that movements of wages of
unskilled men explain a large fraction of the movements in county crime
rates. Ihlanfeldt (2006) finds that greater neighborhood employment
opportunities for young males lower neighborhood crime rates. Levitt
(2004a, pp. 170-171), however, questions the importance of the direct
link between wages or unemployment rates and property crime rates.
The economic model of crime suggests that closer proximity of
potential criminals and potential victims would increase local crime
rates. Bowes and Ihlanfeldt (2001) investigate the impact of location
and transportation costs on census tract crime rates and find that
lowering transportation costs between potential criminals and victims
(by the presence of a nearby rapid transit rail station) increases crime
rates. This locational effect suggests that new casinos could reduce
local crime rates by displacing crime-ridden locations with new
construction. Alternatively, new casinos could lower the opportunity
cost of crime by bringing new criminal opportunities to the local area.
If this effect were important, new casinos would increase crime rates.
Another consideration that may be important in investigating the
link between casinos and crime is that casino openings are typically
followed, often with a lag, by the opening of hotels, entertainment
venues, retail shops, and other public accommodations. These new public
accommodations could raise or lower the opportunity cost of crime in the
same ways discussed above for new casinos. Thus, because new casinos and
other new public accommodations are linked, leaving these other new
public accommodations out of the model explaining crime rates could lead
to a specification error. Then finding that new casinos increase local
crime rates after a lag of a few years could actually be the finding
that casinos cause hotels and other public accommodations with a lag and
the increased presence of these public accommodations causes a
contemporaneous increase in crime rates. (4)
As with new casinos, the activities associated with new hotels and
other public accommodations could reduce crime rates by providing
increased legal employment opportunities, by displacing crime-ridden
locations, or through some other mechanism. Or, they might increase
crime by bringing together potential victims and potential criminals.
Tourists visiting casinos or other sites may be particularly vulnerable
to crime because they often carry large amounts of cash and other
valuables and because they are outside their normal home territory. (5)
Thus, the economic model of crime suggests that the opening of new
casinos, the level of activity at existing casinos, and the extent of
other public accommodations may help explain local crime rates. Here, I
use dates of opening of new casinos, leads and lags around these opening
dates, a measure of the level of casino activity, and the number of
hotel rooms in the specification explaining each county's crime
rates. Previous studies of casinos and crime have accounted only for the
introduction of casinos--they have not examined the impact of the level
of casino activity or the extent of other public accommodations. In some
cases, local areas have seen substantial new activity including multiple
large casinos, new hotels, new restaurants, new entertainment venues,
and new shopping areas. In other areas, introduction of casinos may mean
that the area has a single relatively small casino. This study is the
first to empirically examine the impacts of the level of casino activity
and the level of related activity, in this case the number of hotel
rooms, on local crime rates.
It is possible that opening a new casino in Indiana increases
crimes rates in bordering states. I do not investigate that possibility
here for two reasons. First, some of the Indiana casinos are in the
Chicago area, and explaining crime rates in Chicago is beyond the scope
of this paper. Second, most Illinois counties did not report crime rates
during the period covered here, making investigation of spillovers
infeasible.
III. DATA
The UCR crime data used to examine links between casinos and crime
have severe and well known defects, discussed below. Furthermore,
researchers have lacked useful, broadly applicable measures of casino
activity. (6) Here, I test the hypothesis that new casinos caused
increases in crime rates in the counties of Indiana over the years 1994
to 2004 using the standard data sources previous investigators have used
plus a measure of casino activity and a unique annual panel data set on
hotel accommodations. The counties of Indiana are particularly well
suited for a study of the links among casinos, hotels, and crime. In
1993, the Indiana legislature created the Indiana Gaming Commission
(IGC) and authorized it to issue up to 11 riverboat casino licenses
(IGC, 1994). This Riverboat Gaming Act specified that the city of Gary
would receive the first one or two licenses. The IGC set up local
referendums on the desirability of casinos in the city or county. In the
1993 elections, four counties rejected casino referendums and five
counties and two cities passed casino referendums. The IGC then
established a sequence of locations for consideration for riverboat
casino licenses. These included four Lake Michigan cities or counties,
four Ohio River counties, and Patoka Lake in the southern interior of
Indiana. The Army Corps of Engineers, owner of Patoka Lake, subsequently
denied permission for a riverboat casino. In 2003, the legislature
approved a replacement land-based casino north of the lake in Orange
County near French Lick. At the same time, the legislature created a
Historic Hotel Preservation Commission to work with the IGC to develop
the historic resort hotels in the area (IGC, 2003).
As a result of this and subsequent activity, Indiana, which had no
casinos before 1995, opened 10 casinos in seven counties over the period
1995-2000. Four of the riverboat casinos are in Lake County in the
extreme northwest of Indiana near Chicago. Six other counties have one
riverboat casino each. Indiana also opened an 11th casino in an 8th
county in 2006, which is outside the period for which crime rate data
are currently available. (This 11th casino does, however, enter the data
through the variable indicating a 2-yr lead in casino opening.)
Indiana's casinos also have a readily available measure of
casino activity that we can use in addition to the dummy variables
related to dates of casino opening that other researchers have used. The
IGC's monthly revenue reports show "turnstile" counts of
casino admissions for the entire period covered here (7) (IGC,
1996-2004). As each riverboat casino owes $3 in admissions taxes for
each patron admitted to the casino, it must accurately count admissions.
In the early years of Indiana's riverboat casino operations,
patrons had to be readmitted for a new "excursion" every 2 h
even if they had not left the casino. In 2002, the state allowed
dock-side operations and amended the admissions tax scheme to eliminate
the readmission process. This change did not affect the definition of
the turnstile count because throughout the period each casino had
reported turnstile admissions, which is a count of the number of patrons
actually entering the casino. That is, before the 2002 change, patrons
who were readmitted for a new excursion had not passed through the
turnstile again (Klacik et al., 2003, pp. 5-6).
Two of the Lake County casinos share dock-side facilities. With the
2002 admissions tax changes, these casinos got IGC permission to operate
a single turnstile granting admission to both casinos and to evenly
split the admissions tax liabilities between them. (8) Thus, before this
change, these casinos operated two turnstiles and all patrons went
through one or the other. (In a minority of cases, patrons would go
through both [Klacik, Littlepage, and Payton 2001].) After the change,
all patrons for the two casinos would pass through a single turnstile,
and one-half of the total admissions were assigned to each casino.
Beginning with September 2002, the monthly reports of turnstile
admissions and admissions taxes for these two casinos show the totals
evenly split between them. With both casinos in the same county, this
change had no effect on county turnstile admission totals. This 2002
change seems to have had little effect on the turnstile count, as the
average turnstile count for these two casinos for the preceding years
2000 and 2001 was 1,753,209 and for the following years 2003 and 2004
was 1,738,138.
Indiana is also a good choice for studying the link between casinos
and crime because there are adequate offense data for Indiana counties
in the annual FBI UCR program data. (9) While researchers have often
used the UCR as the source for crime rate data, these crime reports have
some very important limitations: (10)
* the reported data are voluntary self reports of state or local
law enforcement agencies, which may not report at all in some years or
may report incompletely, with errors, or nonuniformly across
jurisdictions;
* the UCR only records the most serious crime in incidents in which
multiple crimes are committed;
* some agencies only report state totals; to get county offense
numbers, the ICPSR allocates offenses reported by these agencies to the
counties in proportion to each county's share of the state's
population.
* Most interesting for our purposes, the UCR includes imputations
of some incompletely reported offense data. That is, in adjusting for
incomplete reports the ICPSR replaces some incompletely reported data
with values not actually from the reporting agencies.
