ArticlePDF Available

Crime Rates And Local Labor Market Opportunities In The United States: 1979-1997

Authors:

Abstract and Figures

The labor market prospects of young, unskilled men fell dramatically in the 1980s and improved in the 1990s. Crime rates show a reverse pattern: increasing during the 1980s and falling in the 1990s. Because young, unskilled men commit most crime, this paper seeks to establish a causal relationship between the two trends. Previous work on the relationship between labor markets and crime focused mainly on the relationship between the unemployment rate and crime, and found inconclusive results. In contrast, this paper examines the impact of both wages and unemployment on crime, and uses instrumental variables to establish causality. We conclude that both wages and unemployment are significantly related to crime, but that wages played a larger role in the crime trends over the last few decades. These results are robust to the inclusion of deterrence variables, controls for simultaneity, and controlling for individual and family characteristics. © 2001 by the President and Fellows of Harvard College and the Massachusetts Institute of Technolog
Content may be subject to copyright.
Crime Rates and Local Labor Market Opportunities
in the United States: 1979-19951
September 17, 1998
Eric D. Gould
Hebrew University
mseric@mscc.huji.ac.il
Bruce A. Weinberg
Ohio State University
weinberg.27@osu.edu
David B. Mustard
University of Georgia
mustard@terry.uga.edu
Abstract:
The relationship between crime and labor market conditions is typically studied by
looking at the unemployment rate. In contrast, this paper argues that wages of low-skill
people are better measures of labor market conditions than the unemployment rate. As
the wages of those most likely to commit crime (unskilled men) have been falling in the
past few decades, we examine the impact of this trend on the crime rate giving special
attention to issues of endogeneity. We conclude that wages are a significant determinant
of crime and are more important than the unemployment rate. As theory predicts,
economic factors are more important for property crime than violent crime. These results
are robust to various measures of wages, the inclusion of deterrence variables, controls
for simultaneity, and controlling for individual and family characteristics.
JEL Codes: K4, J0.
1 We appreciate comments from Stephen Bronars, Rachel Friedberg, Richard Freeman,
Saul Lach, and seminar participants at the Labor Studies Group at the NBER Summer
Institute 1998, the AEA Meetings in Chicago 1998, Hebrew University, Tel Aviv
University, University of Georgia, University of Akron, and Ohio State University. We
also thank the Falk Institute for financial support.
1
Section I: Introduction
This paper examines the degree to which changes in crime rates for the United
States from 1979-1995 can be explained by changes in the labor market opportunities for
those most likely to commit crime. From 1980 to 1994, the crime rate in the United
States increased despite the aging of the population and a tripling of the prison population
(DiIulio (1996)). Although most crime is committed by relatively few people who are
multiple offenders, Freeman (1996) shows that the propensity to commit crime for those
not in jail rose precipitously throughout this period.
Economists usually explain crime rates by examining how the propensity to
commit crime responds to the payoffs and punishments of illegal activity (Becker (1968),
Ehrlich (1973, 1996), Levitt (1997)). Most of the existing literature has focused on the
likelihood of apprehension and the severity of punishment, the direct costs of engaging in
crime. This study examines the indirect costs to crime -- the opportunity cost of wages in
the legal sector. More specifically, this study estimates the impact of changing labor
market opportunities for young, unskilled workers on crime rates. This approach is
motivated by two important factors. First, the people most likely to commit crime are
young, unskilled males (Freeman (1996)). Second, the wages and employment rates of
unskilled men have been falling dramatically since the early 1970s (Katz and Murphy
(1992) and Juhn (1992)). These trends have led researchers to implicate the shifting wage
and industrial structure of the economy as a possible explanation for the increasing trends
in crime during this period (Wilson (1996)). This paper investigates whether a causal
relationship can be established.
The existing literature has found moderate, but inconclusive evidence that
unemployment rates are positively associated with crime.2 This paper differs from the
existing literature in two ways. First, instead of concentrating only on the unemployment
rate, we also measure the labor market prospects of potential criminals with the wages of
low-skilled workers. Second, the existing literature fails to consider the potential
2 Freeman (1983) reviews the literature relating crime rates with unemployment and other
labor market variables. None of the studies he reviewed look specifically at the wages of
unskilled workers or the time period covered in this study.
2
endogeneity of crime with the observed labor market outcomes. In contrast, we employ a
variety of instrumental variable strategies to establish the causal relationship from the
changes in labor market conditions to the changes in crime rates.
Although we examine both the wages and the unemployment rates of young
unskilled men, there are several reasons to believe that wages are theoretically a better
measure of their labor market prospects. First, changes in wages are likely to be more
exogenous to the individual than a change in his employment status. Young people enter
and exit the labor force and unemployment for many reasons, most are unrelated to crime.
In contrast, the wage represents the price of the worker’s skills, which is set exogenously
by the market. Secondly, changes in wages directly affect more workers than a change in
the unemployment rate. A change in the market price of skill affects all uneducated
workers, whereas an increase their unemployment rate directly affects only a tiny fraction
of workers who enter unemployment. Thirdly, unemployment is often short-lived and
highly cyclical. Given the potentially long-lasting effects of criminal activity, crime
should be more responsive to long-term changes in labor market conditions than to short-
term fluctuations. For these reasons, wages are more likely to capture the exogenous
changes in long-term labor market opportunities than the unemployment rate, which is
dominated by many short-term choices and influences.
While Freeman (1996), Wilson (1996) and Raphael and Winter-Ebmer speculate
that the declining wages and employment opportunities of unskilled men have contributed
to their increasing involvement in crime, to the best of our knowledge, Grogger (1998) is
the only paper to examine the relationship between wages and crime.3 Grogger (1998)
uses a structural model with individual-level data from the NLSY, and estimates the
relationship between the wage offer and the property crimes committed by the individual.
In contrast, we focus on a variety of property and violent crimes, and exploit the
differences in the timing of wage changes across geographic areas to see if they can
3 Although the focus of their paper is not on wages, Cornwell and Trumbull (1994)
analyze wages in various sectors. However, their paper only looks at counties in North
Carolina for seven years and they aggregate all crimes into one category. We use counties
throughout the whole United States for 17 years and analyze seven types of crime.
3
explain the timing of the changes in various types of crime.4
The analysis uses three basic strategies -- the first two use aggregate data and the
third incorporates individual level data. Our first strategy is to run panel regressions using
annual, county-level data from 1979-1995 with county and time fixed-effects. This
approach exploits year-to-year variation in county-level wages to explain year-to-year
changes in the county-level crime rate. In this way, we control for any time-invariant
local variation in crime that is correlated with local labor market conditions.
The second aggregate analysis performs a ten-year difference (1979-1989)
regression at the Metropolitan Area level to exploit the low frequency variation in the
data. Given the long-term consequences of criminal activity, crime should be more
responsive to low frequency changes in labor market conditions. This approach also
attenuates measurement error problems in panel regression analyses.5 To the extent that
the propensity to commit crime is a function of the rate at which others engage in illegal
activity, these interactions are captured in both of our aggregate analyses.6
While it is natural to use geographic data on labor market conditions since we are
explaining geographic variations in crime rates, our first two strategies allow us to control
only for geographic characteristics. As a third strategy, we use individual level data from
the NLSY79 to test whether local labor market conditions can explain individual criminal
activity, even after controlling for a wealth of personal characteristics such as education,
ability, parental background, etc.
All three strategies indicate that young, unskilled men are responsive to the
opportunity costs of crime. These results are consistent with Grogger (1998), who found
that youth crime was highly responsive to wages.7 However, endogeneity is likely to bias
4 Topel (1994) shows that there are very significant differences in local labor market
conditions.
5 Griliches and Hausman (1986) and Levitt (1995) discussion advantages of the “long
regression” in the presence of measurement error.
6 Ehrlich (1981) argues that additional crime by one person could reduce unexploited
illegal opportunities for others; Sah (1991) emphasizes the effects of higher crime rates
on law enforcement resources; Glaeser, Sacerdote, and Scheinkman (1996) focus on the
effects of one person’s crime on the crime preferences of others in the community.
7 Grogger (1998) finds that a 10% decrease in the potential wages of youths causes a 10%
4
estimates of the relationship between crime and labor market conditions. Cullen and
Levitt (1996) and Willis (1997) argue that high-income individuals or employers leave
areas with higher or increasing crime rates. On the other hand, Roback (1982) maintains
crime rates force employers to pay higher wages as a compensating differential to
workers. Consequently, the direction of the bias is not clear.
We control for potential endogeneity with a number of strategies. First, we use the
changes in the proportion of workers in high-wage industries, rather than changes in the
level of employment in high-wage industries, to proxy for the wage offers of unskilled
men. Although employers may be driven away from high-crime areas, we do not suspect
that employers are driven away disproportionately in high-wage industries. In fact, Willis
(1997) indicates that low-wage employers in the service sector are more likely to relocate
due to increasing crime rates, thus biasing the results against our instrument.
Also, we use wages at the state-level to proxy for the job prospects of young men
at the county level. Although high-income people or employers may leave the county to
avoid higher county crime rates, it is unlikely they would move out of the state. If this
assumption is correct, a high-income person who moves out of the central city to the
suburbs should not affect the average wage in the state, thus leaving the state variable
exogenous to the county crime level.
Last, following the strategy employed by Bartik (1991) and Blanchard and Katz
(1992), we use Census data to generate instruments for the change in labor demand by
interacting three sources of variation that are exogenous to the change in county-level
crime rates: (1) the initial industrial composition within the county, (2) the national trends
in the industrial composition over the sample period, and (3) the demographic changes
within industries at the national level. Our results indicate that endogeneity is not
responsible for the negative relationship between the wages of unskilled workers and the
various crime rates.
The paper is organized as follows. Section II presents general trends in crime
rates, wages, and employment, and discusses our wage proxies. The literature and the
increase in crime.)
5
main issues are described in Section III. Section IV presents the panel regressions using
annual data. Section V presents the ten-year (1979-1989) difference regression analysis.
Section VI shows the results using the individual-level data and Section VII concludes.
Section II: Trends in Crime Rates, Industrial Composition, and Wages
The aggregate crime data, reported to the FBI by local police authorities, come
from the Uniform Crime Reports. Crime rates are offenses per 100,000 people, and the
arrest rates are the ratios of arrests to offenses. Offenses and arrests are reported for the
individual violent crimes (murder, rape, robbery and aggravated assault) and property
crimes (burglary, larceny and auto theft). The violent and property crime indexes
aggregate their respective individual crimes. The overall crime index aggregates all seven
individual crimes. The UCR data are described in more detail in the Appendix.
There are many reasons to be wary of self-reported crime data. First, not every
crime is reported to the police. This under-reporting produces measurement error in the
offense and arrest rates, which could vary by the type of crime or county of jurisdiction.8
Also, the methods of collecting and reporting data vary by local authorities. Although the
accuracy and comparability of self-reported data across counties may be suspect, our
inclusion of county fixed effects eliminates the effects of (time-invariant) cross-county
variations in reporting methods. Furthermore, the trends in the crime rates should be
accurate as long changes in reporting practices are not correlated across counties over
time.9
Figure 1 shows the standardized log offense rates for the property and violent
crime indices for the entire United States. The property crime index follows a cyclical
8 For example, in 1994 the National Criminal Victimization Surveys indicates that 36.1%
of rapes, 40.7% of sexual assaults, 55.4% of robberies, 51.6% of aggravated assaults,
26.8% of personal larcenies without contact, 50.5% of the household burglaries, and
78.2% of motor vehicle thefts and theft attempts were reported. Murder, which has
virtually no underreporting, is not subject to this type of bias. Sourcebook of Criminal
Justice Statistics 1995, Table 3.38, page 250.
9 Ehrlich (1996) discusses reporting biases in the crime data. One method of addressing it
is to work with the logarithms of the crime rates, which are likely to be proportional to
the true crime rates. We use this strategy in this paper.
6
pattern that peaks in 1980, declines by 8% until 1984, increases by 6% until 1991, and
then declines through 1995. The global-peak for property crime in 1980 was
approximately 3% larger than the local-peak in 1991. Property crime increased through
the latter half of the 1980’s, but the absolute levels were not extraordinary.
Although violent crime is also cyclical, the absolute level is more than 10% larger
in 1991 than at the local-peak in 1981. During the whole period, violent crime rose by
14% until 1991, and then steadily declined by 8% as of 1995. Thus, the pattern for violent
crime is much more consistent with the common perception of increasing crime through
the 1980’s and declining since the early 1990’s.10
Eighty-seven percent of all crime is property crime. Therefore, the overall crime
rate is almost identical to the property crime rate.11 Consequently, results for the overall
crime index are dominated by the results for the property crime index. The property crime
is dominated by larceny (67%) and burglary (21%). Thus, results for the property crime
index will be determined by these two categories of crime. Violent crime is composed
mainly by aggravated assault (62%) and robbery (32%), while rape (5%) and murder
(1%) have only a minor influence on the overall violent crime rate. However, their
seriousness gives them a disproportionate influence over social welfare and public policy.
The trends in our panel sample of 352 counties are displayed in Figure 2 and are
similar to the national trends in Figure 1. The sample consists of all counties with a
population over 100,000 in 1989 and with non-missing data for all 17 years from 1979 to
1995. These conditions select large counties where crime rates are the highest and the
data are the most reliable. The total population covered by our sample as of 1995 is over
142 million people. Since we are concentrating on densely populated counties, the trends
are significantly magnified in our sample in comparison to the national trends.12
However, the size of our sample and the shape of the trends (in comparison to Figure 1)
10 Some readers may be surprised that the murder rate hit a global peak in 1980 at 10.2
murders per 100,000 people, and never got above 9.8 which was the second peak in 1991.
