Crime Causation: Economic Theories

The motivation behind most early applications of Becker's model was to examine the impact of legitimate labor market experiences (e.g., unemployment) and sanctions on criminal behavior. Broadly speaking, the empirical findings are that (1) poor legitimate labor market opportunities of potential criminals, such as low wages and high rates of unemployment, increases the supply of criminal activities; and (2) sanctions deter crime.

The empirical evidence on the relationship between unemployment and criminal activity has been the subject of much investigation (see literature review by Freeman, 1999a). Unemployment could be taken to influence the opportunity cost of illegal activity. High rates of unemployment growth could be taken to imply a restriction on the availability of legal activities, and thus serve to ultimately reduce the opportunity cost of engaging in illegal activities. Although theoretically well-defined, most empirical studies of the unemployment-crime relationship have provided mixed evidence.

Not all early studies used aggregate time-series data to test the relationship between unemployment and crime. Thornberry and Christenson use individual level data from the 1945 Philadelphia cohort to find that unemployment had significant effects on crime. Farrington et al., using data from the CSDD, showed that property crime rates were higher when offenders were unemployed.

Witte and Tauchen (1994) exploit the panel data dimensions of the Philadelphia cohort used by Thornberry and Christenson. Instead of primarily focusing on crime as a function of unemployment, they use a richer set of controls, like deterrence, employment status, age, education, race, and neighbourhood characteristics. The results reported by Tauchen and Witte on the relationship between employment and crime were consistent with the previous findings of Thornberry and Christenson and Farrington. Recent work, of which Levitt and Witt et al. (1999) are representative, proceeded to use pooled time-series cross-section data and find, inter alia, positive associations between unemployment and property crime.

One problem with most work and crime models is that they assume both activities are mutually exclusive. This may be a problematic assumption when considering disadvantaged youths (see Freeman, 1999b). The fact that a youth can shift from crime to an unskilled job and back again or can commit crime while holding a legal job means that the supply of youths to crime will be quite elastic with respect to relative rewards from crime vis-à-vis legal work or to the number of criminal opportunities.

From the 1970s through the 1990s the labor market prospects for unskilled workers in most OECD countries has deteriorated considerably. In particular, the real earnings of young unskilled men fell, while income inequality rose. This suggests that as the earnings gap widens, relative deprivation increases, which in turn leads to increases in crime. Empirical research into the relationship between earnings inequality and crime generally find that more inequality is associated with more crime. For example, in a study based on a sample of the forty-two police force areas in England and Wales, Witt et al. (1999) report a positive association between earnings inequality and crime rates for vehicle crime, theft, and burglary. For the United States, see the evidence reviewed in Freeman (1999a).

Much of the empirical work on testing the Becker model has focused on the role of deterrence in determining criminal activity. Deterrence refers to the effect of possible punishment on individuals contemplating criminal acts. Deterrence may flow from both criminal justice system actions and from social actions (i.e., the negative response of friends and associates to criminal behavior). To date, attempts to measure deterrent effects have concentrated on the effects of the criminal justice system. See Nagin (1998) for a survey of this literature.

This section discusses a variety of practical problems that arise in testing for deterrent effects. In particular, we consider three estimation issues: measurement error, endogeneity, and nonstationarity.

Models of criminal behavior are usually estimated using official reported crime statistics. Such recorded offenses are influenced both by victims' willingness to report crime and by police recording practices and procedures. At the level of the individual police department, both administrative and political changes can lead to abnormalities in reported data or to failures to report any data. For example, the measurement error in crime rates may arise because hiring more police leads to more crimes reported. Consequently, estimates derived from regressing crime rates on the number of police (or on arrest rates) may be severely distorted by the impact of measurement error.

The potentially serious problem of simultaneity between sanctions and crime has been the subject of much debate. Here, the main point is that increases in sanctions may cause decreases in crime, but increases in sanctions may be in response to higher crime rates. Since the 1970s there has been a considerable effort to find instruments (i.e., exogenous factors) to identify the effects of sanctions on the supply of crime. For example, Levitt (1996) uses instrumental variables to estimate the effect of prison population on crime rates. Prison-overcrowding litigation in a state is used as an instrument for changes in the prison population.

In order to identify the effect of police on crime, Marvell and Moody and Levitt (1997) proposed different procedures. Marvell and Moody are concerned with the timing sequence between hiring police and crime. Using lags between police levels and crime rates to avoid simultaneity, they test for causality in the spirit of Granger. Although they find Granger causation in both directions, the impact of police on crime is much stronger than the impact of crime on police. In a recent paper Levitt (1997) uses the timing of elections (when cities hire more police) as an instrumental variable to identify a causal effect of police on crime. He finds that increases in police instrumented by elections reduces violent crime, but have a smaller impact on property crime.

A substantial problem that has been ignored in the vast majority of empirical studies is nonstationarity of crime rates. A time-series is said to be nonstationary if (1) the mean and/or variance does not remain constant over time; and (2) covariance between observations depends on the time at which they occur. In the United States, index crime rate appears strongly nonstationary, for the most part being integrated of order one with both deterministic and stochastic trends (a random variable whose mean value and variance are time-dependent is said to follow a stochastic trend). See, for example, Witt and Witte (2000). Here, the authors have attempted to estimate and test a model using linear nonstationary regressor techniques like cointegration and error correction models. The empirical results suggest a long-run equilibrium relationship between crime, prison population, female labor supply, and durable consumption.