The link between unemployment and crime rate fluctuations: An analysis at the county, state, and national levels

Abstract

Cantor and Land (1985) developed a theoretical model that proposed two pathways through which economic activity – as indexed by the aggregate unemployment rate – could affect the rate of criminal activity. The first is by increasing levels of criminal motivation within the population as deteriorating economic conditions affect social strain and social control; the second is by influencing the availability and vulnerability of criminal targets and thus the number of criminal opportunities. Although much empirical research has applied this theoretical model, few analyses have done so at disaggregated units of analysis. We present the most comprehensive analysis to date by empirically evaluating this model with data on 400 of the largest US counties – and examine the effects of aggregation on results as these county data are combined to the state and national levels – for the years 1978–2005. For seven Index crimes at each of the three levels of analysis, and with or without controls for structural covariates at each level, the directional effects hypothesized by Cantor and Land are found for 78 out of 84 estimated relationships. Even after taking into account the lack of statistical independence of these estimates by drawing on recently developed statistical theory, this is a very unlikely outcome. In accordance with expectations based on theory and prior research, (a) some of these relationships are weak and not statistically significant, and (b) the strongest and most consistent patterns of relationships for both the crime opportunity and crime motivation effects are found for three property crimes: burglary, larceny, and motor vehicle theft. Suggestions for further research on this topic are given.

Introduction

Some 25 years ago, Cantor and Land (1985, p. 317) raised the question: Does aggregate unemployment have a positive, negative, or null effect on levels of crime in capitalist societies? They noted the central importance of answers to this question both to theories of crime and to the formation of social policy. They also observed a lack of consensus of findings from various studies to that date, with some finding a positive unemployment and crime (hereafter U–C) relationship, but others finding a negative or null relationship.

In addressing this question, Cantor and Land (hereafter C&L) developed a theoretical model that proposed two paths through which economic activity – as indexed by the aggregate unemployment rate – could affect the rate of criminal activity. The first is by altering criminal motivation through the impact of changing economic conditions on social strain and social control. The second is by influencing the availability and vulnerability of criminal targets and thus the number of criminal opportunities. These effects were hypothesized to be countervailing – a downturn in aggregate economic activity, for example, would increase motivation but decrease opportunity, and the operation of these two opposing effects could account for the relatively weak overall U–C relationships found in most studies. Cantor and Land also posited that the two impacts would likely occur with different timings – changes in opportunity would be coincident with changes in aggregate economic activity, whereas changes in motivation would be delayed and occur after a period of sustained unemployment.

The Cantor and Land (1985) article has been termed a “seminal work” by Arvanities and DeFina (2006, p. 139), because, by showing the complexity of a seemingly simple relationship, it built a foundation for many subsequent empirical studies. The C&L (1985) article used data on crime rates for the United States at the national level, the level of analyses adopted by many subsequent studies. However, with the increased availability of annual data on crime and structural covariates for multiple units of observation such as cities, counties, and states in the US in recent years, researchers have started to apply panel regression models to assess the C&L model. For example, Arvanities and DeFina (2006), noting specifically the benefits to panel analyses at finer units of aggregation (2006, pp. 152–153), used pooled annual data on the fifty US states from 1986 to 2000, to test various features of the C&L model.

In the present paper, we continue this line of investigation by examining the U–C relationship at a still-more disaggregated unit of analysis – the county – between 1978 and 2005. Such an approach has several advantages. First, counties are more homogenous units than states, with less variation in structural covariates within these units over time, and thus biases due to aggregation are mitigated. Second, due to several factors cited below, such as the countervailing effects of the crime opportunity and crime motivation mechanisms through which macroeconomic downturns affect crime rates, the effects of the C&L explanatory variables are not expected to be strong and highly statistically significant. Increasing the number of observations with county-level data enhances statistical power and the ability to detect whether or not the inclusion of these variables produces models that are preferred as compared to models that do not include them. Finally, to our knowledge, no prior work has focused on the original C&L model using county data. Thus, our first objective is to test the original C&L model specification using panel data at the county level.

This approach has another advantage, in that we can aggregate the county-level data into corresponding annual observations on the states in which they are located, as well as up to the national level. Therefore, our second objective is to consider how the process of aggregation affects conclusions regarding the C&L model, and so we also conduct panel regression analyses at the state level and time series analysis at the national level. Our purpose here is not to conduct full analyses at the state and national level (as noted above, past research has done this), but rather to determine the effects of aggregation on our understanding of the C&L model. We use these multiple analyses, all adopting the same model specification, to provide a better and more consistent interpretation of different results found in past research that are based on varying aggregation levels. This juxtaposition of estimates of the C&L model at the three levels of analysis facilitates the most extensive empirical assessment to date.