11 All percentage breakdowns of the crime indices are from the 1995 Sourcebook of Criminal
Justice Statistics.
12 Glaeser and Sacerdote (1995) explain why crime is more concentrated in highly
populated areas.
7
demonstrate that our sample is representative of the entire United States.13
So far we have looked only at the raw crime data with no adjustments for changes
in the demographic compositions within each county. Figure 3 plots the property and
violent crime trends after adjusting for changes in the age, sex, and racial composition.
After controlling for these factors, the trends for both types of crime rose steadily
throughout the 1980’s and peaked in the early 1990’s. In 1991, the adjusted property
crime rate hits a global-peak at 15% higher than the local-peak in 1980, and 20% higher
than it was at the beginning of the period in 1979. The upward trend in unadjusted
violent crime found before in Figure 2 is now accentuated as the adjusted rate rose by
over 45% until 1991, when it started to decline. Figure 3 demonstrates that changes in
demographics explain much of the decline in both types of crime during the 1990’s.
However, after making these demographic adjustments, both crime series indicate a
dramatic increase in the incidence of crime during this time period, particularly since
1984. The individual property and violent crimes (adjusted and adjusted) are depicted in
Figures 4-7.
As the adjusted property and violent crime rates increased, the labor market
prospects for young, unskilled men deteriorated. In Figure 7, the average wages of all
workers and of non-college, male workers (workers with a high-school degree or less) are
plotted over time. The average wage of non-college men declined quickly from 1979 to
1982, leveled off until 1988, and then declined rapidly throughout the remainder of the
sample period. This 17% overall decline in unskilled wages represents a significant fall
in the opportunity cost of crime in the legal sector, and as Juhn (1992) shows, this wage
decline was not offset by an increase in employment.
For two of our three basic analyses, we use the wages of unskilled men directly in
our regressions. In our analysis of the 1979-89 differences between Census years, we
calculate the wages of unskilled men in each area directly from the Census. In our
analysis of individual crime behavior using NLSY79, local wage measures are also
calculated from the 1980 Census. Unfortunately, annual wage measures for specific
13 Levitt (1997) shows similar trends using the same data for 59 large cities.
8
demographic groups are not available at the county level for all 17 years. Consequently,
we are forced to develop wage proxies for our panel analysis that includes all 17 years.
One of our proxies for the wages of unskilled men is the retail wage.14 The retail wage is
the lowest of the eight industrial sectors identified in the county-level data, and
employees in the retail sector are younger and less skilled than those in other sectors. To
test whether the retail wage is a good proxy, we performed a ten-year difference
regression (1979-1989) using Census data of the average wages of non-college men on
the average retail wage at the MA level. The regression yielded a point estimate of 0.78
(standard error = 0.04) and an R-squared of 0.71. Therefore, changes in the retail wage
are a powerful proxy for changes in the wages of non-college men. Figure 9 presents the
downward trend in the average retail wage in our county sample over time. From 1979 to
1995, the average retail wage fell by almost 14%, but the trend is not steady. The retail
wage declined until 1982, increased until 1986, and then continually fell through 1995.
This pattern is very similar to the trend in wages for non-college men in Figure 8.
Our second proxy for the annual county-level wages of unskilled men is the
percent of workers employed in high-wage industries, which are determined by dividing
the eight major industries into two categories based on their average wage within the
industry. Consequently, the average wage in all high-wage industries (Manufacturing,
Wholesale, Transportation, Construction) is about 50% larger than the average wage in
low-wage industries (Retail, Services, FIRE, and Government).15 As with wages, we do
not have county-level industry employment figures for specific demographic groups, but
the overall industrial shifts during this period reduced the wages of less-skilled men
relative to other groups (Katz and Murphy (1992)). So given our classification, a shift
from high-wage to low-wage industries represents a shift from high-wage to low-wage
employment for less-skilled men. Using Census data, a ten-year difference regression
14 All income and employment variables are from the Regional Economic Information System
(REIS), a component of the Bureau of Commerce. Workers were categorized into eight
industries: Manufacturing, Wholesale, Transportation, Construction, Retail, Services, FIRE, and
Government. Due to missing values, Agriculture was excluded.
15 Average wages for each industry were calculated in the same manner described for the
retail wage.
9
(1979-1989) of the average wages of non-college men on the fraction of high-wage
employment at the MA level yields a coefficient of 1.91 (standard error = 0.21) and an R-
squared of 0.28. Therefore, the percent of high-wage employment appears to be a
reasonable proxy for the wages of unskilled men, although less powerful than the retail
wage.
Figure 10 shows the proportion of workers employed in high-wage industries over
time. At its peak, employment in high-wage industries was only 35% of the workforce,
and this percentage declined over time by over 7 percentage points. This trend may
understate the effects on the opportunities for young workers if firms reduced
employment by laying off or reducing hires of young workers. Even ignoring this
possibility, employment opportunities in high wage sectors clearly declined over time.
The sample period clearly shows that the propensity to commit crime went up at
the same time that the labor market conditions for unskilled men deteriorated. These
trends seem to be related, particularly since young, unskilled men are the most likely to
commit crime.16 However, the timing of both of these trends could be spuriously
correlated. The goal of this paper is to establish whether the relationship is causal.
Section III: Crime and the Labor Market
The traditional economic approach to crime models the decision to commit crime
within the context of utility maximization -- a risk-neutral person commits crime if the
expected benefits outweigh the expected costs. The classic papers in this area by Becker
(1968) and Ehrlich (1973) focused more on the direct costs to committing crime,
measured by the probability of getting caught and the severity of punishment. These
direct costs differ by the type of crime committed, and data on them are much easier to
obtain than data on the potential benefits to crime. Although the causality could go in
either direction, the literature has been mostly concerned with establishing the
relationship between deterrence variables and the crime rate.17
16 Freeman (1996) reports that two-thirds of prison inmates in 1991 had not graduated
from high school.
17 Levitt (1997) is the most recent attempt to untangle the effect of police size on crime.
10
Less attention has been given to the direct benefits to crime, primarily because of
the lack of data. The direct gains from crime differ depending on the nature of the crime.
Some crimes (such as robbery, larceny, burglary and auto theft) can be used for self-
enrichment, whereas other crimes (murder, rape and assault) are much less likely to yield
material gains to the offender.18 Offenders who commit the latter crimes are more likely
to derive benefits from interdependencies in utility with the victim. This notion of
interdependence of utility between offender and victim for certain crimes is supported by
the fact that murder, rape, and assault occur frequently between people who know each
other, whereas the victim and offender have no relationship in the vast majority of
property crimes.19
In addition to the direct costs (the probability and severity of punishment), there
are also indirect costs to crime. Engaging in criminal activity jeopardizes one’s prospects
in the legal labor market. Engaging in criminal activity indirectly diverts time and
resources away from investing in human capital, thus leading to a loss of potential wages.
Also, the time involved in committing crime results in a direct loss of opportunity wages.
Time spent in prison also entails a loss of opportunity wages.
The degree to which legal alternatives affect criminal behavior also depends on
the type of crime. Legal-sector opportunities should be a greater factor for burglary and
larceny, than rape and murder, where pecuniary considerations are lower. However,
holding everything else constant, a reduction in legal opportunities should make one more
likely to engage in any form of criminal activity, regardless of motives, due to the forgone
18 For example, in 1992 the average monetary loss was $483, $840, $1278 and $4713 for
larceny, robbery, burglary and auto theft, respectively, compared with average monetary
losses of $27 and $89 for rape and murder. Crime in the United States 1992.
19 For offenses committed in 1993 the offenders were classified as non-strangers to the
victims in 74.2% of rapes, 51.9% of assaults, and 19.9% of robberies (1994 Sourcebook
of Criminal Justice Statistics, p. 235, Table 3.11). Historically murder victims knew their
offenders (Supplementary Homicide Reports). During the 1990s this relationship has
changed, and now slightly less than half of the murder victims know their offenders. For
example, in 1993, 47.7% of all murders were committed by people who were known to
the victim, 14.0% were committed by strangers, and in 39.3% of the cases the
relationship between victim and offender was unknown (Crime in the United States 1993,
p. 20, Table 2.12).
11
earnings during the crime and potentially in jail.
Freeman (1994, 1996) summarized the literature on the effects of labor market
opportunities on crime rates. Most of the literature discusses the relationship between
unemployment and crime rates. The results are inconclusive, although they generally
point to a positive, but small, relationship between the two. Although there have been
substantial earnings changes in the last few decades, including the declining absolute and
relative wages of unskilled men who are most likely to commit crime, little research has
looked at the relationship between wages and crime (except for Grogger (1998)).20
For many reasons, wages are likely to be a better measure for the legal opportunity
costs of crime than unemployment.21 As Grogger (1998) points out, many criminals
commit crime while employed, and committing crime does not preclude holding a job. In
addition, workers move in and out of the work force and unemployment for many
reasons, most of them unrelated to the decision to commit crime (Topel and Ward
(1992)). Unemployment is also highly cyclical, and for most workers, is temporary and
does not measure their long-term legal-sector prospects. The decision to engage in crime
can have long-term consequences, and may not be affected by short-term fluctuations in
the unemployment rate. Also, the decision to be in the labor force or unemployed is a
choice, and is therefore endogenous to many other factors, including the decision to
commit crime. Furthermore, changes in the unemployment rate will only directly affect
the tiny fraction of workers who enter unemployment, whereas a decrease in the price of
skill will affect all workers in that skill group. Finally, wages are more exogenous to the
individual since the worker has some control over his employment status, but cannot
affect how the market rewards his basket of skills.22 For these reasons, the trend in wages
are likely to better capture the exogenous changes in the labor market prospects of
20 Lochner (1999) argued that labor-market ability, even more than wages, lowers crime.
21 Wages could also be correlated with reputational sanctions. See Lott (1992).
22 Endogeneity problems will arise if his choice of working affects the level of his skills
by affecting his investment in human capital. For this reason, we use wage residuals in
the empirical section to abstract from changes in observed levels of skills, and therefore
measure the changes in the structure of skill prices. Although the results for the levels are
not presented, the results are similar either way.
12
unskilled men.23
To our knowledge, Grogger’s (1998) study of young men using individual-level
data from the NLSY is the only paper that has focused on wages as a determinant of
crime. Grogger estimates a structural model of time spent in the criminal sector and the
legitimate labor market sector. He found that the criminal participation of young men in
crime is positively related to their potential wages, explaining “three-quarters of the
observed rise in youth crime" (page 32).
This paper is the first to study this issue at an aggregate level with a non-structural
approach. Both approaches have to rely on several untested assumptions and deal with
their own data problems, and therefore, should be considered complements to each other.
One big difference is that Grogger’s study is limited to property crime, whereas we
examine both property and violent crime, and the individual categories within these
indices. In addition, Grogger studies the relationship between the individual’s wage and
his criminal activity, whereas our identification comes from whether changes in local
labor market conditions are associated with changes in crime rates. This approach
permits us to control for time-invariant regional differences in crime and labor market
conditions, which otherwise may lead to a spurious correlation between crime and wages.
The use of aggregate-level data has further advantages. The recent crime
literature has emphasized the external effects of one person’s criminal activity on peer
activity (Glaeser, Sacerdote, and Scheinkman (1996)). If the psychic costs to crime are
lower when others are committing more crimes (i.e. people have less shame when others
are doing it), the crime decision of one person can affect the level of crime of others.24
Since an individual-level analysis is limited to a selected sample, this external effect may
not be captured within the sample. An aggregate-level analysis will capture both the
direct effect of an individual’s decision to commit crime and the external effect, although
it will not separately identify the two effects.
An aggregate analysis also captures other environmental factors in the decision to
23 Lott (1992) argues that reputational sanctions are positively correlated with the wage.
24 An increase in aggregate crime can also affect the decision of the individual in a more
indirect way if it leads to reduction in the likelihood of apprehension. See Sah (1991).
13
commit crime. Since most criminals are young men, the age and sex distribution will be
important contributing factors, which may be magnified or attenuated by external effects
when they are concentrated together.25 Different age and gender groups may also be
characterized as easier targets of crime. Thus, as we saw in Section II, which showed
large differences between the adjusted and unadjusted crime trends, it is important to
control for demographic changes to identify the effects of the changes in wages.
Some aggregate characteristics will have ambiguous effects on the level of crime.
For example, a decline in the standard of living could be considered a decline in the labor
market opportunities of the workers in that area, and therefore lead to greater crime.
However, general economic welfare may affect the opportunities for property crime. If
there is less material wealth to steal, then the crime rate may decline if general economic
conditions deteriorate. Our empirical strategy isolates the effect of the changes in wages
of those most likely to commit crime -- unskilled men -- after controlling for the changes
in the general economic prosperity of the area.
The crime rate in an area could be affected by its income distribution. However,
isolating this effect introduces a variety of problems establishing the direction of
causality. Although high-income people have more wealth to steal (leading to higher
crime), they also have the resources to self-protect with garages, alarms, guards, and other
measures. 26 They also have the resources to move out of high-crime areas, thus reducing
the observed average wage in response to crime (Cullen and Levitt (1996)). If employers
leave areas in response to increasing crime rates, this could be another mechanism that
leads higher crime rates to cause a decrease in wages (Willis (1997). However, if
remaining employers raise wages in high-crime areas as a compensating differential, then
higher crime rates could cause wages to rise, even in the face of an out-migration of
workers (Roback (1982)).
To identify whether the declining wages of unskilled men entice them into
25 Economists have not explored why most crime is committed by men. Concerning the
age issue, Grogger (1997) contends that crime declines as age increases because age is a
proxy for the wage.