It is important to examine the imputation of incompletely reported
data more closely. Some agencies report offenses for only part of the
year. For agencies reporting 3-11 mo of data, the ICPSR inflates the
reported data up to a 12-moequivalent. For example, if an agency
reported 6 mo of data, the ICPSR doubles the number of reported offenses
to get a 12-mo number. For agencies reporting 0, 1, or 2 mo of data, the
ICPSR discards the reported data and replaces the number of offenses
with an estimate based on reports of agencies reporting 12 mo of data
within the same state and in cities or counties of the same type based
on urbanization and population. If there are no cities or counties of
that type having agencies reporting 12 mo of data, the ICPSR does not
estimate the missing values; instead it shows the number of offenses as
0.
Thus, in some cases, the idea that the UCR provides offense data is
an illusion, as the coded value is a number but it is not a reported
number of offenses. Recall that the ICPSR has coded some incompletely
reported values as 0. This listing of the number of crimes as
"0" can be particularly important for research relating new
casinos to local crime rates. For example, Iowa reported no offense data
for 1991. Thus for 1991, the year in which Iowa's first casinos
opened, (11) all Iowa counties show UCR crime totals of 0 for all
crimes. Tunica County, Mississippi, which had its first casino open in
1992, (12) reported no crime data for 1990-1998; so that, its offense
totals for those years are coded as 0. Illinois has casinos, but few
Illinois counties reported offense data over the period 1993-2004; so
that, the UCR shows most Illinois county offense totals as
"0." Thus, 0s in the UCR are often not actual offense totals.
It is important to recall that not all imputed values are 0s, however;
so that, some nonzero values are also not actual reported offense totals
for the county. Finally, the ICPSR's imputation method for UCR data
changed beginning with 1994, causing an important break in the
continuity of the crime data. The ICPSR cautions that researchers should
not compare UCR county-level crime data from 1993 and earlier with UCR
data from 1994 and later. (13)
Both E&T and G&M used UCR offense data in constructing
their dependent variables and used data from before and after the 1994
break in continuity. E&T, however, dealt with the ICPSR's
changing imputation methods to try to get consistent crime data over
time. E&T (p. 38) deleted post-1993 county-year observations having
inadequate data based on the coverage variable, discussed below, and
pre-1994 observations based on the proportion of the county's
Census Bureau population for which crimes were reported. G&M used
every county in every year 1977 through 1996, including the counties and
years in Mississippi, Iowa, and Illinois where many crime rates
resulting from incomplete reporting are inaccurate. Furthermore, G&M
treated the offense rates uniformly across the 1994 break in continuity.
Problems with these crime rate data suggest that the results reported in
G&M should be treated with caution.
Here, unlike previous studies, I avoid the worst of these problems.
I use data from Indiana counties for 1994 through 2004. Beginning my
analysis with 1994 data, I avoid the break in the data at 1993-1994. I
examine the statistical relationships among crime, casinos, and hotels
after deleting some observations with crime rate data having problems
created by incomplete reporting, as explained below.
As part of the change in imputations, beginning with 1994 the ICPSR
reports a variable called "coverage," which is the percentage
of the crime rate data for each county that is not imputed. This
percentage is 0, showing 100% imputation, for only 14 instances of
Indiana's counties over the period 1994 through 2004. I delete
those 14 observations with coverage equaling 0. Also, beginning with
1994, the ICPSR reports the county population of jurisdictions reporting
crime data. For example, a county may include an incorporated city with
a municipal police force and a county police force for the remainder of
the county. If only the city agency reports crime data, the ICPSR data
show the population of the city as the county population of
jurisdictions reporting crime data. (The meaning of this population
variable and the coverage variable, which depends on the population
figure, is complicated by the fact that some jurisdictions, such as
parks or toll roads, have 0 population. Thus, in some cases, there will
be reported crimes for jurisdictions with 0 population, and these crime
reports will not affect the coverage number.) To deal with this varying
population coverage, I also delete those observations for which the
UCR's county population of jurisdictions reporting crime data is
less than 60% of the county's population as reported by the Census
Bureau. For Indiana's counties for 1994 through 2004, there were
503 county-year observations having nonzero values for the coverage
variable and having ratios of county population of jurisdictions
reporting crime data to total county population of at least 0.6. Note
that I am truncating the sample based on coverage and population
reporting ratios, not the crime rate dependent variable--some of the
remaining crime rates are 0 and some of the deleted observations have
nonzero crime rates. I also deleted six observations in county panels
having only one observation, as these observations have no effect on
parameter estimates in a fixed-effects model. This leaves a data set of
497 observations. This smaller data set with adequate crime rate data
has information on 69 Indiana counties, including five counties having
casinos.
Data on hotel rooms by county over a long time period have not
previously been available. I have constructed this data set for Indiana
using a historical series of publications including AAA TourBook (AAA
Publishing, 1995-2005), Mobil Travel Guide, Great Lakes, and other
sources. I began by listing, for each year, all Indiana lodging
establishments (with their addresses and number of rooms) included in
the AAA TourBook (Illinois, Indiana, Ohio edition), assuming that the
guide covers lodging available for the year preceding the Guide's
copyright date. (This is clearly the case, as in a few cases the guide
for one year states that a hotel was scheduled to open on a date from
the previous year.) I then added any establishments that were in the
appropriate Mobil Travel Guide but had not been included in the AAA
guide. I then filled in any intervening years for establishments that
moved in and out of the guides. I then checked Internet sources
(including websites of the Association of Indiana Convention and
Visitors Bureaus, Hotel-Guides.us (2007), triprewards.com,
ChoiceHotels.com, and Dayslnn.com) to see if any listed establishment
was still operating in early 2007. For those that were, I filled in up
through 2004 from the most recent year having a guide listing. I also
added any casino hotels listed in IGC (various years) riverboat casino
evaluation reports but not included in the AAA or Mobil guides. Finally,
in some cases, I was able to fill backwards in time using Internet
travel sites, including TravelPost.com, that show the
establishment's date of opening. I then used each property's
town or street address to assign it to a county, using State of Indiana
(2007) or, for properties near county borders, Mapquest.com. The result
is a database of lodging properties by Indiana county for the years 1994
through 2004. I found 7,526 property-year observations in 78 of
Indiana's 92 counties. (14) For each county and year, I added the
rooms of all establishments to get the county's total rooms for
that year. The 14 counties for which I found no lodging establishments
have 0 rooms for all years.
IV. HOTEL ROOMS FOLLOW CASINOS
To test whether or not finding that casinos cause rising crime
rates after a lag of a few years could actually be finding that casinos
cause hotel rooms with a lag and new hotel rooms cause a contemporaneous
increase in crime rates, I add the number of hotel rooms to the crime
equation. It seems clear that new casinos and new hotels rooms are
linked, as we sometimes observe that casino openings are followed, often
with a lag, by new or expanded hotels. To test this more formally, I
estimate the parameters of an equation explaining hotel rooms by Indiana
county for the period 1994 through 2004 using a fixed-effects estimator
including both county and year effects. With 92 counties and 11 yr, we
have 1,012 observations. The equation to be estimated is
[rooms.sub.it] = [x.sub.it][beta] + [[alpha].sub.i] +
[[delta].sub.t] + [[epsilon].sub.it].
Here [rooms.sub.it] is the number of hotel rooms in county i in
year t. The row vector [x.sub.it] contains the values of the independent
variables for county i at time t, and [beta] is a vector of parameters
to be estimated. I use a dummy variable for the year of casino opening,
5 yr of lag dummy variables indicating year after casino opening,
population, population squared, population density, and real per capita
income, and various other demographic variables as independent variables
to explain the number of hotel rooms in each county. (See below for data
sources.) The constants [[alpha].sub.i] capture unobserved influences of
county i on hotel rooms that do not vary over the time period of the
sample. The constants [[delta].sub.t], capture unobserved effects of
year t on hotel rooms that are the same for all counties. The term
[[epsilon].sub.it] is the unobserved random error.
