The effects of labor markets and income inequality on crime: evidence from panel data.

• Author: Doyle, Joanne M.

1. Introduction

   In recent years, crime rates have remained a focal point of debate in political as well as academic circles. The release of new crime statistics is typically followed by a barrage of partisan political approvals and disclaimers depending on which party or interest group benefits. Incumbents are always quick to accept the credit for any decrease in crime rates while opponents are just as quick to challenge the reliability of the statistics or argue rates would have somehow fallen faster had they been in office (Thomas 1996). Among academicians, there remains little consensus with respect to the factors that underlie criminal activity, how to appropriately model criminal activity, and what public policies might serve to lessen criminal activity. In a recent symposia on the economics of crime in Journal of Economic Perspectives, John J. DiIulio (1996) argues that economists have not focused adequate attention on modeling crime using sophisticated quantitative and modeling skills that are part of the economist's toolkit. He laments the fact that much of the research in this area has remained the domain of sociologists and criminologists who tend to use less sophisticated empirical analyses. While we do not want to debate the relative merits of the noneconomist's approach to the empirical study of the economic model of crime, we believe this is a field that needs to be open to all attempts at serious inquiry.

   The economics literature has relied on Becker's (1968) economic model of crime for guidance in developing and testing empirical models of crime. This model assumes that people choose optimal allocations of time devoted to criminal activity to maximize their expected utility by weighing the expected benefits from crime against the expected costs of crime. The benefits consist of income derived from crime, as in the case of property crime, as well as ill-defined psychic benefits (tastes and preferences). These benefits will be negatively affected by public policies designed to deter crime since such efforts increase the likelihood of being caught and punished. The costs of crime consist of the legal income that a criminal forgoes when engaging in crime because crime is time consuming and may result in incarceration, further limiting legal income opportunities. These costs may also include the disutility of being incarcerated, which in turn is influenced by society's deterrence efforts.

   A test of Becker's model involves empirically testing for evidence that actual and potential criminals respond to these costs and benefits of crime. Much of the existing research has focused on verifying the empirical relationship between deterrence efforts and crime. While we consider this to be an important objective, in this paper, we direct our attention to analyzing the effects of labor market conditions, income distribution, and demographics on crime. We choose this focus in response to Freeman (1996), who asks whether the rising rate of criminal involvement of young males is the result of the collapse in the job market for the less skilled and whether income inequality has a significant effect on crime. If this is so, he urges society to place more effort on improving legitimate job opportunities for potential and actual criminal offenders.

   We estimate a model of property crime using a panel data set of the U.S. over the years 1984-1993. We focus on property crime under the assumption that property crime is more likely motivated by financial gain and thus the benefits and costs of property crime are more likely to be captured with economic variables than are the benefits and costs of violent crime. For the purpose of comparison, we also present results from estimating the model for violent crime. The use of aggregate data, whether at the country, state, city, or police jurisdiction level, has been criticized since the model at hand is one of individual behavior. In spite of this criticism, such data have nevertheless been used in circumstances where individual data are not available.(1) Given our focus on labor markets, income distribution, and demographics, we believe that aggregate data are well suited for our study since we can use a rich set of variables such as average market wages, sector-specific wages, unemployment rates, and income inequality measures. Furthermore, variables such as income inequality are not readily available at aggregation levels lower than the state level. Using panel data instead of a simple time series or cross-section allows us to control for unobserved heterogeneity across states, which greatly reduces the likelihood of an omitted variable bias.

   We find econometric evidence that partly supports Freeman's arguments. In particular, we find, not surprisingly, that the proportion of young males in the population has a significant positive effect on property crime yet a significant negative effect on violent crime. More importantly, we find strong evidence that favorable labor market conditions have significant negative effects on both property crime and violent crime. We measure labor market conditions using an expected wage that takes into account wages, unemployment compensation, and the unemployment rate. We further analyze labor market conditions by replacing the expected wage with a vector of sector-specific average wages. We find that property crime is most responsive to wages in low-skilled sectors. Surprisingly, we find that, contrary to popular belief, income inequality has no independent effect on crime rates.

   The paper is divided into five sections. In section 2, we review some relevant research on econometric models of crime. Particular attention is paid to the effect of economic variables in models of crime from studies that use aggregate data as well as individual level data. In section 3, we present an econometric model of crime and discuss the variables included in the model as well as the econometric techniques employed. Section 4 discusses the data and the results. We conclude with section 5.

2. Review of Recent Economic Models of Crime

   Previous research on the effect of economic conditions on crime rates has not come to a consensus on the role labor markets play in influencing criminal behavior. Within the context of the Becker model, legal wages represent the opportunity cost to crime. Higher wages should reduce criminal activity. However, empirical support of this hypothesis is ambiguous, partly because the variables used to capture market wages differ drastically across studies. In several earlier studies conducted in the 1970s and 1980s with aggregate data, researchers used income measures such as median family income or mean per capita income. Although these variables may capture labor market wages (opportunity cost of crime), they may simultaneously capture the benefits to crime, especially property crime where high wealth in a given region signals good opportunities for crime. So good legal opportunities should lower crime, but high returns to crime should increase crime. It is not surprising then that coefficient estimates of the elasticity of crime with respect to income vary greatly across studies. For example, Trumbull (1989) uses mean income per capita as a measure of the opportunity cost to crime. Using a cross-section of data on counties in North Carolina, he finds that mean income has no effect on crime, most likely due to the negative effect of high income on crime being offset by the positive effect high income can have on crime by providing better income-producing criminal opportunities. In a survey of the literature, Eide (1994) reports that there is no systematic relationship between the income measures applied and the estimates obtained in a variety of crime studies, most likely for the reasons mentioned above. It simply is not clear what the various income variables are measuring.

   Freeman (1996) argues for the inclusion of income inequality measures in empirical models of crime. In fact, conventional wisdom maintains a positive relationship between income inequality and crime. However, a causal link is not well documented. Several authors have used income inequality measures in their statistical models, but they differ according to whether inequality captures costs or benefits to crime. Ehrlich (1996) uses an income distribution variable to capture the opportunity costs of crime and finds it to be statistically significant. The argument is that higher inequality results in more people at the bottom end of the income distribution, and these individuals will be more prone to crime because the cost in terms of legal income forgone is quite low. However, this argument does not establish a causal link between income inequality and crime per se but instead uses inequality as a proxy for the opportunity cost of crime, which we believe does have a causal link. Still other authors have used income differentials to measure the benefits to crime. For example, Mathur (1978) uses the Gini coefficient for this purpose and finds its effect on crime to be ambiguous.

   Unemployment rates have been used in crime studies as an additional aspect of the labor market that may influence criminal activity. To the extent that wages are not flexible enough to clear labor markets, unemployment rates may provide important information regarding job availability. High unemployment rates indicate a lack of legal opportunities and should lead to increases in crime. However, Trumbull (1989) finds that, on the contrary, high unemployment rates have a negative effect on crime. This result can be explained by recognizing that unemployment rates, like income measures, may capture more than a lack of legal job opportunities. In the sociology literature, Cantor and Land (1985), Chiricos (1987), and Smith, Devine, and Sheley (1992) empirically test the relationship between crime rates and unemployment rates, arguing that, in addition to the unemployment rate...