26 Lott and Mustard (1997) and Ayres and Levitt (1998) show that self-protection lowers
14
criminal activity, the empirical strategy must establish the direction of causality. To do
this, we need instruments correlated with the changes in crime within an area only
through changes in the wages within the area. In the next section, we develop an
empirical framework to control for the demographic changes, isolate the effects of the
declining wages of unskilled men on crime, and control for potential endogeneity.
Section IV: Analysis Using Annual Data, 1979-1995
This section provides an empirical analysis of the preceding discussion, and uses
17 years of panel data (1979-1995) for 352 counties described in Section II. In each
specification, county fixed-effects control for much of the cross-sectional variation as we
explain the within-county trends in the crime rates over this period. Yearly fixed-effects
are also included to take out the national trends. We expect that the labor market
variables can help explain the national trends, but we identify the effects of these
variables from the within-county deviations from the national trends to abstract from any
spurious correlation at the national level. Because demographic changes will alter the
costs and benefits to crime as described in Section III, each specification also controls for
changes in the age, sex, and race composition of the county.
The empirical strategy identifies the importance of the labor market opportunity
costs for those most likely to commit crime. As unskilled men commit most crime, two
variables are used to directly proxy for their opportunity wages, since direct measures for
the wages of unskilled men are not available at the county level on an annual basis. The
unemployment rate for unskilled workers is not available on an annual basis, and
therefore, is not included in this analysis. As discussed in Section II, the wage proxies for
young, unskilled men at the county level are the retail wage and the percent of workers
employed in high-wage industries. To control for the general level of prosperity in the
county, we use log per capita income in the county.
Table 1 displays the coefficient estimates for various combinations of the labor
market variables after controlling for demographic changes. We present each combination
crime by carrying concealed weapons and purchasing Lojack, respectively.
15
to show how sensitive the labor market proxies are to each other, so the effect of each is
clearly identified. The first specification for the two indices (columns 1 and 6) shows that
each is very responsive to the retail wage when other labor market controls are excluded.
The coefficient for property crime (-0.429) is a bit larger than the estimate for violent
crime (-0.301), and both are statistically significant. Since the retail wage declined on
average by 13.6%, these coefficient estimates predict a 5.9% (-13.6 multiplied by -0.429)
rise in the property crime rate and a 4.1% rise in violent crime due to the decline in the
retail wage. These “predicted” effects, evaluated at the average change in the
independent variables between 1979 and 1995, are reported below each coefficient in
brackets (the standard errors are below the coefficients in parentheses).
The second specification in Table 1 (columns 2 and 7) shows the effect of the
retail wage on crime after controlling for the overall level of welfare in the county as
proxied by the income per capita of the county. For property crime, the retail wage effect
is reduced somewhat by the income per capita variable as it falls from -0.429 to -0.313,
but it still remains statistically significant. The coefficient on income per capita is also
significant at -0.249. The negative coefficient on income per capita could mean that the
costs of property crime represented by this variable are stronger than the potential benefits
it also represents. However, if the retail wage measures the wages of unskilled workers
imperfectly, per capita income may be picking up some of the variation in the wages of
just those workers who are most likely to commit crime. Another possibility, addressed
later in this section, is whether this result stems from endogeneity.
For violent crime, adding the income per capita variable does not affect the retail
wage coefficient (it changes from -0.301 to -0.295). Unlike the result for property crime,
the coefficient for income per capita is insignificant for the violent crime index, thus
suggesting that the overall welfare of the county is not an important determinant of the
violent crime rate. Like property crime, violent crime is significantly affected by the retail
wage, which proxies for the wage level offered to those most likely to commit crime.
Since the retail wage is not a perfect measure of the wages of young, unskilled
workers, we use our second proxy -- the percent of workers in high-wage industries.
Columns 3 and 8 in Table 1 present the effect of this variable when the other labor market
16
variables are excluded. The next columns (4 and 9) show what happens when income per
capita is added to the specification. This wage proxy is inversely related to property
crime and positively related to violent crime. For both indices, the addition of the income
per capita measure has no effect, although the income per capita measure is significantly
negative now for both crime indices (before it was significant only for property crime).
The positive effect of this wage proxy for unskilled men for violent crime is contrary to
the hypothesis that declining wages entice unskilled workers into a life of crime, but as
we will see in the next table, this result is not robust for the individual violent crimes.
The last specification for each crime index in Table 1 uses all three labor market
variables together -- income per capita, the retail wage, and the percent of high-wage
workers. The coefficient estimates for each of the variables are not very affected by the
inclusion of the other variables. For property crime, the coefficient on the percent of
high-wage workers is virtually identical to when the retail wage was excluded, and the
retail wage coefficient gets a bit larger in magnitude. Both of these variables, as well as
the negative coefficient on income per capita, are statistically significant. The results for
violent crime follow a similar pattern. Both proxies for the wages of unskilled workers
are unaffected by the inclusion of the other.
These results suggest that our two wage proxies pick up potentially important, but
different aspects of our targeted variable -- the opportunity labor market cost of crime for
unskilled workers. These results could stem from the fact that both are imperfect
measures, or the retail wage may pick up the current opportunity wage and the percent of
high-wage workers is proxying more for the long-term opportunity wage. Since they are
unaffected by the inclusion of the other, we use them together as our “core” specification.
However, either variable can be excluded with little effect on the other coefficients.
Table 2 presents the “core” specification for each crime classification within each
crime index. This table shows that the retail wage is significant statistically and
economically for each individual property and violent crime. Counties with a larger than
average drop in the retail wage experience higher than average increases in each type of
crime, with an elasticity response of roughly 0.30% for both the property and violent
crime indices. Within property crime, the biggest effect is on burglary (-0.593) and
17
larceny (-0.299). The coefficient on auto theft is significantly negative (-0.172), but is
small and is counteracted by the positive coefficient estimate on the percent of workers in
high-wage industries (0.018). The overall effect of both wage proxies, evaluated at their
average change between 1979 and 1995, is to increase burglary by 12.7% (8.1% from the
retail wage and 4.6% from the percent of high wage workers), increase larceny by 10.2%,
and decrease auto theft by 11.4%. Although the results for auto theft run counter to the
proposed hypothesis, because burglary and larceny compose 88% of property crime, they
are very significant factors for property crime as a whole. Since the coefficient estimates
on both proxies are similar when they are specified alone, the results for auto theft
suggest that the retail wage and the percent of workers in high-wage industries pick up
very different phenomena.
For violent crime in Table 2, the retail wage coefficient is significantly negative
for all the individual crimes, while the percent of workers in high-wage industries is
insignificant. The combined effects of both variables, evaluated at their mean changes
over time, are to increase aggravated assault by 3.1%, murder by 7.4%, robbery by 5.4%,
and rape by 8.8%. All of these numbers would be stronger if we looked only at the effect
of the retail wage, since the coefficient on the percent of workers in high-wage industries,
although not significant, is generally positive and therefore decreases violent crime.
Table 3 re-runs the regressions in Table 2 but includes the arrest rates as
independent variables. Missing values for arrest rates are more numerous than for
offense rates, so the sample is reduced to 245 counties. The inclusion of the arrest rates
does not meaningfully alter the coefficient estimates on the labor market variables. The
magnitude and statistical significance of the retail wage variable for all the individual
property and violent crimes are comparable to Table 2. The same is true for the percent
of workers in high-wage industries. The arrest rates have a large and significantly
negative effect for every classification of crime. Because the numerator of the dependent
variable appears in the denominator of the arrest rate (the arrest rate is defined as the ratio
of total arrests to total offenses), measurement error in the offense rate leads to a
18
downward bias in the coefficient estimates of the arrest rates (“division bias”).27
However, the results indicate that the labor market coefficients are robust to the usual
inclusion of the arrest rates, as well as the decrease in the number of counties. To work
with the broadest sample possible, the remaining specifications exclude the arrest rates.28
Up to now, our results may be contaminated by the endogeneity of crime and
observed wages at the county level. An increase in crime may lead employers to relocate,
thus reducing labor demand and wages within the county (Willis (1997)). Similarly, if
avoiding crime is a normal good, an increase in crime will cause high-wage individuals to
move out of the county (Cullen and Levitt (1996)). However, it is likely that crime-
induced migration will occur mostly across county lines within states rather than across
states. High-wage earners may leave the county because of increases in the crime rate,
but their decision to leave the state is exogenous to increases in the crime rate. Under this
assumption, higher crime rates will reduce (measured) county-level wages, but have little
impact on state-level wages. Therefore, to control for the endogeneity between county-
level wages and crime, we use the average wage for all workers in the state as a substitute
for income per capita for the county, and substitute the average wage for non-college men
in the state for the retail wage in the county. Another advantage of this procedure is that
it is possible to estimate the annual wages of specific demographic groups at the state
level using the CPS, whereas before we had to proxy for the wages of unskilled men at
the county level. Although we could use a two-step IV approach with these state-wide
variables, we prefer to substitute them straight into the equation as independent variables
because the retail wage was only a proxy measure of the variable of interest.
Endogeneity is less of a problem with our second proxy for the county-level
27 Levitt (1995) analyzes this issue and why the relationship between arrest rates and
offense rates is so strong.
28 Ideally we would like to incorporate additional deterrence variables into the analysis.
Unfortunately, conviction and sentencing data are available only at the county level for
four states (Mustard (1999)). However, the exclusion of these deterrence variables does
not bias our results for the economic variables, because they are likely to be uncorrelated,
as the results in Table 3 demonstrate with the arrest rates. The use of further deterrence
variables will also open up a variety of endogeneity issues which are beyond the scope of
this study (Levitt (1997) for a study of the endogeneity of crime and police force size).
19
wages of unskilled men -- the percent of workers employed in high-wage industries.
Although increases in crime rates may cause existing employers to relocate and new
employers to locate elsewhere, this will alter the level of employment but not necessarily
the share of employment in high-wage industries. Unless increases in crime drive out
high-wage employment disproportionately, the percent of employment in high-wage
industries should be exogenous to the crime rate. The results in Willis (1997) suggest
that employment is unrelated to property crime and that violent crime drives out low-
wage employment in services much more than in high-wage industries.29 Thus, using the
percent of workers in high-wage industries may bias our results against finding significant
negative effects for this variable -- which explains why the estimated coefficient on this
variable is frequently positive (although not significant) in Tables 2 and 3 for violent
crime.
Table 4 presents the regression results using the state-wide wage variables and the
county-level percent of workers in high-wage industries. Our wage measures are the
residuals from regressing individual wages from the CPS on education, experience,
experience squared, and controls for race and marital status. The construction of these
variables is described in detail in Appendix II. Using the residuals allows us to abstract
from wage changes due to changes in observable characteristics of workers, and thus
more accurately reflect changes in the structure of wages. Using residuals attenuates any
remaining endogeneity issues where the county crime rate affects the state-level
composition of workers. However, very similar results are obtained by using the state
wages rather than the residuals.
Comparing the coefficient estimate for the non-college wage in the state in Table
4 to the coefficient on the retail wage for the county in Table 2, the coefficient estimates
are larger and still significant for both of the crime indices. For the property index, the
retail wage coefficient was -0.333 in Table 2, while the non-college state wage is -0.483
29 Using data from neighborhoods in Los Angeles, Willis (1997) reports that an increase
of violent crime is associated with an decrease of 14 jobs per square mile. Broken down
by industry, 9 of those jobs are lost in services and “all other” employment, 2 are lost in
manufacturing, and one each is lost in wholesale and transportation.
20
in Table 4 (both are significant). For the violent crime index, the comparable estimates
are -0.282 in Table 2 compared to -0.460 in Table 4. The state wage for non-college
workers is considerably stronger than the county retail wage for auto theft, burglary,
aggravated assault, and robbery. The coefficient flips signs for murder and rape but is
insignificant for murder. This result suggests that endogeneity may be a larger issue for
murder and rape, or for all crimes, the state-level wages capture different aspects of the
wages of unskilled men at the county level than the county-level retail wage.
The results for the percent of workers in high-wage industries are very similar to
those in Table 2. Again, excluding this variable does not significantly affect the
coefficient estimates on the state wage variables. The state wage for all workers has a
small effect for most crimes, which suggests that the negative effect for property crimes
using county-level income per capita in Tables 1 and 3 may be due to the endogenous
outmigration of high-income people from high-crime areas instead of the imperfect nature
of the retail wage as a proxy for the wages of unskilled men.
The combined effect of both proxies for the wages of unskilled workers in Table 4
predicts a 10% increase in property crime and a 3.6% increase in violent crime, when
evaluated at their average within-county change over the sample period. Since the
adjusted index for all property crime increased by only 16.5% over the sample period, the
predicted increase in property crime from the two wage proxies explains about 60% of
the increase. For violent crime, the two variables explain roughly 8.8% of the 41%
increase in the adjusted series.
For the individual crimes, the combined effects are the largest for burglary
(predicting a 14.1% increase), larceny (8.8% increase), aggravated assault (5.3%
increase), and robbery (8.0% increase). Apart from auto theft, these offenses typically
have an economic motive, in contrast to rape or murder. Therefore, we expect our wage
proxies to be stronger factors in these crimes than for rape and murder. The auto theft
result for the percent of high-wage industries is a bit of a puzzle, but as noted before, the
wage measure of unskilled men in Table 4 and the retail wage (from earlier tables) have
the expected signs and even after excluding the percent of high-wage workers.