The direction of causality between casinos and hotels raises an
econometric issue here if the State of Indiana granted casino licenses
to localities because they had a large number of hotel rooms. The recent
French Lick Resort Casino, seems to illustrate such an occurrence. It
was constructed in French Lick, which for many years has been the site
of two large resort hotels. This casino, however, opened in 2006, which
is outside the range of this study. As for the remaining counties, the
Riverboat Gaming Act originally envisioned all of the Indiana casinos to
be on riverboats, so that all of Indiana's counties bordering Lake
Michigan or the Ohio River (or Patoka Lake, discussed above) were
eligible for casinos and no other counties were eligible (IGC, 1994).
Waterside location had little to do with the number of hotel rooms, and
indeed the counties in the Indianapolis area with the state's
largest number of hotel rooms were not eligible for casinos. Other
relatively large hotel markets (including South Bend, Fort Wayne,
Lafayette, Richmond, and Terre Haute) were also not eligible. Thus, the
state did not generally locate casinos in areas having the most hotel
rooms.
Table 1 shows the results of estimating the parameters of the
lodging rooms equation using all 92 counties for all 11 yr. Table Al
shows the average values of the dependent and independent variables. The
figures in parentheses are ratios of estimated coefficients to estimated
standard errors. The results confirm the conjecture that hotel rooms
follow casinos with a lag. Lags 3, 4, and 5 have statistically
significant, positive coefficients. The values peak 4 yr after the
casino's opening, indicating an increase of 241 rooms in the new
casino's county. Other variables are also positive and highly
statistically significant, including population and its square,
population density, and real per capita income. The "within"
[R.sup.2] goodness-of-fit measure used here, and throughout the paper,
shows that variation over time in the explanatory variables within each
county explains almost 75% of the variation in rooms.
We turn now to our main concern, casinos and crime.
V. RESULTS FOR CASINOS AND CRIME
To investigate the link between casinos and crime, I begin by
adopting the basic model and method used by G&M. That is, I explain
differing crime rates across counties and across time using a linear
regression model including fixed effects for both county and year.
Equation (2) shows the model:
[offenses.sub.it] = [x.sub.it][beta] + [[alpha].sub.i] +
[[delta].sub.t] + [[epsilon].sub.it]
The dependent variable [offenses.sub.it] is the rate of offenses
per 100,000 population in one of the six categories in county i in year
t. The row vector [x.sub.it] contains the values of the independent
variables for county i at time t, and [beta] is a vector of parameters
to be estimated. The constant [[alpha].sub.i] captures county fixed
effects for county i. The constant [[delta].sub.t] captures time fixed
effects for year t. The term [[epsilon].sub.it] is the unobserved random
error. As dependent variables, I use crime rates for three categories of
property crime--larceny, burglary, and motor vehicle theft--and three
categories of violent crime--aggravated assault, rape, and robbery. (15)
I include the same independent variables used by G&M, except
that they also used a variable measuring the extent to which the county
issued permits to carry concealed handguns under "shall issue"
requirements. I exclude this variable here because Indiana had a
statewide shall-issue requirement over the entire period examined here.
(Maltz and Targonski, 2002, p. 314) The explanatory variables in the
vector [x.sub.it] include dummy variables for the year in which the
first casino opened in the county, dummy variables for 1-yr and 2-yr
leads before casino opening and 1-yr through 5-yr lags after casino
opening. It also includes control variables that vary across counties
and year, including population density, real per capita income, real per
capita receipts of income maintenance payments and unemployment
compensation, real retirement payments per person over 64yr of age, and
percentages of population in various age and race categories. My source
for casino opening dates is IGC (various years). My sources for the
remaining data are U.S. Census Bureau and U.S. Department of Commerce
sources. (16) I augment the model with three additional variables.
First, I add two variables to the G&M specification: the number of
hotel rooms in the county and the turnstile measure of casino activity.
Later, I will add a deterrence variable, which will also require
changing to instrumental variables estimation.
A. An Aside on the Impacts of the Defective Data
We will spend little time looking at the crime equations for the
full sample because as discussed above, the dependent variable is
defective for about one-half of the sample. We can, however, examine the
impact of the defective crime rate data on some of the results. Table 2
shows the estimated parameters for the hotel room and casino variables
from Equation (2) for each category of crimes for both the full and
reduced samples. For each crime category, the first column shows results
using the reduced sample of 497 observations, and the second column
shows the results for the same specification using the full sample of
1,012 observations. In all cases, the estimated equations include county
and year fixed effects. Many of the results are qualitatively similar in
the full and reduced samples. (The crime data for Indiana are not
especially low in quality, as they are for Illinois, for example.) There
are, however, some important differences, especially for the burglary
equation. For burglary, we see that eliminating the bad data reduces the
coefficients of rooms and turnstile by half and eliminates their
statistical significance. It has the opposite effect on the 3-yr and
4-yr lag variables, doubling the coefficients and making them
statistically significant. Thus, moving from the complete sample to the
reduced sample changes some of the results. We now turn to analysis of
the results and some additional specifications using the reduced sample.
B. Results and Analysis Using the Reduced Sample
The results for the reduced sample in Table 2 show that increased
casino activity (turnstile) reduces all crime rates, although the
negative coefficient is not statistically significant for burglary or
rape. The coefficients on the casino lead and lag variables are
generally insignificant, except for aggravated assault where they are
large and negative, burglary where they are positive and at the
borderline of significance at Lag 3 and positive and significant at Lag
4, motor vehicle theft where they are negative at the casino opening and
at the 2-yr lead, and robbery where the coefficient on casino opening is
negative. The coefficient on the hotel rooms variable is also always
negative, but it is statistically significant only for larceny, motor
vehicle theft, and at the borderline of significance for rape. Before
interpreting these results further, we examine one more important
issue--allowing for deterrence in determining crime rates.
The economic model of crime suggests that increasing the
opportunity cost of crime reduces the crime rate. The specification used
up to this point leaves out components of the opportunity cost of crime
associated with the probability of arrest and conviction, often known as
deterrence. Many of the papers cited above that apply the economic model
of crime try to measure the impact of deterrence variables on crime
rates. Some researchers (e.g., Gould, Weinberg, and Mustard, 2002;
Levitt, 1998) use arrest rates, while others use measures of enforcement
effort such as the number of police (e.g., Levitt, 2002). While the
results in the deterrence literature are mixed, researchers have found
that deterrence variables can be important in crime equations.
Introducing deterrence variables to the crime equation leads to
important potential sources of bias in estimating the crime
equation's parameters. There are two main issues. The first is
spurious correlation resulting from measurement error in the number of
offenses (Gould, Weinberg, and Mustard, 2002; Levitt, 1998). If the
dependent variable is the crime rate (offenses/ population) and we use
the arrest rate (arrests/ offenses) as an explanatory variable, then the
number of offenses is both the numerator of the dependent variable and
the denominator of this right-hand-side variable. Any measurement error
in the number of offenses will appear on both sides of the equation,
creating a negative bias in the estimated coefficient of the arrest
rate. Gould, Weinberg, and Mustard (2002, p. 51) call this
"division bias." As we have seen, in this data set,
measurement error in the number of offenses is an important problem.
The second source of bias arises from the fact that that we can
expect that measures of deterrence will be jointly determined with crime
rates. That is, a rise in local crime rates will almost certainly cause
local officials to increase crime deterrence efforts, such as hiring
more police. This creates a simultaneous equations bias in OLS estimates
of the effects of deterrence variables in the crime equation.