Table 4 tells the same basic story--for the most common crimes in each category
21
(burglary, larceny, aggravated assault, and robbery), the wage proxies of unskilled men
are important factors. The similarity of the results for the retail wage and the state wage
of non-college men suggests that endogeneity is not responsible for the negative
relationship between the wages of less-skilled workers and the increase in crime. The
results are more negative with the state wages, which is the opposite direction of the bias
predicted by the endogeneity argument. This suggests that the state-level wages for
unskilled workers are better measures for the wages of unskilled workers within the
county than the county-level retail wage.
Although the point estimates are comparable in magnitude for violent crime and
property crime as a whole, violent crime increased more dramatically than property crime
during this period. Therefore, the decline in the wages for unskilled men explains up to
60% of the increase in adjusted property crime and only 8% of the increase in adjusted
violent crime. 30 The findings for the individual crimes are consistent with the hypothesis
that monetary incentives play a larger role for economically motivated crimes such as
burglary and robbery than for murder and rape. Furthermore, these results are robust to
the inclusion of the arrest rates, a decrease in sample size, and several experiments with
the specification of our labor market proxies.
Section V: Analysis of 10-Year Differences, 1979-1989
This section studies the relationship between economic conditions and crime rates
using ten-year differences, which has many implications. First, it emphasizes low-
frequency variations in economic conditions. The previous analysis was based on annual
data and thus explained changes in crime rates within a given year with the
contemporaneous change in our wage proxies. Given measurement error in our
independent variables, long-term changes may suffer less from attenuation bias than
estimates based on annual data (Griliches and Hausman (1986) and Levitt (1995)). Also,
30 We compare the predicted change in crime to the “adjusted” increase in crime after
controlling for changes in demographics. As noted in Section II, the “adjusted” crime
indexes grew much faster than the unadjusted.
22
given the long-term consequences of criminal activity, crime should be more responsive
to low-frequency changes in labor market conditions. The model estimated in the
previous section was conservative in the sense that it did not estimate the full effect of
lagged changes in wages on the current changes in crime. The “long regression” strategy
employed here is less demanding on the identification of the full effects of changes in our
wage measures on crime, even in the absence of measurement error.
The use of two Census years (1979 and 1989) for our end points has two further
advantages. First, it is possible to construct detailed measures of labor market conditions
for specific demographic groups, which we were unable to do on an annual basis with
county-level data. Second, we can better link each county to the appropriate local labor
market in which it resides. In most cases, the relevant labor market exceeds the county of
residence. Therefore, labor market conditions in each county are measured using labor
market variables for the SMSA/CMSA in which it lies. Consequently, the sample in this
analysis is restricted to those that lie within metropolitan areas.31
To measure the labor market prospects of potential criminals, we compute the log
wage of non-college men after controlling for observable characteristics and the
employment rate of non-college men. To control for changes in the standard of living on
criminal opportunities, we include the mean log household income in the MA. The
construction of these variables is discussed in Appendix III. The regressions also control
for the same set of demographic changes employed in the previous section. The estimates
presented here are for the ten-year differences of the dependent variables on similar
differences in the independent variables. Thus, analogous to the previous section, our
estimates are based on cross-county variations in the changes in economic conditions
after eliminating fixed-county effects and aggregate time-series effects. Cross-sectional
31We also constructed labor market variables at the county group level. Unfortunately,
due to changes in the way the Census identified counties in both years, the 1980 and 1990
samples of county groups are not comparable, introducing noise in our measures. The
sample in this section differs from that in the previous section in that it (1) excludes
counties not in MA’s (or are in MA’s which are not identified on both the 1980 and 1990
PUMS 5% samples); (2) includes counties with fewer than 100,000 residents which are
within MA’s.
23
models not reported here yield similar estimates. Table 5 presents descriptive statistics.
Table 6 presents weighted least squares results for the indices and individual
crimes.32 We focus on property crimes before considering violent crimes. The estimates
indicate a strong positive effect of household income on crime rates, which is consistent
with household income as a measure of criminal opportunities. The wages of non-college
men have a large negative effect on property crimes. The estimated elasticities range from
-0.785 for larceny to -2.282 for auto theft. The 13.7% drop in wages for non-college men
between 1979 and 1989 is associated with a 13.9% increase in overall property crimes.
The unemployment rate among non-college men has a large positive effect on property
crimes. Previous researchers have often found a weak relationship between crime and
unemployment rates (Freeman (1994) surveys the literature). Including wage measures
reduces the estimated effects of unemployment, making these strong effects notable. The
estimated elasticities for unemployment range between 2.2 and 2.5. However,
unemployment increased by only 0.6% over this period, so changes in unemployment
rates are responsible for small changes in crime rates (between 1 and 2 log points). Given
the large variations in crime (Glaeser, Sacerdote, and Scheinkman (1996)), the economic
variables explain a reasonable amount of the variations in crime.
With violent crime, the estimates for aggravated assault and robbery are quite
similar to those for the property crimes. Given the pecuniary motives for robbery, this
similarity is expected and resembles the results in the previous section. Some assaults
may occur during property crimes, leading them to share some of the characteristics of
property crimes. Because assault and robbery constitute 94% of violent crimes, the
violent crime index follows the same pattern. The crimes with the weakest pecuniary
motive are murder and rape. As expected, the relationship between the economic
variables and these crimes is quite weak. None of the variables is significant in the
murder equation. The negative relationship between household income and rape
indicates that rape declines with income. In general, the weak relationship between our
economic variables and murder and rape suggests that our conclusions are not due to a
32 Although not presented, similar regressions which include the fraction of households
24
spurious correlation between economic conditions and crime rates generally.
As discussed previously, we are concerned that our results may be biased upward
or downward by causality running from criminal activity to labor market conditions. We
use instrumental variables to address the direction of causality. Following Bartik (1991)
and Blanchard and Katz (1992), we develop instruments for the change in labor demand
within each MA. The idea is to interact three sources of variation that are completely
exogenous to the change in crime within each MA: (1) the initial industrial composition,
(2) the national industrial composition trends, and (3) the demographic composition
trends within each industry at the national level.33
An example with two industries provides the intuition behind the instruments.
Autos (computers) constitute a large share of employment in Detroit (San Francisco).
The national employment trends in these industries are markedly different. Therefore, the
decline in the auto industry’s share of national employment will adversely affect Detroit’s
demand for labor more than San Francisco’s. Conversely, the growth of the high-tech
sector at the national level translates into a much larger positive effect on San Francisco’s
demand for labor than Detroit’s. In addition, if biased technological change causes the
auto industry to reduce its employment of unskilled men, this affects the demand for
unskilled labor in Detroit more than in San Francisco. A formal derivation of the
instruments is in Appendix IV. We obtain eight instruments, which are used to identify
exogenous variation in the three labor market variables. After controlling for the
demographic variables, the partial R2 between our set of instruments and the three labor
market variables ranges from 0.242 to 0.327.34
The IV estimates in Table 7 show no effect of household income on crime,
with female heads and the fraction beneath the poverty line yield similar results.
33 Bartik (1991) and Blanchard and Katz (1992) interact the first two sources of
variation. Since we instrument for labor market conditions of specific demographic
groups, we also exploit cross-industry variations in the changes in industrial shares of 4
demographic groups (gender interacted with educational attainment).
34 This instrument strategy is more appropriate for identifying long-term changes in labor
demand rather than picking up year-to-year variation.
25
whereas the WLS estimates show a strong relationship between household income and
crime. These IV estimates are similar to those from the annual sample in the previous
section.35 One explanation for the difference between the IV and WLS estimates is that
high crime rates may force employers to raise wages, causing the WLS estimates to be
biased upwards. Only in the case of auto theft does mean household income remain
positive and significant. Mean household income may measure auto theft opportunities
better than the opportunities for other crimes. The IV estimates of the wage and
unemployment rates of non-college men are quite similar to the WLS estimates,
indicating that reverse causality is not responsible for these effects.
Among the violent crimes, the IV estimates for aggravated assault, murder, and
rape are generally smaller and estimated imprecisely. As before, robbery follows the
same pattern as the property crimes. High income levels reduce rape although higher
wages for non-college men are associated with higher rape rates.
To summarize this section, the use of ten-year changes enables us to exploit the
low-frequency relation between wages and crime, and use direct measures for the wages
and unemployment rates of unskilled workers calculated from the Census. Increases in
the unemployment of non-college men increase property crime, while increases in their
wages reduce property crime. Violent crimes are less sensitive to economic conditions
than property crimes. Our estimates for property crime exceed those from annual data,
which is consistent with a greater impact of long term economic changes on crime or
measurement error in our independent variables. Using IV methods to control for reverse
causality has little effect on the relationship between the wages and unemployment rates
for less-skilled men and crime rates. Our estimates imply that declines in labor market
opportunities of less-skilled men were responsible for substantial increases in property
crime.
35 The specification in Table 7 instruments for all three labor market variables. The
coefficient estimates are very similar if we instrument only for either one of the three
variables.
26
Section VI. Analysis Using Individual-level Data
The results at the county and MA-levels have shown that aggregate crime rates
respond to labor market conditions in the expected manner. Aggregate crime data are
attractive because they show how the criminal behavior of the entire population responds
to labor market conditions. However, it is impossible to control for detailed individual
characteristics without individual data. In this section, we link individual data on criminal
behavior of male youths from the NLSY79 to labor market conditions measured at the
state level. The goal is to see whether local labor market conditions still have an effect on
each individual, even after controlling for his education, cognitive ability, parental
background, etc.
The analysis explains criminal activities by each male individual such as
shoplifting, theft of goods worth less than and more than $50, robbery (“using force to
obtain things”), and the fraction of individual income from crime.36 We focus on these
offenses because the NLSY does not ask about murder and rape and our previous results
indicate that crimes with a monetary incentive are more sensitive to changes in wages and
employment than other crime. The data come from self-reporting of the number of times
individuals engaged in various forms of crime during the twelve months prior to the 1980
interview. While we expect economic conditions to have the greatest effect on less skilled
workers, we calculate average wages and unemployment rates for non-college and college
educated men in each state (from the 1980 Census), and see if they can explain the
criminal activity of each individual. To allow the effects to vary by the education of the
individual, we interact labor market conditions (and household income, which is included
to capture criminal opportunities) with the respondent’s educational attainment. The level
of criminal activity of person i is modeled as follows:
36 The means for the crime variables are: 1.24 times for shoplifting; 1 time for stealing
goods worth less than $50; .41 times for stealing goods worth $50 or more; and .32 times
for robbery. On average, 5% of the respondent’s income was from crime.
27
ii
COL
siiCOL
COL
siiCOLsiiCOL
HS
siiHS
HS
siiHSsiiHSi
ZUCOLWCOLIncHouseCOL
UHSWHSIncHouseHSCrime
εβββ βββ +Γ++++ ++=
321
321 .
i
HS and i
COL are indicator variables for whether the respondent had no more than a
high school diploma or some college or more as of May 1, 1979. si
IncHouse denotes the
mean log household income in the respondent’s state of birth (to avoid endogenous
mobility); and HS
si
HS
si UW, ( COL
si
COL
si UW,) denote the regression-adjusted mean log wage
and unemployment rate of high school (college) men in the respondent’s state of birth.
Our measures of individual characteristics, i
Z, include years of school (within college
and non-college), AFQT, mother’s education, family income, family size, age, race, and
Hispanic background. Details of the sample construction are in Appendix V.
Table 8 presents the results. For shoplifting and both measures of theft, the
economic variables consistently have the expected signs and are generally statistically
significant among less educated workers. Thus, higher household incomes raise property
crime among less educated men as do lower wages and higher unemployment rates for
less educated men. The implied effects of the change in economic conditions (reported in
brackets) are generally quite large for household income and wages, but much smaller for
unemployment. As expected, economic conditions have no effect on criminal activity for
more highly educated workers.37 The estimates are insignificant and often have
unexpected signs. Robbery estimates are insignificant and have the wrong signs, however
robbery is the least common crime in the sample. The fraction of total income from crime
also shows the expected pattern for less educated workers and the estimates are
statistically significant. (The numerator and denominator are both affected by labor
market conditions to reinforce our results). Among the other covariates (not reported),
age and education are typically significant – crime increases with age in this young
37 When labor market conditions for low education workers are used for both groups, the
estimates for college workers remain weak, indicating that the difference between the two
groups is not due to the use of separate labor market variables. Estimates based on
educational attainment at age 25 are similar to those reported.
28
sample, and decreases with education.38 The other covariates are generally not significant.
The analysis in this section strongly supports our previous findings that labor
market conditions are important determinants of criminal behavior. Low-skilled workers
are clearly the most affected by the changes in labor market opportunities, and these
results are robust to controlling for a wealth of personal and family characteristics. In
addition, this section helps explain why previous studies that have concentrated the
unemployment rate have failed to show any consistent relationship between crime and
labor market variables.
Section VII: Conclusion
The existing economic literature on crime emphasizes the effects of deterrence
measures. Studies that examine the effects of labor market conditions typically do so
using the unemployment rate. Such studies frequently find a weak positive relationship
between crime and unemployment. To the best of our knowledge, no work controls for
the potential endogeneity between crime and labor market conditions.
We study the effects of economic conditions on crime by using the wages and
unemployment of the group most likely to commit crime -- young, unskilled men. We
show that wages are theoretically and empirically a better measure for the opportunity
cost of crime than the unemployment rate, which helps explain why previous studies have
found mixed results for explaining crime rates with labor market variables. We also
employ a number of strategies to control for endogeneity.
The empirical strategy uses time-series variation within a geographic area (after
controlling for national trends) to identify the effect of various measures for the
opportunity cost of crime for young, unskilled men. Our results indicate that economic
conditions are important determinants of crime. After controlling for endogeneity, our IV
estimates from the “long regression” indicate that the wage declines of unskilled men
have contributed to a 13.5% increase in burglary, 7.1% increase in larceny, 9.2% increase
in aggravated assault, and an 18% increase in robbery. Similar results were obtained from
38 Estimates are similar when the sample is restricted to respondents 18 and over.
29
panel regressions with annual data. These four categories represent 88% of all property
crime and 94% of all violent crime. Although the unemployment rate is a significant
factor, the average increase in unemployment was very small. Therefore, the “predicted”
increase in most crimes due to the increase in unemployment is in the 1% to 2% range.
As expected, economic conditions have a larger impact on crimes with a pecuniary
motive than crimes such as rape and murder where monetary considerations are smaller.
These findings are robust to various measures for the opportunity costs of crime, the
inclusion of arrest rates, controls for endogeneity, both the short and long time period
regression strategies, and controlling for personal and family characteristics.
30
References
Ayres, Ian; and Levitt, Steven D. “Measuring Positive Externalities from Unobservable
Victim Precaution: An Empirical Analysis of Lojack.” Quarterly Journal of
Economics, February 1998, 113, 43-77.
Bartik, Timothy J. Who Benefits from State and Local Economic Development Policies?
Kalamazoo, MI: W. E. Upjohn Institute for Employment Research, 1991.
Becker, Gary. “Crime and Punishment: An Economic Approach.” Journal of Political
Economy, 1968, 76:2, 169-217.
Blanchard, Oliver Jean; and Katz, Lawrence F. “Regional Evolutions.” Brookings Papers
on Economic Activity, 1992, 0:1, 1-69.
Cornwell, Christopher; and Trumbull, William N. "Estimating the Economic Model of
Crime with Panel Data." The Review of Economics and Statistics, 1994, 76:2,
360-366.
Crime in the United States. Washington, DC: U.S. Department of Justice, Federal Bureau
of Investigation, 1992-1993.
Cullen, Julie Berry; and Levitt, Steven D. “Crime, Urban Flight, and the Consequences
for Cities,” NBER Working Paper #5737, September 1996.
DiIulio, John, Jr. “Help Wanted: Economists, Crime and Public Policy.” Journal of
Economic Perspectives, Winter 1996, 10, 3-24.
Ehrlich, Isaac. “Crime, Punishment, and the Market for Offenses.” Journal of Economic
Perspectives, Winter 1996, 10, 43-68.
Ehrlich, Isaac. “On the Usefulness of Controlling Individuals: An Economic Analysis of
Rehabilitation, Incapacitation, and Deterrence.” American Economic Review, June
1981, 307-322.
Ehrlich, Isaac. “Participation in Illegitimate Activities: A Theoretical and Empirical
Investigation.” Journal of Political Economy, 1973, 81:3, 521-565.
Freeman, Richard B. “Why Do So Many Young American Men Commit Crimes and
What Might We Do About It?” Journal of Economic Perspectives, Winter 1996,
10, 25-42.
Freeman, Richard B. “The Labor Market.” In Crime, edited by James Q. Wilson and
Joan Petersilia. San Francisco: Institute for Contemporary Studies Press, 1995,
171-191.
Freeman, Richard B. “Crime and the Job Market.” NBER Working Paper #4910, 1994.
Freeman, Richard B. “Crime and the Employment of Disadvantaged Youths.” In Urban
Labor Markets and Job Opportunity, edited by George Peterson and Wayne
Vroman, Urban Institute, 1992, 201-237.
Freeman, Richard B. “Crime and Unemployment.” In Crime and Public Policy, edited by
James Q. Wilson. San Francisco: Institute for Contemporary Studies Press, 1983.
Glaeser, Edward; and Sacerdote, Bruce. “Why Is There More Crime in Cities?”
Working Paper, November 1995.
Glaeser, Edward; Sacerdote, Bruce; and Scheinkman, Jose. “Crime and Social
Interactions." Quarterly Journal of Economics, 1996, 111, 507-548.
Griliches, Zvi; and Hausman, Jerry. “Errors in Variables in Panel Data.” Journal of
Econometrics, 1986, 31, 93-118.
31
Grogger, Jeff. “Market Wages and Youth Crime.” Journal of Law and Economics,
October 1998, 16, 756-791.
Hashimoto, Masanori. “The Minimum Wage Law and Youth Crimes: Time Series
Evidence.” Journal of Law and Economics, 1987, 30, 443-464.
Juhn, Chinhui. “Decline of Male Labor Market Participation: The Role of Declining
Market Opportunities,” Quarterly Journal of Economics, February 1992, 107, 79-
121.
Katz, Lawrence F.; and Murphy, Kevin M. “Changes in Relative Wages, 1965-1987:
Supply and Demand Factors.” Quarterly Journal of Economics, February 1992,
107, 35-78.
Levitt, Steven. “Using Electoral Cycles in Police Hiring to Estimate the Effect of Police
on Crime.” American Economic Review, June 1997, 87, 270-290.
Levitt, Steven. “Why Do Increased Arrest Rates Appear to Reduce Crime: Deterrence,
Incapacitation, or Measurement Error?” NBER Working Paper #5268, 1995.
Lochner, Lance. “Education, Work and Crime: Theory and Evidence.” Working Paper,
September 1999.
Lott, John R., Jr. “An Attempt at Measuring the Total Monetary Penalty from Drug
Convictions: The Importance of an Individual’s Reputation.” Journal of Legal
Studies, January 1992, 21, 159-187.
Lott, John R., Jr.; and Mustard, David B. “Crime, Deterrence, and Right-to-Carry
Concealed Handguns.” Journal of Legal Studies, January 1997, 26, 1-68.
Mustard, David. “Re-examining Criminal Behavior: The Importance of Omitted Variable
Bias.” Working Paper, September 1999.
Raphael, Steven; and Winter-Ebmer, Rudolf (1999) "Identifying the Effect of
Unemployment on Crime." Working Paper, August 1998.
Roback, Jennifer. “Wages, Rents, and the Quality of Life.” Journal of Political Economy,
1982, 90:6, 1257-78.
Sah, Raaj. “Social Osmosis and Patterns of Crime.” Journal of Political Economy, 1991,
99:6, 1272-95.
Sourcebook of Criminal Justice Statistics. Washington, DC: US Department of Justice,
Bureau of Justice Statistics, 1994-1995.
Topel, Robert H. “Wage Inequality and Regional Labour Market Performance in the US.”
In Labour Market and Economic Performance: Europe, Japan and the USA
edited by Toshiaki Tachibanaki. New York: St. Martin’s Press, 1994, 93-127.
Topel, Robert H.; and Ward, Michael P. “Job Mobility and the Careers of Young Men.”
Quarterly Journal of Economics, 1992, 107:2, 439-79.
Uniform Crime Reports, Federal Bureau of Investigation. Washington, DC.
Willis, Michael. “The Relationship Between Crime and Jobs.” University of California--
Santa Barbara, Working Paper, May 1997.
Wilson, James Q.; and Herrnstein, Richard J. Crime and Human Nature, New York:
Simon & Schuster, 1985.
Wilson, William Julius. When Work Disappears: The World of the New Urban Poor,
New York: Alfred A. Knopf, 1996.
32
Appendix
Appendix I. The UCR Crime Data
The number of arrests and offenses from 1977-1995 was obtained from the
Federal Bureau of Investigation's Uniform Crime Reporting Program, a cooperative
statistical effort of over 16,000 city, county, and state law enforcement agencies. These
agencies voluntarily report the offenses and arrests in their respective jurisdictions. For
each crime, the agencies record only the most serious offense during the crime. For
instance, if a murder is committed during a bank robbery, only the murder is recorded.
Robbery, burglary and larceny are often mistaken for each other. Robbery, which
includes attempted robbery, is the stealing, taking, or attempting to take anything of value
from the care, custody, or control of a person or persons by force, threat of force or
violence, and/or by putting the victim in fear. There are seven types of robbery: street and
highway, commercial house, residence, convenience store, gas or service station, bank,
and miscellaneous. Burglary is the unlawful entry of a structure to commit a felony or
theft. There are three types of burglary: forcible entry, unlawful entry where no force is
used, and attempted forcible entry. Larceny is the unlawful taking, carrying, leading, or
riding away of property or articles of value from the possession or constructive
possession of another. Larceny is not committed by force, violence or fraud. Attempted
larcenies are included. Embezzlement, "con" games, forgery and worthless checks are
excluded. There are nine types of larceny: items taken from motor vehicles, shoplifting,
taking motor vehicle accessories, taking from buildings, bicycle theft, pocket picking,
purse snatching, theft from coin operated vending machines, and all others.39
When zero crimes were reported for a given crime type, the crime rate was
changed to 0.1 before taking the natural log of the crime rate. Since 1985, some counties
have shown zeros for all their offense and arrest categories, even though they should have
been recorded as missing. This occurred because the ICPSR has been unable to
distinguish the FBI's legitimate values of 0 from values of 0 that should be missing. To
39 Appendix II of Crime in the United States contains these definitions and the definitions of the
33
address this, we assigned missing values for all offense and arrest rates when the numbers
of all offenses and arrests were all 0 for a given county in a given year. If an individual
offense or arrest category had a value of 0 and that county had non-zero values for other
crime categories, we used the raw data.
Appendix II. Construction of the State-level Wage Residuals from the CPS
To construct the CPS data set we used the merged outgoing rotation groups for
1979-1995. The data on each survey correspond to the week prior to the survey. We
employed these data rather than the March CPS because the outgoing rotation groups
contain approximately three times as many observations as the March CPS. Unlike the
March CPS, non-labor income is not available on the outgoing rotation groups surveys.
We use the log weekly wages of non-college men and of all workers after
controlling for observable characteristics. We estimate wages for workers who worked or
held a job in the week prior to the survey. The sample was restricted to those who usually
work 35 or more hours a week, were between 18 and 65, and were working in the private
sector (non self-employed) or for government (the universe for the earnings questions).
To estimate weekly wages, we used the edited earnings per week for workers paid
weekly, and used the product of usual weekly hours and the hourly wage for those paid
hourly. Those with top-coded weekly earnings were assumed to have earnings 1.5 times
the top-code value. All earnings figures were deflated using the CPI-U to 1982-1984=
100. Workers whose earnings were beneath $35 per week in 1982-1984 terms were
deleted from the sample, as were those with imputed values for the earnings questions.
To control for changes in the human capital stock of the workforce, we control for
observable worker characteristics when estimating wages. We employ a two-step
procedure. In the first stage, we regress log weekly wages upon individual worker
characteristics: years of completed schooling, a quartic in potential experience, and
dummy variables for Hispanic, black and marital status. We estimate a separate model for
each gender and education group and each year, which permits the effects of each
other offenses.
34
explanatory variable to vary across the four gender-education groups and to change over
time. Adjusted wages were estimated using the mean residuals from these regressions.
Appendix III. Description of the Census Data
We employed the 5% sample of the 1980 and 1990 Censuses to estimate the mean
log weekly wages of non-college men, the unemployment rate among non-college men,
and the mean log household income in each MA for 1979 and 1989. We also used the
Census to estimate industry employment shares at the national and MA levels and the
employment shares for each gender-education group within each industry which were
used to construct the industrial composition instruments.
Wage information is from the wage and salary income in the year prior to the
survey. For 1980, we restrict the sample to persons between 18 and 65 who worked at
least one week, were in the labor force for 40 or more weeks and usually worked 35 or
more hours per week. The 1990 census does not provide data on weeks unemployed. To
generate an equivalent sample of high labor force attachment individuals, we restrict the
sample to people who worked 20 or more weeks in 1989 and who usually worked 35 or
more hours per week. People currently enrolled in school were eliminated from the
sample in both years. Individuals with positive farm or non-farm self-employment
income were excluded from the sample. We excluded people who earned less than $40
per week in 1979 dollars and those whose weekly earnings exceeded $2,500 per week. In
the 1980 census, people with top-coded earnings were assumed to have earnings 1.45
times the top-coded value. The 1990 census imputes individuals with top-coded earnings
to the median value for those with top-coded earnings in the state. These values were
used. Individuals with imputed earnings (non top-coded), labor force status, or individual
characteristics were excluded from the sample. The procedures used to control for
individual characteristics in the CPS were also used for the Census data.40
The mean log household income was estimated using the income reported for the
year prior to the survey for persons not living in group quarters. We estimate the
40 The 1990 Census categorizes schooling according to the degree earned. Dummy
35
employment status of non-college men using the current employment status because the
1990 Census provides no information about weeks unemployed in 1989. The sample is
restricted to people between 18 and 65 not currently enrolled in school. The
unemployment rate was constructed as the number of unemployed people divided by the
number of people in the labor force.
The industry composition instruments require industry employment shares at the
national and MA level and the employment shares of each gender-education group within
each industry. These were estimated from the Census. The sample included all persons
between 18 and 65 not currently enrolled in school who resided in MA’s. Individuals
with imputed industry affiliations were dropped from the sample. Our classification has
69 industries at roughly the 2-digit level of the SIC. Individuals were weighted using the
person weight in the 1990 Census. The 1980 Census is a flat sample.
Appendix IV. Construction of the MA Level Instruments from the Census
This section outlines the construction of the instruments for labor demand. We
exploit inter-city variations in industrial composition interacted with industrial
differences in growth and technological change favoring particular groups to construct
instruments for the change in demand for labor of all workers and workers in particular
groups at the city (MA) level.
Let fict| denote industry i’s share of the employment at time t in city c. This
expression can be read as the employment share of industry i conditional on the city and
time period. Let fi t| denote industry i’s share of the employment at time t for the nation.