One could potentially deal with measurement error biases and
simultaneous equations bias using instrumental variables (IV). This
requires having at least one instrumental variable that is correlated
with the arrest rate or other deterrence variable but is uncorrelated
with the error in the crime equation, [[epsilon].sub.it] in Equation
(2). Levitt (1997) used local election cycles for a wide range of U.S.
cities as instruments, arguing that the timing of local elections may
affect the hiring of new police but the timing of local elections is not
related to local crime. Using local election cycles as instruments will
not work when we, unlike Levitt, are dealing with only one state.
Indiana has most municipal elections on the same 4-yr cycle, so that
there will be no variation in timing of elections across counties. With
no cross-section variation in election timing, we cannot use election
cycles as an instrumental variable in a panel equation.
Here we will use the aggregate arrest rate for the four categories
of violent crime (aggravated assault, robbery, rape, and murder) as the
measure of deterrence in the property crime equations. Using the violent
crime arrest rate to measure deterrence in the property crime equations
avoids the division bias. Furthermore, for the property crime equations,
we have a variable having the properties we seek in an instrument for
the violent crime arrest rate: the percentage of local government
employment in total county employment. (17) First, we would expect that
local government employment would be higher where local deterrence
effort is higher. Second, this variable is statistically significant is
explaining the violent crime arrest rate. (18) Third, this variable is
not significant in explaining any of the property crime rates. (19)
Unfortunately, we cannot make the analogous argument for using the
property crime arrest rate as the measure of deterrence in the violent
crime equations. The problem here is that the percentage of local
government employment in total county employment is significant in
explaining violent crime rates. This correlation of the proposed
instrument with the dependent variable rules it out as an instrument for
the deterrence variable. Therefore, we proceed below with estimates
including the deterrence variable for the property crimes only.
Table 3 shows the estimated parameters for the deterrence, hotel
room, and casino variables from Equation (2) with various specifications
of the right-hand side variables [x.sub.it]. (Table A2 shows the
estimated parameter values for the other independent variables.) In all
cases, the estimated equations include county and year fixed effects.
For each category of offense, the first column shows results using OLS,
reproduced from Table 2. The second column shows the same equation
estimated using OLS with the violent crime arrest rate added. The third
column for each crime category shows the equation including the
deterrence variable and estimated using instrumental variables with the
percentage of local government employment in total county employment as
the instrument.
Looking at the center columns, for each category of property crime
(OLS) the deterrence variable has a negative coefficient; but it is only
at the borderline of statistical significance in a one-tailed test for
burglary and motor vehicle theft. The statistical significance of
deterrence disappears in the instrumental variable estimation, shown for
each category of property crime in the columns on the right. Adding the
deterrence variable has little effect on the estimated parameters and
ratios of parameters to standard errors of the casino timing, turnstile,
and rooms variables.
We can examine the patterns of coefficients on lead, casino
opening, lag variables, and turnstile in Table 3 to look for impacts of
new casinos on crime. The patterns are the same in all three
specifications. There are no significant effects of the timing of casino
openings on rates of larceny, although the coefficients on all of the
lag variables are positive. (The five lag variables are not jointly
significantly different from 0 in any of the three specifications.) The
coefficient on the turnstile variable shows a large and significant
negative effect of casino activity on larceny. The coefficient on the
turnstile variable shows a significant and negative effect of casino
activity on motor vehicle thefts. The casino opening and 2-yr lead
variables are negative and statistically significant in explaining motor
vehicle thefts, while the lag coefficients are small, insignificant, and
vary in sign. (Again, the five lag variables are not jointly
significantly different from 0 in any of the three specifications.) The
burglary equations show a different result. For burglary, the
coefficient on the turnstile variable is small and not statistically
significant. The coefficients for the 3-yr lag are positive and near the
borderline for significance and for the 4-yr lag are positive and
significant, suggesting that increased burglary rates follow new
casinos. The lags at years 3 and 4 are jointly significant at the .95
level for all three specifications. For the first OLS burglary
specification, all five lag variables are also jointly significant at
the .95 level (F = 2.32; critical value 2.21), while the five lag
variables are at the borderline of significance in the second and third
specifications.
VI. IMPLICATIONS OF THE RESULTS
These results suggest that introducing a new casino increases local
burglary rates after a lag of a few years. Otherwise, contrary to the
results in G&M and E&T, these results do not show that
introducing new casinos increases local crime rates. The results
suggest, moreover, that increasing casino activity, in this case the
turnstile count of casino patrons, leads to reductions in local crime
rates. (This effect of casino activity is, however, small and
statistically insignificant for burglary.) These results suggest that
for some crime categories, new casinos may actually reduce local crime
rates. Furthermore, new casinos lead to new hotel room construction,
which may lead to additional reductions in some crime rates.
We can use the estimated parameter values and the average of the
turnstile variable to calculate an estimate of the quantitative effect
on crime rates of the opening of a new casino in an Indiana county.
Table 4 shows the averages of the turnstile variable (in thousands) by
county, for counties having casinos in the reduced sample, by years
since the opening of the county's first casino. In calculating the
effects of new casinos, I use only the results for the first column for
each crime category from Table 3. I begin with the average crime rate
for the counties and years included in the reduced sample and then add
the lead or lag coefficient plus the product of the turnstile
coefficient and the average turnstile count for the appropriate lag from
Table 4. The results show approximate average effects of new casinos on
crime rates over time. (I do not attribute to casinos any of the effects
of hotel rooms on crime rates.) Figures 1, 2, and 3 show these net
effects of the lead, casino open, lag, and turnstile variables on the
predicted crime rates for larceny, burglary, and motor vehicle theft,
respectively. The figures also show the 95% confidence bounds above and
below the estimated values. The net effect of a new casino on larceny
rates is small, as the large and significant turnstile effect roughly
cancels out the large (but insignificant) lag effects. We can see that
county burglary rates rise a few years after opening a new casino
because the turnstile effect is too small to cancel out the large
effects increasing burglary rates 3 and 4 yr after the casino's
opening. The large negative turnstile effect substantially reduces the
rate of motor vehicle thefts.
[FIGURE 1 OMITTED]
[FIGURE 2 OMITTED]
[FIGURE 3 OMITTED]
VII. CONCLUSION
This paper examines the link between casinos and crime using
Indiana's counties for the years 1994 through 2004. This paper uses
a specification of the equations explaining crime rates which includes
the number of hotel rooms in the county and a count of casino patrons as
a measure of casino activity. I have also deleted defective crime rate
data that have plagued earlier studies. As a first result, I find that
construction of new hotel rooms follows the introduction of casinos into
the county. These new hotel rooms seem to reduce the levels of larceny
and motor vehicle theft. Turning to casinos and crime, I find very
limited support for the proposition that new casinos increase local
crime rates. Opening new casinos appears to increase the number of
burglaries in the county after a lag of a few years. Opening new casinos
appears, however, to reduce the number of motor vehicle thefts and
aggravated assaults. Increased casino activity, measured using turnstile
count of casino patrons, seems to reduce rates of larceny, motor vehicle
theft, aggravated assault, and robbery. These results do not match those
of earlier studies that show large increases in a broad range of local
crime rates after opening new casinos. The results presented here,
however, are based on only a small segment of the nation's
experience with the casino industry, and this narrow focus limits the
applicability of the results. Expanding investigations like this to
other states will be limited by the availability of adequate crime data,
annual data by county on hotels or other establishments, and consistent
measures of casino activity.