The growth in industry i’s employment nationally is given by
GROW f
f
i
i
io
≡ −
|
|
11
Our instrument for the change in total labor demand in city c is
GROW TOTAL fGROW
cici
i
|0.
variables were included for each educational category.
36
We estimate the growth in total labor demand in city c by taking the weighted average of
the national industry growth rates. The weights for each city correspond to the initial
industry employment shares in the city. These instruments are analogous to those in
Bartik (1991) and Blanchard and Katz (1992).
We also construct instruments for the change in demand for labor in four
demographic groups. Our groups are defined on the basis of gender and education (non-
college educated and college educated). Let fg cti| denote demographic group g’s share of
the employment in industry i at time t in city c ( fgt| for the whole nation). Group g’s
share of the employment in city c at time t is given by
f f f
gct g cti ict
i
| | |
.
The change in group g’s share of employment can be decomposed as
(
)
(
)
f f f f f f f f
gcgcgci i cic
igcigci i c
i
| | | | | | | |1 0 0 1 0 1 0 1
− ≡
.
The first term reflects the effects of industry growth rates. The second term reflects
changes in each group’s share of employment within industries. The latter can be thought
of as arising from industrial differences in biased technological change.
In estimating each term, we replace the MA-specific variables with analogues
constructed from national data. All cross-MA variation in the instruments is due to cross-
MA variations in initial industry employment shares. In estimating the effects of industry
growth the demand for labor of each group, we replace the MA-specific employment
shares ( fgci|0) with national employment shares ( fgi|0 ). We also replace the actual end of
period shares ( fic|1) with estimates ( $|
fic1). Our estimate of the growth term is
(
)
GROW f f f
gc gi i cic
i
≡ −
|0 | |
$1 0 .
The date 1 industry employment shares for each MA are estimated using the industry’s
initial employment share in the MA and the industry’s employment growth nationally.
$|
|
|
ffGROW
fGROW
gc
ici
jcj
j
1
0
0
=.
37
To estimate the effects of biased technology change, we take the weighted average of the
changes in each group’s national employment share
(
)
TECH f f f
gc gigi i c
i
≡ −
||0 |1 0 .
The weights correspond to the industry’s initial share of employment in the MA.
Appendix IV. NLSY79 Sample
This section describes the NLSY79 sample. The sample included all male
respondents with valid responses for the variables used in the analysis (the crime
questions, education in 1979, AFQT, mother’s education, family size, age, black and
Hispanic background; a dummy variable for family income missing was included in order
to include respondents without valid data for family income). The number of times each
crime was committed was reported in bracketed intervals. Our codes were as follows: 0
for no times; 1 for 1 time; 2 for 2 times; 4 for 3 to 5 times; 8 for 6 to 10 times; 20 for 11
to 50 times; an 50 for 50 or more times. Following Grogger (1998) we code the fraction
of income from crime as: 0 for none; .1 for very little; .25 for about one quarter; .5 for
about one half; .75 for about three quarters; and .9 for almost all.
38
Figure 1: United States National Trends in Crime Indices
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
1979
1981
1983
1985
1987
1989
1991
1993
1995
Violent Crime
Property Crime
Plotted values are the log offense rate (offenses per 100,000 people) relative to the year 1979.
The Property Crime Index is the sum of Auto Theft, Burglary, and Larceny. The Violent Crime
Index is the sum of Aggravated Assault, Robbery, Murder, and Rape.
Figure 2: County Sample Trends in Unadjusted Crime Indices
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
1979 1981 1983 1985 1987 1989 1991 1993 1995
Overall Index
Property Index
Violent Index
Plotted values are the coefficients on the time dummies from regressions of the log offense rates
on time dummies and county fixed effects for 352 counties with population over 100,000.
Population was also used as weights.
39
Figure 3: County Sample Trends in Adjusted Crime Indices
-0.1
0
0.1
0.2
0.3
0.4
0.5
1979 1981 1983 1985 1987 1989 1991 1993 1995
Overall Index
Property Index
Violent Index
Plotted values are the coefficients on the time dummies of regressions of the log offense rates on
time dummies, county fixed effects, and controls for age distribution (using the percent of the
population in five different age groups), the sex composition, the percentage of the population
that is black, and the percentage that is neither white nor black. The sample consists of 352
counties with population over 100,000. Population was also used as weights.
Figure 4: County Sample Trends in Unadjusted Property Crimes
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
1979 1981 1983 1985 1987 1989 1991 1993 1995
Auto Theft
Burglary
Larceny
Plotted values are the coefficients on the time dummies of regressions of the log offense rates on
time dummies and county fixed effects for 352 counties with population over 100,000.
Population was also used as weights.
40
Figure 5: County Sample Trends in Adjusted Property Crimes
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
1979 1981 1983 1985 1987 1989 1991 1993 1995
Auto Theft
Burglary
Larceny
Plotted values are the coefficients on the time dummies of regressions of the log offense rates on
time dummies, county fixed effects, and controls for age distribution (using the percent of the
population in five different age groups), the sex composition, the percentage of the population
that is black, and the percentage that is neither white nor black. The sample consists of 352
counties with population over 100,000. Population was also used as weights.
Figure 6: County Sample Trends in Unadjusted Violent Crimes
-0.2
-0.1
0
0.1
0.2
0.3
0.4
1979 1981 1983 1985 1987 1989 1991 1993 1995
Aggravated Assault
Murder
Robbery
Rape
Plotted values are the coefficients on the time dummies of regressions of the log offense rates on
time dummies and county fixed effects for 352 counties with population over 100,000.
Population was also used as weights.
41
Figure 7: County Sample Trends in Adjusted Violent Crimes
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
1979 1981 1983 1985 1987 1989 1991 1993 1995
Aggravated Assault
Murder
Robbery
Rape
Plotted values are the coefficients on the time dummies of regressions of the log offense rates on
time dummies, county fixed effects, and controls for age distribution (using the percent of the
population in five different age groups), the sex composition, the percentage of the population
that is black, and the percentage that is neither white nor black. The sample consists of 352
counties with population over 100,000. Population was also used as weights.
Figure 8: Standardized Log Average Wages Over Time (CPS data)
-0.2
-0.15
-0.1
-0.05
0
0.05
1979
1981
1983
1985
1987
1989
1991
1993
1995
All Full-Time Workers
Full-Time Non-College Men
Plotted values are the coefficients of regressing log wages on year-specific dummy variables.
42
Figure 9: Log of Retail Income Per Retail Worker Over Time
-0.14
-0.12
-0.1
-0.08
-0.06
-0.04
-0.02
0
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
Each point represents the coefficients from a county-level weighted regression of log retail
income per retail worker on time dummy variables with average population for the county used
as weights.
Figure 10: Percent of All Workers Employed In High-Wage Industries
20
22
24
26
28
30
32
34
36
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
High Wage industries include manufacturing, wholesale, transportation, and construction. Low
wage industries include retail, services, FIRE, and government. Agriculture has been excluded
due to missing values. Percentages were calculated by taking a weighted average of the county-
level observations using the average county population as weights.
Table 1: County Level Regressions for Property and Violent Crime Offense Rate Indexes, 1979-1995
Property
Crime
Index
Property
Crime
Index
Property
Crime
Index
Property
Crime
Index
Property
Crime
Index
Violent
Crime
Index
Violent
Crime
Index
Violent
Crime
Index
Violent
Crime
Index
Violent
Crime
Index
Log Income Per
Capita -0.249**
(0.051)
[-5.1]
-0.377**
(0.046)
[-7.7]
-0.206**
(0.052)
[-4.2]
-0.013
(0.075)
[-0.3]
-0.185**
(0.067)
[-3.8]
-0.040
(0.076)
[-0.8]
Log Retail Income
Per Retail Worker -0.429**
(0.044)
[5.9]
-0.313**
(0.050)
[4.3]
-0.333**
(0.050)
[4.5]
-0.301**
(0.063)
[4.1]
-0.295**
(0.072)
[4.0]
-0.282**
(0.073)
[3.8]
Percent of All
Workers in High
Wage Industries
-0.005**
(0.001)
[3.8]
-0.004**
(0.001)
[3.0]
-0.004**
(0.001)
[3.0]
0.003*
(0.001)
[-2.3]
0.003**
(0.001)
[-2.3]
0.003*
(0.002)
[-2.3]
Observations 5984 5984 5984 5984 5984 5984 5984 5984 5984 5984
Partial R20.016 0.020 0.004 0.015 0.023 0.004 0.004 0.001 0.002 0.004
** indicates significance at the 5% level. * indicates significance at the 10% level. Standard errors in parentheses. Numbers in
brackets represent the “predicted” percent increase of the crime rate due to the mean change in the independent variable, computed by
multiplying the coefficient estimate by the mean change in the independent variable between 1979-1995 (multiplied by 100).
Observations are for 352 counties for 17 years. Regressions include county and time fixed effects and demographic controls (percent
of population age 10-19, age 20-29, age 30-39, age 40-49, age 50-64, and age 65 and over, percent male, percent black, and percent
non-black and non-white). Regressions weighted by mean of population size of each county. Partial R2 are after controlling for county
and time fixed effects, and demographic controls.
Table 2: County Level Regressions for Various Offense Rates, 1979-1995
Overall
Crime
Index
Property
Crime
Index
Auto
Theft Burglary Larceny Violent
Crime
Index
Aggravated
Assault Murder Robbery Rape
Log Income Per
Capita -0.178**
(0.051)
[-3.7]
-0.206**
(0.052)
[-4.2]
-0.347**
(0.092)
[-7.1]
-0.240**
(0.064)
[-4.9]
-0.156**
(0.052)
[-3.2]
-0.040
(0.076)
[-0.8]
0.197**
(0.096)
[4.0]
0.494**
(0.195)
[10.1]
-0.158*
(0.096)
[-3.2]
0.315
(0.104)
[6.5]
Log Retail Income
Per Retail Worker -0.346**
(0.049)
[4.7]
-0.333**
(0.050)
[4.5]
-0.172**
(0.088)
[2.3]
-0.593**
(0.064)
[8.1]
-0.299**
(0.050)
[4.1]
-0.282**
(0.073)
[3.8]
-0.227**
(0.092)
[3.1]
-0.759**
(0.186)
[10.4]
-0.393**
(0.092)
[5.4]
-0.597**
(0.099)
[8.1]
Percent of All
Workers in High
Wage Industries
-0.005**
(0.001)
[3.8]
-0.004**
(0.001)
[3.0]
0.018**
(0.001)
[-13.7]
-0.006**
(0.001)
[4.6]
-0.008**
(0.001)
[6.1]
0.003*
(0.002)
[-2.3]
-0.000
(0.001)
[0.0]
0.004
(0.004)
[-3.0]
-0.000
(0.002)
[0.0]
-0.001
(0.002)
[0.7]
Observations 5984 5984 5984 5984 5984 5984 5984 5984 5984 5984
Partial R20.023 0.023 0.020 0.033 0.025 0.004 0.001 0.003 0.006 0.006
See notes to Table 1.
Table 3: County Level Regressions for Various Offense Rates with Arrest Rates, 1979-1995
Overall
Crime
Index
Property
Crime
Index
Auto
Theft Burglary Larceny Violent
Crime
Index
Aggravated
Assault Murder Robbery Rape
Log Income Per
Capita -0.276**
(0.054)
[-5.7]
-0.313**
(0.054)
[-6.4]
-0.504**
(0.102)
[-10.3]
-0.376**
(0.070)
[-7.7]
-0.259**
(0.054)
[-5.3]
-0.091
(0.075)
[-1.9]
0.211**
(0.097)
[4.3]
0.354**
(0.149)
[7.3]
-0.284**
(0.094)
[-5.8]
0.109
(0.097)
[2.2]
Log Retail Income
Per Retail Worker -0.309**
(0.053)
[4.2]
-0.273**
(0.053)
[3.7]
-0.122
(0.100)
[1.7]
-0.558**
(0.068)
[7.6]
-0.224**
(0.054)
[3.1]
-0.260**
(0.074)
[3.5]
-0.244**
(0.095)
[3.3]
-0.405**
(0.146)
[5.5]
-0.417**
(0.092)
[5.7]
-0.327**
(0.095)
[4.5]
Percent of All
Workers in High
Wage Industries
-0.004**
(0.001)
[3.0]
-0.004**
(0.001)
[3.0]
0.024**
(0.002)
[-18.2]
-0.005**
(0.001)
[3.8]
-0.008**
(0.001)
[6.1]
0.005**
(0.002)
[-3.8]
0.000
(0.002)
[0.0]
0.015**
(0.003)
[-11.4]
0.002
(0.002)
[-1.5]
0.002
(0.002)
[-1.5]
Arrests per Offenses -0.002**
(0.0003) -0.009**
(0.001) -0.004**
(0.0005) -0.012**
(0.001) -0.007**
(0.0005) -0.003**
(0.0002) -0.003**
(0.0002) -0.002**
(0.0002) -0.006**
(0.0003) -0.004**
(0.0002)
Observations 4165 4165 4165 4165 4165 4165 4165 4165 4165 4165
Partial R20.047 0.088 0.052 0.112 0.090 0.056 0.047 0.039 0.089 0.086
Sample includes 245 counties for 17 years (1979-1995). See notes to Table 1 for further details.
Table 4: County Level Regressions for Various Offense Rates Using State Variables From the CPS, 1979-1995.
Overall
Crime
Index
Property
Crime
Index
Auto
Theft Burglary Larceny Violent
Crime
Index
Aggravated
Assault Murder Robbery Rape
Log Mean Weekly
Wage of All
Workers in the State
(Residuals)
-0.086
(0.161)
[-0.4]
-0.048
(0.163)
[-0.2]
-0.364
(0.283)
[-1.8]
0.039*
(0.213)
[0.2]
-0.224
(0.163)
[-1.1]
-0.211
(0.237)
[-1.1]
-0.431
(0.299)
[-2.1]
-1.606**
(0.608)
[-8.0]
-0.112
(0.301)
[-0.6]
-0.551
(0.325)
[-2.7]
Log Mean Weekly
Wage of Non-
College Men in the
State (Residuals)
-0.479**
(0.116)
[5.4]
-0.483**
(0.118)
[5.4]
-1.405**
(0.204)
[15.7]
-0.718**
(0.154)
[8.0]
-0.178
(0.118)
[2.0]
-0.460**
(0.171)
[5.1]
-0.408*
(0.216)
[4.6]
0.563
(0.438)
[-6.3]
-0.583**
(0.217)
[6.5]
1.113**
(0.234)
[-12.4]
Percent of All
Workers in High
Wage Industries in
the County
-0.006**
(0.001)
[4.6]
-0.006**
(0.001)
[4.6]
0.014**
(0.002)
[-10.6]
-0.008**
(0.001)
[6.1]
-0.009**
(0.001)
[6.8]
0.002
(0.001)
[-1.5]
-0.001
(0.002)
[0.7]
0.005
(0.004)
[-3.8]
-0.002
(0.002)
[1.5]
0.001
(0.002)
[-0.8]
Observations 5984 5984 5984 5984 5984 5984 5984 5984 5984 5984
Partial R20.018 0.017 0.054 0.012 0.017 0.009 0.007 0.002 0.006 0.007
See notes to Tables 1. Residual wages for non-college men were obtained by regressing individual wages on education levels,
experience, experience squared, and dummy variables for Hispanic, Black, and marital status.
Table 5: Descriptive Statistics, 1979-1989 Changes. Mean Standard Deviation
Change in Real Mean Log Wages of Non-College Men -.137 .091
Change in Real Mean Log Household Income .087 .118
Change in Unemployment Rate of Non-College Men .006 .025
Log Changes in Crime Rates (Unadjusted)
Index Crime -.008 .299
Property Crime -.028 .304
Auto Theft .167 .489
Burglary -.251 .361
Larceny .021 .308
Violent Crime .172 .382
Aggravated Assault .281 .498
Murder -.135 .804
Robbery .018 .547
Rape .075 .577
Counties weighted by mean of 1979 and 1989 populations. Sample includes 582 counties that lie within MA’s. Counties with missing
crime data are excluded from sample as are counties that lie within MA’s which are not identified in both the 1980 and 1990 PUMS
5% samples. Changes in crime rates are not adjusted for changes in demographic composition.
Table 6: Effects of Economic Conditions on Crime Rates. Ten Year Differences, 1979-1989. Weighted Least Squares Estimates.
Overall
Crime
Index
Property
Crime
Index
Auto
Theft Burglary Larceny Violent
Crime
Index
Aggravated
Assault Murder Robbery Rape
Change in Mean 0.692** 0.671** 2.088** 0.207 0.566** 0.770** 1.050** 0.679 0.853** -0.860**
Log Household (0.187) (0.190) (0.300) (0.221) (0.195) (0.266) (0.346) (0.578) (0.378) (0.399
Income in MA [6.0] [5.8] [18.2] [1.8] [4.9] [6.7] [9.1] [5.9] [7.4] [-7.5]
Change in Mean -1.008** -1.015** -2.282** -0.976** -0.785** -0.784** -0.861** 0.065 -1.016** 0.774**
Log Weekly Wage (0.207) (0.210) (0.333) (0.244) (0.216) (0.295) (0.384) (0.640) (0.419) (0.442)
of Non-College Men
in MA (Residuals) [13.8] [13.9] [31.3] [13.4] [10.8] [10.7] [11.8] [-0.9] [13.9] [-10.6]
Unemployment Rate 2.102** 2.226** 2.516** 2.540** 2.160** 0.776 0.694 0.805 2.920** -1.874*
of Non-College Men (0.463) (0.471) (0.746) (0.548) (0.484) (0.661) (0.860) (1.435) (0.939) (0.992)
[1.3] [1.4] [1.6] [1.6] [1.3] [0.5] [0.4] [0.5] [1.8] [-1.2]
Observations 582 582 582 582 582 582 582 582 582 582
Partial R20.074 0.075 0.108 0.072 0.056 0.019 0.017 0.005 0.028 0.014
** indicates significance at the 5% level. * indicates significance at the 10% level. Standard errors in parentheses. Numbers in
brackets represent the “predicted” percent increase of the crime rate due to the mean change in the independent variable, computed by
multiplying the coefficient estimate by the mean change in the independent variable between 1979-1989 (multiplied by 100).
Dependent variable is log change in crime rate from 1979-1989 in county. Sample consists of 582 counties. Regressions include
county and time fixed effects and demographic controls (percent of population age 10-19, age 20-29, age 30-39, age 40-49, age 50-64,
and age 65 and over, percent male, percent black, and percent non-black and non-white). Regressions weighted by mean of population
size of each county. Partial R-squares are after controlling for county and time fixed effects, and demographic controls. Wage
residuals control for educational attainment, a quartic in potential experience, Hispanic background, black, and marital status.
Table 7: Effects of Economic Conditions on Crime Rates. Ten Year Differences, 1979-1989. Instrumental Variables Estimates.
Overall
Crime
Index
Property
Crime
Index
Auto
Theft Burglary Larceny Violent
Crime
Index
Aggravated
Assault Murder Robbery Rape
Change in Mean -0.013 -0.035 1.688** -0.818* -0.146 -0.167 0.499 0.778 -0.103 -3.192**
Log Household (0.382) (0.389) (0.608) (0.459) (0.399) (0.543) (0.701) (1.166) (0.772) (0.828)
Income in MA [-0.1] [-0.3] [14.7] [-7.1] [-1.3] [-1.5] [4.3] [6.8] [-0.9] [-27.8]
Change in Mean -0.868** -0.874** -2.528** -0.987** -0.521 -0.406 -0.673 0.684 -1.341* 2.340**
Log Weekly Wage (0.364) (0.370) (0.579) (0.467) (0.379) (0.517) (0.667) (1.110) (0.735) (0.789)
of Non-College Men
in MA (Residuals) [11.9] [12.0] [34.6] [13.5] [7.1] [5.6] [9.2] [-9.4] [18.4] [-32.1]
Unemployment Rate 2.54** 2.777** 2.991* 3.019** 2.707** 1.441 2.089 -0.139 3.359* -1.453
of Non-College Men (1.004) (1.021) (1.596) (1.204) (1.046) (1.425) (1.841) (3.061) (2.026) (2.175)
[1.6] [1.7] [1.9] [1.9] [1.7] [0.9] [1.3] [-0.1] [2.1] [-0.9]
Observations 582 582 582 582 582 582 582 582 582 582
** indicates the coefficient is significant at the 5% significance level. * indicates significance at the 10% level. Standard errors in
parentheses. Numbers in brackets represent the “predicted” percent increase of the crime rate due to the mean change in the
independent variable, computed by multiplying the coefficient estimate by the mean change in the independent variable between 1979-
1989 (multiplied by 100). Dependent variable is log change in crime rate from 1979-1989 in county. Sample consists of 582 counties.
Regressions include county and time fixed effects and demographic controls (percent of population age 10-19, age 20-29, age 30-39,
age 40-49, age 50-64, and age 65 and over, percent male, percent black, and percent non-black and non-white). Regressions weighted
by mean of population size of each county. Wage residuals control for educational attainment, a quartic in potential experience,
Hispanic background, black, and marital status. The coefficients on all three presented independent variables are IV estimates using
augmented Bartik-Blanchard-Katz instruments for the change in total labor demand, and in labor demand for four gender-education
groups.
Table 8: Effects of Economic Conditions on Crime Rates. Estimates from the NLSY79.
Dependent Variable: Number of times
committed each crime: Shoplifted Stole Property
less than $50 Stole Property
more than $50 Robbery Share of Income
From Crime
Mean Long Household Income in State * 3.325** 1.883** 1.090 -.683 .080**
Respondent HS Graduate or Less (.864) (.791) (1.019) (.491) (.029)
[ .289] [ .164] [ .095] [- .059] [ .007]
Mean Log Weekly Wage of Non-College -5.352** -3.495** -3.108* 1.006 -.060**
Men in State (Residuals) * Respondent HS (1.414) (1.410) (1.613) (.905) (.030)
Graduate or Less [ .733] [ .479] [ .426] [- .138] [ .008]
Unemployment Rate of Non-College Men 5.624** 4.021** 4.415** -.415 .158**
in State * Respondent HS Graduate or (2.349) (1.616) (1.620) (1.332) (.038)
Less [ .035] [ .025] [ .027] [- .003] [ .001]
Mean Long Household Income in State * .743 1.349 -.070 .236 .024
Respondent Some College or More (1.301) (1.178) (.343) (.496) (.044)
[ .065] [ .117] [- .006] [ .021] [ .002]
Mean Log Weekly Wage of Non-College -.041 -2.703 1.048 -.330 .110
Men in State (Residuals) * Respondent (2.693) (2.135) (.713) (1.089) (.103)
Some College or More [ .000] [ .022] [- .008] [ .003] [- .001]
Unemployment Rate of Non-College Men -10.908 4.113 -.539 -3.456 -.185
in State * Respondent Some College or (8.073) (10.269) (1.386) (2.516) (.400)
More [ .021] [- .008] [ .001] [ .007] [ .000]
Observations 4585 4585 4585 4585 4585
R2.008 .008 .016 .002 .040
** indicates significance at the 5% level. * indicates significance at the 10% level. Standard errors correcting for within state
correlation in errors in parentheses. Numbers in brackets represent the “predicted” increase of the number of times the crime is
committed / share of income from crime due to the change in the independent variable, computed by multiplying the coefficient
estimate by the change in the independent variable between 1979-1989. Dependent variables are self-reports of the number of times
each crime was committed / fraction of income from crime in the past year. Additional controls include years of completed school (and
a dummy variable for some college or more), AFQT, mother’s education, log of a three-year average of family income (and a dummy
variable for family income missing), log of family size, a quadratic in age, and dummy variables for black and Hispanic background.
Wage residuals control for educational attainment, a quartic in potential experience, Hispanic background, black, and marital status.
... Previous works have mostly investigated the causes of crime from the perspective of total crime (Avio 1988;Craig 1987), which assigns equal weight to various types of crime under the total gure without measuring the different severity of each crime category such as homicide, aggravated assault, sexual assault, robbery, burglary, vandalism and auto theft (for example, the severity of homicide is far greater than vandalism or burglary); therefore, using a general measurement such as total crime is prone to weighing errors. For this reason, some researchers chose to analyse the determinants of several types of crime such as homicide, aggravated assault, sexual assault, robbery, burglary, vandalism, or theft (Entorf and Spengler 2000;Gould et al., 2002;Ihlanfeldt and Mayock, 2010;Kelly, 2000;Neumayer, 2005). ...
... While some studies found that drug abuse among adolescents has a profound effect on youth crime (Brook et al., 1996;Dawkins, 1997;Moss et al., 2014), it should be noted that drug crime is not limited to a specific age since the addiction is likely to be carried into adulthood. This study has classi ed crimes with reference to Entorf and Spengler (2000), Gould et al. (2002), and Ihlanfeldt and Mayock (2010). Besides total crime, violent crime and property crime, drug crime is also taken into account in this paper. ...
... Previous studies have revealed that areas with higher unemployment rates tend to be associated with more prevalence of crimes (Carmichael and Ward, 2001;Entorf and Spengler, 2000;Gould et al., 2002;Nordin and Almén, 2017;Raphael and Winter-Ebmer, 2001). Some studies suggested the inverse relationship in individual level, such as Borland and Hunter's (2000) tial employers. ...
Article
This study uses a panel data of Statistical Area Level 2 (SA2; generally known as suburbs) in Brisbane regions of Australia from 2002 to 2010, to examine the effects of various factors on four categories of crime (including total crime rate, property crime, violent crime and drug crime). The results show that the conventional socioeconomic and demographic factors, such as "unemployment rate", "income" and "population density", have the expected influence on criminal behaviours. Moreover, the demographic factor "male25-29" had positive effect on drug crime possibly because of the highest percentage of illicit drug use in Australians' late 20s. The demographic variable "male30-49", as an indicator of male population with a higher opportunity costs of committing crime, had the expected discouraging effect on violent crime but not on drug crime. Drug crime may be less age-specific probably due to addiction developing that continues through life. Furthermore, under Australia's Youth Allowance policy, "youth allowance" variable, as an indicator of proportion of youth receiving allowance from government, showed its inhibiting influence on different types of crime, especially on drug crime. This finding implies that adolescents, who may be compelled to resolve to drug-related crime for money, could be restrained when they are financially supported and periodically supervised by the social welfare program of the government.
... A recent extension of their work by Ha et al. (2020) however shows that the unemployment-crime relationship is not stable over time, and that more research is needed to explain this phenomenon. In order to establish a causal relationship between unemployment rates and crime, Gould et al. (2002) employ an instrumental variables model. Their main results show that higher unemployment rates lead to more crime. ...
... Higher unemployment rates or lower consumer confidence may increase both the likelihood and level of different types of strain individuals experience, and therefore increase criminal coping. As positive effects of times of economic decline on crime were also found in macromacro research (e.g., Gould et al., 2002;Rosenfeld and Fornango, 2007) and in the macro-micro study by Farrall et al. (2020), we formulate the following hypothesis: ...