Nevertheless, the new results presented here provide guidance for
future work on casinos and crime. I introduce a measure of casino
activity in addition to variables related to the timing of casino
opening. The measure of casino activity, turnstile count of patrons
entering the casinos, is often statistically significant and its effect
is negative, suggesting that increased casino activity reduces crime
rates. It may also suggest that casinos aimed at attracting large
numbers of patrons are more likely than small casinos to reduce local
crime rates. I also test whether or not the number of hotel rooms
affects crime rates. The estimated effect of the number of hotel rooms
in the county on crime rates is always negative, but when restricting
the sample to observations with relatively complete data, this variable
is only significant in the equations for larcenies and motor vehicle
theft. The statistical significance of the turnstile variable suggests
that leaving out a measure of casino activity when estimating the effect
of casinos on crime is a serious specification error. The significance
of the hotel room variable in equations for some crime rates suggests
that a measure of the level of other public accommodations in addition
to casinos may be necessary to avoid specification error.
These results provide some support for the economic model of
crime--the statistically significant negative effect of the turnstile
measure of casino activity may suggest that increasing casino activity
increases the opportunity cost of crime by providing increased
legitimate employment opportunities.
Finally, researchers need to exercise care in using county-level
Uniform Crime Reports data to study links between casinos and crime. In
all cases, the UCR's coded number of crimes at the county level is
a number, but in many cases it is not a reported number of crimes.
REFERENCES
AAA Publishing. AAA TourBook. Heathrow. FL: AAA Publishing,
1995-2005 annual editions.
Albanese, J. "The Effect of Casino Gambling on Crime."
Federal Probation, 48, 1985. 39-44.
American Gaming Association. "States with Gaming." 2009,
www.americangaming.org/Industry/factsheets/general_info_detail.cfv?id=15. Accessed May 20, 2009.
Association of Indiana Convention and Visitors Bureaus. "CVB
Directory." 2007, www.aicvb.org/cvbdirectory.html.
Becker, G. S. "Crime and Punishment: An Economic
Approach." Journal of Political Economy, 76, 1968, 169-217.
Bowes, D. R., and K. R. Ihlanfeldt. "Identifying the Impacts
of Rail Transit Stations on Residential Property Values." Journal
of Urban Economics, 50, 2001, 1-25.
Cornwell, C., and W. N. Trumbull. "Estimating the Economic
Model of Crime with Panel Data." Review of Economics and
Statistics, 76, 1994, 360-66.
Ehrlich, I. "Participation in Illegitimate Activities: A
Theoretical and Empirical Investigation." Journal of Political
Economy, 81, 1973, 521-65.
Ehrlich, I. "Crime, Punishment, and the Market for
Offenses." Journal of Economic Perspectives, 10, 1996, 43-67.
Evans, W. N., and J. H. Topoleski. "The Social and Economic
Impact of Native American Casinos." NBER Working Paper 9198, 2002.
Glaeser, E. L, B. Sacerdote, and J. A. Scheinkman. "Crime and
Social Interactions." Quarterly Journal of Economics, 111, 1996,
507-48.
Gould, E. D., B. A Weinberg, and D. B Mustard. "Crime Rates
and Local Labor Market Opportunities in the United States:
1979-1997." Review of Economics and Statistics, 84, 2002, 45-61.
Grinols, E. L., and D. B. Mustard. "Casinos, Crime, and
Community Costs." Review of Economics and Statistics, 88, 2006,
28-45.
Grinols, E. L., and D. B. Mustard. "Correctly Critiquing
Casino-Crime Causality." Econ Journal Watch, 5, 2008, 21-31.
Hotel-Guides.us. 2007. Hotel-Guides.us/indiana/in-hotels.html.
Ihlanfeldt, K. "Neighborhood Crime and Young Males' Job
Opportunity." Journal of Law and Economics, 49, 2006, 249-83.
IGC. "Annual Report to the Governor, 1993" 1994.
www.in.gov/igc/files/CY1993-Annual.pdf. Accessed May 20, 2009.
IGC. "Annual Report to the Governor." 2003.
www.in.gov/igc/files/annual/FY2003-Annual.pdf.
IGC. "Monthly Revenue Reports." 1996-2004,
www.in.gov/igc/2363.htm.
IGC. "Casino Evaluations." (Various years).
WWW.in.gov/igc/2361.htm.
Iowa Racing and Gaming Commission. [No date]. "Chronology Of
The Iowa Racing And Gaming Commission."
www.iowa.gov/irgc/Chronology.htm. Accessed May 20, 2009.
Klacik, D., L. Littlepage, and S. Payton. "Five-Year License
Renewal: Majestic Star Casino, LLC." 2001.
www.in.gov/igc/files/majestic-5.pdf.
Klacik, D., L. Littlepage, S. Payton, L. DeBoer, and C. Sullivan.
"Eight-Year License Renewal: Aztar Indiana Gaming
Corporation." 2003. www.in.gov/igc/files/aztar-8.pdf.
Levitt, S. D. "Using Electoral Cycles in Police Hiring to
Estimate the Effects of Police on Crime." American Economic Review,
87, 1997, 270-90.
--. "Why Do Increased Arrest Rates Appear to Reduce Crime:
Deterrence, Incapacitation, or Measurement Error." Economic
Inquiry, 36, 1998, 353-72.
--. "Using Electoral Cycles in Police Hiring to Estimate the
Effects of Police on Crime: Reply." American Economic Review, 92.
2002, 1244-50.
--. "Understanding Why Crime Fell in the 1990s: Four Factors
that Explain the Decline and Six that Do Not." Journal of Economic
Perspectives, 18, 2004, 163-90.
Maltz, M. D. "Analysis of Missingness in UCR Crime Data."
2006, http://www.ncjrs.gov/pdffiles1/nij/grants/215343.pdf.
Maltz, M. D, and J. Targonski. "A Note on the Use of
County-Level UCR Data." Journal of Quantitative Criminology, 18,
2002, 297-318.
Mapquest.com.
Mississippi Gaming Commission. "History of Licensure for
Operating Casinos." 2009
www.mgc.state.ms.us/pdf/historyofmgclicensees.pdf. Accessed May 20,
2009.
Mobil Travel Guide, Great Lakes. New York: Fodor's Travel
Publications (1995, 1996, 1997, 1998, 1999); Lincolnwood, IL:
Publications International (2000, 2001, 2002); Park Ridge, IL: EMTG, LLC
(2002, 2004, 2005).
Raphael, S., and R. Winter-Ebmer. "Identifying the Effect of
Unemployment on Crime." Journal of Law and Economics, 44, 2001,
259-83.
Sjoquist, D. L. "Property Crime and Economic Behavior: Some
Empirical Results." American Economic Review, 63, 1973, 439-46.
State of Indiana. "City/Town Locator." 2007.
www.in.gov/mylocal/.
Stitt, B. G., M. Nichols, and D. Giacopassi, "Does the
Presence of Casinos Increase Crime? An Examination of Casino and Control
Communities." Crime & Delinquency, 49, 2003, 253-84.
Stokowski, P. "Crime Patterns and Gaming Development in Rural
Colorado." Journal of Travel Research, 34, 1996, 63-69.
TravelPost.com.
Triprewards.com.
Trump Entertainment Resorts, Inc. 2005, "Form 10-K."
period ending December 31, 2004.
U.S. Department of Justice, Federal Bureau of Investigation.
Uniform Crime Reporting Program Data [United States]: County-Level
Detailed Arrest And Offense Data, 1994-2004 [Computer files]. 2nd ICPSR
edition. Ann Arbor. MI: Interuniversity Consortium for Political and
Social Research [producer and distributor], various years.
U.S. Department of Justice, Federal Bureau of Investigation.
"Codebook," in Uniform Crime Reporting Program Data [United
States]: County-Level Detailed Arrest And Offense Data, 2004 [Computer
files]. Ann Arbor. MI: Interuniversity Consortium for Political and
Social Research [producer and distributor], 2006, ICPSR04466-v1.
Walker, D. M. "Do Casinos Really Cause Crime?" Econ
Journal Watch, 5, 2008, 4-20.
Wilson, J. M. "Riverboat Gambling and Crime in Indiana: An
Empirical Investigation." Crime & Delinquency, 47, 2001,
610-40.
(1.) Albanese (1985), Stitt, Nichols, and Giacopassi (2003), and
Stokowski (1996) also study the link between casinos and crime. These
studies use small data sets. One feature of these studies, emphasized by
Walker (2008), is that they assert that the population at risk of crime
in areas having casinos includes casino visitors from outside the local
area in addition to the local population. This suggests that using the
local population as the denominator in calculating the crime rate
overstates crime rates in casino areas. Following G&M, I use local
population in calculating crime rates. Wilson (2001) studies crime in
two Indiana casino cities using weekly and monthly data extending 1 yr
beyond the casino openings.
(2.) U.S. Department of Justice, Federal Bureau of Investigation
(various years). Researchers may freely download these UCR crime rate
data subject to the restriction that the researchers may not distribute
the data to others. On request, the author will provide all of the data
used here except for the crime data. For the crime data, the author will
provide the link to the ICPSR source of the data and the calculations
that convert the raw data to the variables used here.
(3.) See Ehrlich (1996) for a more complete explanation of the
economic model of crime. See also Glaeser, Sacerdote, and Scheinkman
(1996) for a generalization of the economic model of crime to include
social interactions (rather than independent decisions) in explaining
criminal behavior.
(4.) As Walker (2008) and others have pointed out, the new casinos
and other new public accommodations would increase the local area's
population at risk of being involved in crime and thereby increase the
number of crimes. This would increase the calculated crime rate if the
local population rather than the entire population at risk were included
in the denominator of the crime rate.
(5.) See, for example. Walker (2008).
(6.) Walker (2008) discusses problems with G&M's data,
including the defects in the UCR crime data and the lack of a measure of
casino activity. Grinols and Mustard (2008) responds to these comments.
(7.) The monthly reports arc available beginning 3 mo after
Indiana's first casino opened. Here I assume each of the first
turnstile counts for that casino was 100,000. The first recorded
month's turnstile count was 108.205 for March 1996. I also
incorporated two later corrections for monthly turnstile figures for
March 2000 and May 2004. Also, using the IGC's monthly revenue
reports allows us to construct consistent calendar year totals despite
the shift of IGC's annual reporting from calendar year to fiscal
year during 2002.
(8.) Trump Entertainment Resorts, Inc. (2005).
(9.) This is notably in contrast to Illinois and Mississippi. which
lack adequate crime data to investigate the link between casinos and
crime.
(10.) Maltz (2006), Maltz and Targonski (2002). U.S. Department of
Justice. Federal Bureau of Investigation (2006).
(11.) Iowa Racing and Gaming Commission (undated).
(12.) Mississippi Gaming Commission (2009).
(13.) U.S. Department of Justice. Federal Bureau of Investigation
(2006), p. i.
(14.) Indiana has many small counties. In 1995. it had 15 counties
with populations less than 17,000.
(15.) I do not report results for the murder category. The model
explained little of the variation in county murder rates, and the casino
variables were not statistically significant in the murder equation.
(16.) Population: 1990-1999 Intercensal State and County
Characteristics (www.census.gov/popest/archives/est90intercensal/stch-intercensal/stch-intercensal_layout.txt). County Population Estimates by
Age, Sex, Race and Hispanic Origin: April 1. 2000 to July 1. 2005
(http://www.census.gov/popest/counties/asrh/files/cc-est-alldata_layout.txt); income, unemployment compensation, income maintenance payments,
and retirement payments: U.S. Department of Commerce. Bureau of Economic
Analysis, REIS (http://bea.gov/bea/regional/reis); County land area:
http://www.fedstats.gov/qf/download/DataSet.txt
(17.) The local government employment data (U.S. Department of
Commerce, Bureau of Economic Analysis, REIS) had 14 missing values for
counties and years used here. I replaced the missing values with the
most recent prior available value.
(18.) Estimating the equation from Table 3 (and Table Al) with the
violent crime arrest rate as the dependent variable and the local
government employment instrumental variable added as an explanatory
variable yields a t-value for the local government employment variable
of 2.36.
(19.) Estimating the equation from Table 3 (and Table Al) with each
offense rate as dependent variable and the local government employment
instrumental variable added as an explanatory variable yields t-values
for the local government employment variable ranging between 0.23 and
0.50.
ABBREVIATIONS
E&T: Evans and Topoleski
FBI: Federal Bureau of Investigation
G&M: Grinols and Mustard
ICPSR: Interuniversity Consortium for Political and Social Research
IGC: Indiana Gaming Commission
UCR: Uniform Crime Reporting
WILLIAM S. REECE *
* Without implicating them in any of the results, the author thanks
Craig Depken, Earl Grinols, Rich Harrill, Brad Humphreys, Peter Leeson,
David Mustard, Santiago Pinto, Russell Sobel, William Trumbull, seminar
participants at the Alfred P. Sloan Foundation Travel and Tourism
Industry Center at the University of South Carolina, and three anonymous
referees for helpful and generous comments on earlier drafts.
Reece: Professor, Department of Economics, College of Business and
Economics, West Virginia University, PO Box 6025, Morgantown, WV
26506-6025. Phone (304) 293-4039, Fax (304) 293-5652, E-mail william.
reece@mail.wvu.edu
doi: 10.1111/j.1465-7287.2009.00172.x
TABLE 1
Hotel Rooms Regression Results--92 Indiana Counties, 1994-2004
Rooms
Casino open -148.0 * (3.76)
LAG1 -110.3 * (2.80)
LAG2 49.0 (1.24)
LAG3 179.6 * (4.56)
LAG4 241.4 * (6.14)
LAG5 187.0 * (4.43)
Population -37.3 * (7.23)
Pop squared 0.0612 * (22.01)
Density 9.78 * (5.00)
Real income 42.1 * (5.42)
Male -71.4 * (3.40)
White 51.0 * (2.01)
Black 189.1 * (7.52)
Pop 10-19 -53.9 * (2.97)
Pop 20-29 -45.0 * (2.88)
Pop 30-39 -92.3 * (3.85)
Pop 40-49 -28.0 (1.60)
Pop 50-64 -89.5 * (5.19)
Pop over 64 -58.4 * (3.12)
Constant 3176.3 (0.95)
[R.sup.2] .7356
Num obs 1012
Note: t-values are given in parentheses.