... The unemployment rate is a more objective measure of the state of the economy, but it is always backward-looking when published. Although full macro-level analyses find that higher unemployment rates lead to more crime (e.g., Gould et al., 2002), we do not find this effect on micro-level criminal behavior. One of our sensitivity analyses shows that unemployment rates above the trend might actually lead to less crime, which would more fit the routine activity theory by Cohen and Felson (1979) and are in line with the pro-cyclical opportunity effects of Cook and Watson (2014). ...
Article
Full-text available
Since the early 1990s, increasing attention is being paid to the impact of life course transitions on criminal behavior. However, individuals’ life courses do not evolve in a vacuum but rather in the broad context of societal characteristics and developments. In this paper, we analyze whether there is an effect of macroeconomic circumstances on individuals’ criminal careers and whether this effect is conditional on marital status and parenthood. We employ micro-level data from a Dutch large-scale longitudinal study and enrich these with macro-level data. Using logistic panel data models for criminal behavior, we distinguish three types of effects: the macro-effect of experiencing declining economic times, the micro-effect of family composition and the interaction effect, indicating whether the effect of economic decline is differentiated by marital status and parenthood. We quantify economic decline either as an increase in the unemployment rate (objective and backward-looking measure) or as a decrease in consumer confidence (subjective and forward-looking). First, we find that an increase in the unemployment rate has no effect on criminal careers, but a decrease in the consumer confidence is associated with an increase in individual-level crime. Next, we confirm earlier results that marriage has a negative effect on crime, while parenthood has no effect. Finally, our results indicate that the detrimental effect of decreasing consumer confidence on criminal behavior is nullified for married individuals.
... There are many reasons to believe that education will affect crime; first, education increases the returns of reasonable work by raising the opportunity cost of illegal behaviour (Lochner & Moretti, 2004). There are plenty of studies, mainly in developed countries, which establish a negative correlation between earnings level and criminal activity (Freeman, 1982;Gould et al., 2002;Grogger, 1998). Second, education may directly affect the financial or psychic rewards from the crime itself, and finally, education may increase one's patience or risk aversion (Witte, 1980); hence all of these facts have led to a negative relationship between education and crime. ...
... Workers with high-paying jobs are likely to commit less crime than workers with low-paying jobs because the higher paid face a more significant opportunity cost of crime. Most of the casual labour in India is unskilled or low-skilled (Hajela, 2012), and studies have found that unskilled persons are more likely to commit a crime (Gould et al., 2002). ...
Article
Full-text available
Using National Crime Records Bureau and National Sample Survey data, this article explores the Spatio-temporal patterns of different types of crime in India. To explain these spatial patterns, we use spatial panel data regression techniques that account for spatial dependencies between different crimes and their socio-economic predictors at the NSS-region level. The results show that different types of crime have different time consistent spatial clustering of high and low crime regions. Evidence has been found that the prevalence of crime in a region is influenced by the crime rate of its neighbouring regions. This influence is linked to the spatial diffusion of criminal behaviour and harmful social norms that spread across contiguous regions over time. Our results also confirm the importance of various socio-economic variables like female literacy rate, sex ratio, inequality, regional prosperity, etc., in shaping the prevalence of crime in India. These findings could have significant implications for the policies aiming to reduce crime.
... The reason for this line of argument is that advanced countries, especially the United States of American and Canada, where these potentials can be effectively exploited provide adequate incentives for immigrants in the form of scholarship and visa lottery while those without skills and potentials are denied visa, scholarship. Exhibiting desperation to escape poverty, the latter usually enter their destination country illegally without documentation, significantly leading to deportation on arrival, as it is assumed that without skills to attract opportunities, they are likely to commit crime to survive in a new environment (Gould, Weinberg & Mustard 2002;Hanson, Lui & Mcintosh 2017;Connor & Ruiz 2019). ...
Article
Full-text available
The study focused on altering the emergence of a fallow state in Nigeria's fourth republic through channelling its human resources for sustainable development. The objective of the research was to develop an explanatory model that explores the available alternative paths that countries facing human resources management challenges can adopt to alter the possibility of emerging as a fallow state or necessary entering into it. Utilising Marxism to extrapolate the inherent human resources conditions and predictive revolutionary outburst (which is failing to occur in Nigeria) and Ekeh's theory of 2 republics to explain how complex dialectical social interactions provide sufficient conditions to allow Nigeria's perennial decline and decay towards the possibility of a fallow state. The researcher adopted quantitative method, utilising human capital index, human development index and migration data to make correlation analysis to measure the quality of institution that manages human resources in Nigeria and examined the pattern of flight against the postulation of a fallow state. The researcher recommended that there is need to urgently develop a 50 year national human resources investment and development strategy to aggressively channel Nigeria towards sustainable development.
... There is substantial evidence from the unemployment and crime literature which documents significant and modest impacts of unemployment on total and property crimes in the US (Gould et al 2002;Raphael & Winter-Ebmer 2001;Sweden, Öster & Agell, 2007); France (Fougère et al 2009) and for right-wing extremist crime in Germany (Falk et al 2011). These studies instrument unemployment rates to identify a causal impact of unemployment on crime. ...
Book
Full-text available
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, computer, mechanical, photocopying, recording and / or otherwise, without the prior written permission of the authors and the publisher.
... This comparison of descriptive statistics between the two populations might suggest that the Syrian population would have a higher potential to commit a crime as it has higher shares of young and single males with low education levels who are trapped in informal employment with long hours of work and little pay, as the literature on the individual determinants of crime suggests (Freeman, 1991;Levitt, 1998;Grogger, 1998;Raphael & Winter-Ember, 2001;Gould et al., 2002). ...
Article
Full-text available
The impact of immigration on crime continues to stir heated debates in public policy circles around the world. Surveys indicate that host societies favor controversial measures because they are concerned about what they perceive to be an impingement exacted on their security with each new wave of migration. Seeking whether there is any truth to such perceptions, this paper analyzes the refugees’ impact on crime rates, using the case of Turkey which has started to host the world’s largest refugee population within any national borders due to the Syrian civil war. In doing so, the paper employs instrumental variables, difference-in-differences (DiD), and staggered DiD methods to explain if the conflict-fleeing Syrians have pushed Turkey’s crime rates higher in the short and the long run. It also controls for a multitude of time-varying provincial characteristics and presents a battery of robustness checks against various identification threats. As a result, DiD estimates show that refugees do not have any causal effect on the crime rates in Turkey. More strikingly, its IV estimates provide evidence for a rather negative effect on the crime rates per capita whilst finding a null effect on the crime rates per native resident in particular.
... More and better education increases individuals' income from legal sources and potentially avoid the need for illegal returns from crime. In addition, education decreases the likelihood of unemployment which, in turn, is linked to higher crime rates (Gould et al., 2002;Raphael & Winter-Ebmer 2001). As education usually leads to increased future returns, individuals have an opportunity cost that deters them from engaging in crime (Lochner, 2020). ...
Article
Modern slavery is a persistent human tragedy and a growing organisational risk. The United Nations' sustainable development goals highlight the significance of governments in shaping firms' sustainability agenda and combating modern slavery. However, little is known about the effects of the institutional environment on modern slavery risk. This study, therefore, investigates the crucial policy question of whether the quality of the institutional environment has any effect on modern slavery and whether sustainable human development reinforces this relationship. Using data from 167 countries, we find that institutional environment quality is negatively associated with the prevalence of and vulnerability to modern slavery and positively associated with its modern slavery risk mitigation. Our results suggest that democratically elected governments operating in politically stable societies with higher quality of voice and accountability, higher levels of control of corruption, and stricter rule of law are more accountable and responsive to modern slavery risks. We also find that sustainable human development (HDI) has a moderating effect on the relationship between institutional environment quality and modern slavery, and this effect is mainly noticeable in low HDI countries. These results imply that governance reforms alone might not yield the desired effects for all countries and, hence, have significant implications for policymakers, companies, and societal stakeholders.
... Studies such as the one conducted by Gould, Mustard, and Weinberg (2002) have shown that education increases pay, showing that investments in human capital and schooling should reduce crime. Lochner (2004) highlighted the role of education as an investment in human capital that increases future legal employment opportunities. ...
Conference Paper
Previous research showed that public policies regarding expenditures on public order and safety, punishments, employment opportunities, schooling are determinants for the crime rate, mainly property crime. Since there is little study available in the literature on the determinants of money laundering, we wanted to see if unemployment, the level of adult education, emotional well-being, morality, or the health of a tax system has a particular influence on the financial crime rate. For this purpose, we built an econometric model, through which we aimed to capture the influencing socioeconomic factors of money laundering crime rates in 21 countries during 2013 and 2018. The results have shown that unemployment and fiscal health influence 20% of the money laundering rate variation, at a 95% confidence level. Increasing emotional well-being lowers the crime rate of money laundering at a 10% significance level, while education and morality are not statistically significant.
Research
Full-text available
As oportunidades no mercado de trabalho, bem como a formação de capital humano (educação), afetam os custos implícitos de se dedicar às atividades ilegais. No presente texto, investigamos empiricamente o efeito que a taxa de desemprego, os rendimentos reais no mercado de trabalho e o acesso educacional para os jovens (15 a 24 anos) exercem sobre a prevalência de homicídios nos municípios brasileiros. Analisamos ainda as heterogeneidades que dizem respeito à especialização do trabalho e ao ciclo de vida na juventude. A fim de contornar os problemas de endogeneidade, em nossa estratégia de identificação utilizamos um modelo de dados em painel para todos os municípios brasileiros (no período entre 1980 e 2010), com o uso de variáveis instrumentais, seguindo de perto a abordagem desenvolvida por Gould, Weinberg e Mustard (2002). Os instrumentos relacionam-se a mudanças estruturais de longo prazo na indústria que afetam a demanda por trabalho nas cidades. Encontramos os seguintes resultados: i) efeito positivo da taxa de desemprego dos homens sobre a taxa de homicídios, sendo esse efeito maior para jovens adultos (25 a 29 anos); ii) efeito estatisticamente nulo da renda do trabalho por hora trabalhada sobre a taxa de homicídios; e iii) efeito negativo da taxa de atendimento escolar de 15 a 17 anos sobre a taxa de homicídios.
Book
Full-text available
The 2022 Swedish labour market is characterised partly by high unemployment with a high proportion of long-term unemployed and partly by a large labour supply with a very high activity rate by international comparison. At the same time, requirements of the labour force are increasing continuously, and future prospects look different for different groups. To develop an effective labour market policy, knowledge of how entry into the labour market and the establishment process appear for different groups of individuals is needed. This report specifically concerns paths into the labour market. It follows all individuals born in 1979 on their journey into the labour market from 1995 to 2017, i.e. from ages 16 to 38. We analyse successes and setbacks for this cohort in achieving the objective of some degree of self-sufficiency. Full employment is a central objective of economic and social policy and there is good reason to continue pursuing this objective, but we think it should be combined with the objective that the individual also has an earned income they can live on.
Article
Full-text available
This paper uses panel data techniques to examine the relationship between unemployment and a range of categories of crime in New Zealand. The data set covers sixteen regions over the period 1984 to 1996. Random and fixed effects models are estimated to investigate the possibility of a causal relationship between unemployment and crime. Hypothesis tests show that two-way fixed effects models should be used. The regression results provide some evidence for significant effects of unemployment on crime, both for total crime and for some subcategories of crime.
Chapter
In the US, fluctuations in relative wages and employment across regional markets are very large. For example, during the so-called ‘Massachusetts miracle’ of the 1980s, average wages of men in New England increased by 17 percent, while wages in the ‘rustbelt’ states of the Midwest fell by about 8 percent. At the same time, annual weeks worked increased by about 4 weeks per person in New England, but fell by about 2 weeks in the Midwest. Changes such as these are apparently demand driven in the sense that wage and employment fluctuations are positively correlated over time (Juhn, Murphy and Topel, 1991).
Article
Bartik reviews evidence on whether state and local policies affect job growth. He then presents empirical data supporting the intentions of such programs, showing that job growth may lead to a number of positive long-term effects including: lower unemployment, higher labor force participation, higher real estate values, and better occupational opportunities. He also shows that the earnings gains to disadvantaged groups outweigh the resulting increased real estate values for property owners, and concludes by saying that regional competition for jobs may actually be a benefit for the nation as a whole.
Article
Much work on crime has focussed on the effect of criminal sanctions on crime, ignoring (except as a control variable) the effect of labor market conditions on crime. This study reviews studies of time series, cross area, and individual evidence pertaining to the effect of unemployment and other labor market variables on crime and compares the "strength" of the labor market-crime and the sanctions-crime relations.It corcludes that there is a labor market-crime link but that this link is not well estimated by existing studies and is weaker than the sanctions-crime link. The rise in crime in recent years does not appear to be greatly due to the performane of the labor market.
Article
Recently many papers have used the arrest rate to measure punishments in crime-rate regressions. However, arrest rates account for only a portion of the criminal sanction. Conviction rates and time served are theoretically important, but rarely used, and excluding them generates omitted variable bias if they are correlated with the arrest rate. This paper uses the most complete set of conviction and sentencing data to show that arrest rates are negatively correlated with these normally excluded variables. Consequently, previous estimates of arrest rate impacts are understated by as much as 50%. Also, conviction rates, but not sentence lengths, have significant explanatory power in standard crime-rate regressions.
Book
Bartik reviews evidence on whether state and local policies affect job growth. He then presents empirical data supporting the intentions of such programs, showing that job growth may lead to a number of positive long-term effects including: lower unemployment, higher labor force participation, higher real estate values, and better occupational opportunities. He also shows that the earnings gains to disadvantaged groups outweigh the resulting increased real estate values for property owners, and concludes by saying that regional competition for jobs may actually be a benefit for the nation as a whole.