* Statistically significant in a two-tailed test at the .95 level.TABLE 2
Crime Regression Results for Full and Reduced Samples--Property and
Violent Crimes
Larceny
Sample Reduced Full
Rooms -0.245 * (2.02) -0.283 * (2.03)
Turnstile -0.0845 * (2.61) -0.0888 * (2.15)
LEAD2 -303.9 (1.48) -421.5 (1.91)
LEAD1 -141.7 (0.69) -277.8 (1.23)
Casino open -248.1 (1.27) -337.9 (1.56)
LAG1 136.7 (0.76) 76.7 (0.38)
LAG2 102.4 (0.59) 98.7 (0.49)
LAG3 216.9 (1.26) 157.4 (0.78)
LAG4 141.1 (0.82) 85.1 (0.42)
LAG5 58.2 (0.34) 52.3 (0.25)
[R.sup.2] .164 .094
Num obs 497 1012
Burglary
Sample Reduced Full
Rooms -0.046 (0.97) -0.098 * (2.12)
Turnstile -0.0146 (1.14) -0.0287 * (2.12)
LEAD2 30.21 (0.37) -66.4 (0.91)
LEAD1 -10.2 (0.13) -92.9 (1.25)
Casino open -42.1 (0.55) -112.6 (1.58)
LAG1 -9.21 (0.13) -44.2 (0.66)
LAG2 -70.5 (1.02) -60.9 (0.92)
LAG3 131.8 (1.94) 71.6 (1.08)
LAG4 146.3 * (2.16) 74.1 (1.12)
LAG5 84.0 (1.26) 61.7 (0.88)
[R.sup.2] .118 .071
Num obs 497 1012
Motor Vehicle Theft
Sample Reduced Full
Rooms -0.049 * (2.33) -0.053 * (2.91)
Turnstile -0.0444 * (7.87) -0.0432 * (7.99)
LEAD2 -78.4 * (2.19) -70.7 * (2.44)
LEAD1 -49.4 (1.39) -56.2 (1.89)
Casino open -71.4 * (2.11) -63.0 * (2.22)
LAG1 45.2 (1.45) 41.9 (1.58)
LAG2 -6.0 (0.20) 7.2 (0.27)
LAG3 4.3 (0.14) 16.2 (0.61)
LAG4 -11.6 (0.39) -3.81 (0.14)
LAG5 -6.8 (0.23) 2.22 (0.08)
[R.sup.2] .286 .143
Num obs 497 1012
Assault
Sample Reduced Full
Rooms -0.063 (1.58) -0.024 (0.83)
Turnstile -0.0436 * (4.10) -0.0460 * (5.41)
LEAD2 -156.0 * (2.31) -106.5 * (2.34)
LEAD1 -194.8 * (2.90) -135.4 * (2.90)
Casino open -128.9 * (2.01) -59.0 (1.32)
LAG1 -162.0 * (2.74) -65.9 (1.58)
LAG2 -165.0 * (2.87) -68.6 (1.65)
LAG3 -152.6 * (2.69) -89.3 * (2.15)
LAG4 -147.5 * (2.61) -102.0 * (2.46)
LAG5 -79.9 (1.44) -57.5 (1.31)
[R.sup.2] .319 .182
Num obs 497 1012
Robbery
Sample Reduced Full
Rooms -0.0063 (0.71) -0.0087 (1.32)
Turnstile -0.0110 * (4.63) -0.0113 * (5.83)
LEAD2 -8.65 (0.57) -8.21 (0.79)
LEAD1 -2.45 (0.16) -5.16 (0.48)
Casino open -28.3 * (1.99) -19.8 (1.94)
LAG1 9.98 (0.76) 13.3 (1.39)
LAG2 -2.34 (0.18) 0.76 (0.08)
LAG3 -2.03 (0.16) 2.86 (0.30)
LAG4 -6.05 (0.48) -3.10 (0.33)
LAG5 -6.78 (0.55) -2.57 (0.26)
[R.sup.2] .182 .104
Num obs 497 1012
Rape
Sample Reduced Full
Rooms -0.0067 * (1.96) -0.0062 * (2.07)
Turnstile -0.0017 (1.91) -0.0019 * (2.12)
LEAD2 -4.48 (0.78) -7.04 (1.48)
LEAD1 0.421 (0.07) -2.44 (0.50)
Casino open 1.23 (0.23) -2.84 (0.61)
LAG1 2.87 (0.57) -1.99 (0.46)
LAG2 1.55 (0.31) 0.63 (0.14)
LAG3 -1.70 (0.35) 1.30 (0.30)
LAG4 5.31 (1.10) 1.61 (0.37)
LAG5 6.50 (1.37) 7.76 (1.69)
[R.sup.2] .139 .064
Num obs 497 1012
Note: t-values are given in parentheses.
* Statistically significant in a two-tailed test at the .95 levelTABLE 3
Crime Regression Results for Reduced Sample--Property Crimes
Larceny
OLS OLS IV OLS
Violent arrest -14.53 (0.59) -48.52 (0.23)
rate
Rooms -0.245 * (2.02) -0.244 * (2.01) -0.242 * (1.98)
Turnstile -0.0845 * (2.61) -0.0846 * (2.61) -0.0849 * (2.61)
LEAD2 -303.9 (1.48) -303.0 (1.47) -301.0 (1.45)
LEAD1 -141.7 (0.69) -140.3 (0.68) -137.0 (0.66)
Casino open -248.1 (1.27) -247.4 (1.27) -245.6 (1.25)
LAG1 136.7 (0.76) 145.8 (0.81) 167.0 (0.75)
LAG2 102.4 (0.59) 105.7 (0.60) 113.3 (0.62)
LAG3 216.9 (1.26) 215.3 (1.25) 211.7 (1.21)
LAG4 141.1 (0.82) 140.6 (0.82) 139.5 (0.81)
LAG5 58.2 (0.34) 56.5 (0.33) 52.6 (0.31)
[R.sup.2] .164 .164 .160
Num obs 497 497 497
Burglary
OLS OLS IV OLS
Violent arrest rate -14.76 (1.51) 38.87 (0.45)
Rooms -0.046 (0.97) -0.045 (0.95) -0.048 (0.97)
Turnstile -0.0146 (1.14) -0.0147 (1.15) -0.0144 (1.08)
LEAD2 30.2 (0.37) 31.1 (0.38) 27.9 (0.33)
LEAD1 -10.2 (0.13) -8.81 (0.11) -14.0 (0.17)
Casino open -42.1 (0.55) -41.4 (0.54) -44.1 (0.55)
LAG1 -9.21 (0.13) -0.011 (0.00) -33.4 (0.37)
LAG2 -70.5 (1.02) -67.2 (0.97) -79.2 (1.07)
LAG3 131.8 (1.94) 130.2 (1.92) 136.0 (1.91)
LAG4 146.3 * (2.16) 145.8 * (2.15) 147.6 * (2.10)
LAG5 84.0 (1.26) 82.3 (1.24) 88.5 (1.27)
[R.sup.2] .118 .123 .056
Num obs 497 497 497
Motor Vehicle Theft
OLS OLS IV
Violent arrest -6.98 (1.62) -13.04 (0.36)
rate
Rooms -0.049 * (2.33) -0.049 * (2.31) -0.048 * (2.28)
Turnstile -0.0444 * (7.87) -0.0444 * (7.90) -0.0445 * (7.88)
LEAD2 -78.4 * (2.19) -78.0 * (2.18) -77.7 * (2.16)
LEAD1 -49.4 (1.39) -48.7 (1.37) -48.1 (1.34)
Casino open -71.4 * (2.11) -71.1 * (2.10) -70.8 * (2.08)
LAG1 45.2 (1.45) 49.6 (1.58) 53.4 (1.38)
LAG2 -6.02 (0.20) -4.46 (0.15) -3.10 (0.10)
LAG3 4.27 (0.14) 3.52 (0.12) 2.86 (0.09)
LAG4 -11.6 (0.39) -11.8 (0.40) -12.0 (0.40)
LAG5 -6.80 (0.23) -7.62 (0.26) -8.33 (0.28)
[R.sup.2] .286 .291 .287
Num obs 497 497 497
Note: t-values are given in parentheses.
* Statistically significant in a two-tailed test at the .95 levelTABLE 4 Average Turnstile Count (in Thousands) for Counties Having
Casinos in the Reduced Sample
Year Casino Opens Lag 1 Lag 2 Lag 3 Lag 4 Lag 5
899.6 2887.1 3447.9 3519.4 3591.2 3542.9
APPENDIX
TABLE A1
Descriptive Statistics
Crime Rate Equations
Reduced Sample
Variable Observations Average SD Minimum Maximum
Turnstile 497 295.00 1324.5 0.00 11429
Rooms 497 942.63 2232.6 0.00 16202
Larceny 497 1817.07 1073.8 24.23 4685.7
Burglary 497 527.02 298.6 3.01 1417.3
MV theft 497 191.93 185.9 0.00 1225.5
Assault 497 176.84 158.4 5.83 1021.4
Robbery 497 46.63 70.8 0.00 403.7
Rape 497 18.90 15.2 0.00 118.3
Density 497 254.89 385.7 23.31 2178.6
Male 497 49.28 0.96 47.24 52.4
White 497 95.57 5.99 72.35 99.9
Black 497 3.66 5.73 0.02 26.7
Pop 10-19 497 14.82 0.89 12.71 18.4
Pop 20-29 497 13.24 3.06 9.00 29.6
Pop 30-39 497 14.50 1.41 11.10 19.7
Pop 40-49 497 14.99 1.16 11.17 17.6
Pop 50-64 497 15.54 1.83 10.45 22.7
Pop over 64 497 13.13 1.96 7.38 16.9
Real income 497 14.61 2.18 10.84 25.7
Retirement 497 7.67 0.63 5.32 9.4
Maintenance 497 149.38 54.77 41.08 324.3
Unemployment 497 46.19 26.91 9.41 129.1
Viol arrest rate 497 0.75 0.97 0.00 9.50
Local government 497 0.085 0.026 0.043 0.149
Crime Rate Equations
Full Sample
Variable Observations Average SD Minimum Maximum
Turnstile 1012 164.42 952.9 0.00 11429
Rooms 1012 569.99 1616.9 0.00 16202
Larceny 1012 1211.68 1102.8 0.00 4685.7
Burglary 1012 341.53 311.8 0.00 1639.9
MV theft 1012 122.53 156.0 0.00 1225.5
Assault 1012 125.54 136.9 0.00 1021.4
Robbery 1012 27.68 55.2 0.00 403.7
Rape 1012 13.36 14.5 0.00 177.5
Density 1012 173.19 287.7 22.34 2178.6
Male 1012 49.40 0.92 47.24 54.0
White 1012 97.05 4.64 72.35 99.9
Black 1012 2.33 4.39 0.00 26.7
Pop 10-19 1012 14.84 0.86 12.71 18.4
Pop 20-29 1012 12.80 2.56 9.00 29.6
Pop 30-39 1012 14.63 1.33 11.10 19.9
Pop 40-49 1012 15.03 1.08 11.17 17.9
Pop 50-64 1012 15.58 1.69 10.41 22.7
Pop over 64 1012 13.35 1.80 7.38 16.9
Real income 1012 14.18 2.17 9.73 25.7
Retirement 1012 7.55 0.63 5.30 9.4
Maintenance 1012 142.16 50.96 41.08 324.3
Unemployment 1012 45.27 26.98 9.41 159.4
Viol arrest rate 1012 1.80 3.69 0.00 58.0
Local government 1012 0.091 0.026 0.043 0.187
Rooms Equation
Variable Observations Average SD Minimum Maximum
Rooms 1012 569.99 1616.9 0.00 16202
Population 1012 65.56 108.7 5.49 863.29
Popsq 1012 16098 79949 30.2 745268
Density 1012 173.19 287.7 22.34 2178.65
Male 1012 49.40 0.92 47.24 53.95
White 1012 97.05 4.64 72.35 99.91
Black 1012 2.33 4.39 0.00 26.71
Pop 10-19 1012 14.84 0.86 12.71 18.44
Pop 20-29 1012 12.80 2.56 9.00 29.59
Pop 30-39 1012 14.63 1.33 11.10 19.95
Pop 40-49 1012 15.03 1.08 11.17 17.91
Pop 50-64 1012 15.58 1.69 10.41 22.69
Pop over 64 1012 13.35 1.80 7.38 16.87
Real income 1012 14.18 2.17 9.73 25.70
TABLE A2
Crime Regression Results for the Reduced Sample--Property Crimes
Larceny
OLS OLS IV
Density -0.155 (0.06) -0.144 (0.06) -0.119 (0.05)
Male -465.2 * (2.89) -468.5 * (2.91) -476.2 * (2.83)
White -370.3 * (2.62) -372.5 * (2.63) -377.5 * (2.60)
Black -285.2 * (2.04) -285.8 * (2.04) -287.2 * (2.04)
Real income -1.18 (0.03) -0.805 (0.02) 0.080 (0.00)
Retirement -9.51 (0.08) -8.72 (0.07) -6.87 (0.06)
Maintenance 1.40 (0.54) 1.45 (0.56) 1.55 (0.58)
Unemployment 1.761 (0.80) 1.862 (0.85) 2.099 (0.79)
Pop 10-19 -354.7 * (3.27) -354.6 * (3.27) -354.5 * (3.26)
Pop 20-19 -356.9 * (3.63) -355.5 * (3.61) -352.3 * (3.50)
Pop 30-19 -250.0 (1.59) -247.5 (1.57) -241.6 (1.49)
Pop 40-19 -115.5 (1.08) -112.6 (1.05) -105.6 (0.92)
Pop 50-64 -179.8 (1.64) -179.5 (1.64) -179.0 (1.63)
Pop over 64 -209.0 (1.81) -209.2 (1.81) -209.8 (1.81)
Constant 82231 (3.98) 82484 (3.99) 83078 (3.95)
Num obs 497 497 497
Burglary
OLS OLS IV
Density 0.840 (0.84) 0.851 (0.86) 0.811 (0.78)
Male 107.4 (1.69) 104.0 (1.64) 116.2 (1.69)
White 6.77 (0.12) 4.60 (0.08) 12.5 (0.21)
Black 13.4 (0.24) 12.8 (0.23) 15.0 (0.26)
Real income -0.693 (0.04) -0.309 (0.02) -1.704 (0.10)
Retirement 68.9 (1.41) 69.7 (1.43) 66.8 (1.31)
Maintenance -0.602 (0.59) -0.559 (0.55) -0.715 (0.66)
Unemployment 0.485 (0.56) 0.588 (0.68) 0.214 (0.20)
Pop 10-19 -88.0 * (2.06) -87.9 * (2.06) -88.1 * (1.99)
Pop 20-19 -24.9 (0.64) -23.5 (0.61) -28.6 (0.70)
Pop 30-19 -52.8 (0.85) -50.2 (0.81) -59.5 (0.90)
Pop 40-19 12.5 (0.30) 15.6 (0.37) 4.59 (0.10)
Pop 50-64 -4.53 (0.10) -4.29 (0.10) -5.15 (0.11)
Pop over 64 29.1 (0.64) 28.79 (0.63) 29.8 (0.63)
Constant -3997 (0.49) -3740 (0.46) -4676 (0.54)
Num obs 497 497 497
Motor Vehicle Theft
OLS OLS IV
Density 0.065 (0.15) 0.070 (0.16) 0.075 (0.17)
Male 26.2 (0.94) 24.6 (0.88) 23.3 (0.80)
White 31.6 (1.28) 30.6 (1.24) 29.7 (1.18)
Black 37.4 (1.53) 37.1 (1.52) 36.8 (1.51)
Real income 19.16 * (2.71) 19.34 * (2.74) 19.50 * (2.73)
Retirement 21.4 (0.99) 21.8 (1.01) 22.1 (1.02)
Maintenance -0.591 (1.31) -0.571 (1.27) -0.553 (1.19)
Unemployment 0.119 (0.31) 0.168 (0.44) 0.210 (0.46)
Pop 10-19 -10.88 (0.58) -10.85 (0.58) -10.83 (0.57)
Pop 20-19 8.62 (0.50) 9.28 (0.54) 9.86 (0.56)
Pop 30-19 18.40 (0.67) 19.61 (0.72) 20.66 (0.73)
Pop 40-19 37.44 * (2.02) 38.86 * (2.10) -40.10 * (2.00)
Pop 50-64 10.93 (0.57) 11.05 (0.58) 11.14 (0.58)
Pop over 64 24.79 (1.23) 24.67 (1.23) 24.56 (1.22)
Constant -5823 (1.62) -5701 (1.59) -5595 (1.53)
Num obs 497 497 497
Note: t-values are given in parentheses.
* Statistically significant in a two-tailed test at the .95 level