V. Empirical Challenges to Studying the Employment–Crime Relationship
Any study of the causal effect of employment on crime must confront at least two empirical challenges: (1) endogeneity and (2) simultaneity. These are threats to causal inference that can seriously bias empirical estimates of the employment– crime association. Each is discussed in turn, and recent efforts to overcome these challenges are described.
A. Endogeneity: The Selection Problem
One of the most serious challenges to existing studies of employment and crime is the selection problem. It is the problem of endogeneity of employment effects on crime, meaning that individuals who are employed (or are employed in high-quality jobs) differ fundamentally from individuals who are not employed in a way that accounts for their lower crime involvement. One may conceive of such person-level characteristics as ability, planfulness, and agreeableness that might individually or jointly increase the likelihood that an individual will be gainfully employed and simultaneously reduce the likelihood that the person will commit crime. The selection problem arises when these traits are difficult or impractical to observe and measure. The consequence is systematic bias in the estimated effect of employment on crime. Moreover, the direction of the bias under this scenario is predictable: The impact of employment on crime will be overestimated.
Sampson and Laub (1993) found that weak occupational commitment and job instability from ages 17 to 32 were predicted by official delinquency, unofficial delinquency (self-, parent, and teacher report), and early temper tantrums (parent report) during childhood. Caspi, Wright, Moffitt, and Silva (1998) linked youth unemployment (ages 15–21) with a variety of factors that reach far back into childhood. As measured in early childhood (ages 3–5), longer duration of unemployment was predicted by low family occupational status, low intelligence, an unmarried mother at birth, and difficult temperament. As measured in late childhood (ages 7–9), youth unemployment was predicted by these same variables in addition to family conflict and behavior problems. An important contribution of these studies is that they directly address the selection problem and identify underlying factors responsible for the differential sorting of individuals into the labor market, oftentimes long before they do so.
Under these circumstances, causal inference about the nature of the employment–crime relationship is aided by the availability of longitudinal data, which allow researchers to overcome endogeneity of the employment effect on crime attributable to time-stable individual differences, so-called unobserved heterogeneity (e.g., Horney et al., 1995). Such studies have examined the way in which change in employment affects change in crime and have found that the employment–crime relationship (at least among adults) does withstand these more rigorous selection controls and is not seriously biased by endogeneity. However, it is worth noting that the strength of the correlation tends to be weak compared with other time-varying factors, such as drug consumption and living arrangements (e.g., marital living and cohabitation).
The consequences of the selection problem have been brought into sharp focus in recent youth employment research. Paternoster et al. (2003) and Apel et al. (2006) have addressed the selection problem using longitudinal data on employment and antisocial behavior for 3 years. Both studies replicated the positive correlation between intensive employment during the school year and delinquent behavior using conventional methods. However, both also found that intensive work was positively correlated with delinquency only when examined across individuals but that within-individual change in work involvement was not correlated at all with change in delinquent behavior and substance use. They concluded that the criminogenic effect of intensive work among adolescents was driven by a process of selection rather than causation and could be best understood as a spurious correlation.
In one of the most recent statements on the subject of adolescent employment, Apel et al. (2008) exploited interstate variation in child labor laws at the 15-to-16 transition as a source of causal identification. They found in their analysis that work intensity was actually inversely correlated with delinquent behavior; that is, the increase in work involvement from age 15 to 16 attributable to a loosening of child labor restrictions (the magnitude of which varied across states) was actually associated with a substantial decline in delinquent involvement. Once the problem of endogeneity was addressed through the use of longitudinal data and instrumental variables, then, the employment–delinquency association was found to be inverse after all, contrary to most previous youth employment research but well in line with employment–crime research among adults. Moreover, the techniques that Apel et al. used allowed them to interpret this as a causal association.
B. Simultaneity: The Feedback Problem
The contemporaneous, inverse correlation between employment and crime is usually interpreted as the causal effect of employment on crime. However, the correlation may in fact represent the causal effect of crime on employment, which is the feedback problem. This is the problem of simultaneity of causal effects, in that employment and crime mutually influence one another. The practical consequence of simultaneity bias is to systematically overestimate the effect of employment on crime, because the simultaneous inverse effect of crime on employment will be erroneously attributed to the effect of employment on crime.
Labeling theory, for one, anticipates just this sort of feedback effect from crime to employment. This is the notion of secondary deviance, or deviance amplification, among persons toward whom a sanction has been directed. An arrest or conviction, for example, constitutes a social stigma that might lead to exclusion from legitimate employment (Pager, 2003). Many prospective employers may be disinclined to hire individuals with a criminal record because it serves as a signal of sorts about what kind of employee one is likely to be. For example, employers may be sensitive to liability for negligent hiring (Bushway, 2004), or they may perceive offenders as untrustworthy (Waldfogel, 1994). A criminal record may also relegate individuals to the secondary labor market, or to what Nagin and Waldfogel (1995) referred to as “spot market jobs” as opposed to “career jobs.” This effect may be attributable, in part, to state-imposed restrictions on employment in certain industries (e.g., government employment), catering to vulnerable clientele (e.g., children), and professional licensing in certain occupations (Burton, Cullen, & Travis, 1987).
Empirical research confirms that a criminal record in the form of arrest, conviction, or incarceration does indeed hamper an individual’s future employment prospects (e.g., Nagin & Waldfogel, 1995; Waldfogel, 1994; Western, 2002). A criminal record reduces employment, increases unemployment, lowers earnings, slows wage growth, diminishes job tenure, and exacerbates job turnover. Thus, the feedback problem is real, and research that examines the contemporaneous effect of employment on crime must be attentive to simultaneity bias that overstates the preventive effect of employment on crime.
One way that researchers have addressed the feedback problem is through estimation of reciprocal models of employment and crime. Simultaneous equation studies have confirmed that the cross-sectional association between employment and crime is a combination of the effect of employment on crime as well as the effect of crime on employment (e.g., Good et al., 1986; Thornberry & Christenson, 1984). In these studies, isolation of causal effects requires the use of exclusion restrictions (i.e., instrumental variables) or other modeling constraints that are capable of identifying the simultaneous effects in the model. Thornberry and Christenson (1984) imposed cross-time equality constraints on model parameters to identify the reciprocal effects of unemployment and arrest. Good et al. (1986) used the number of job rejections as an instrumental variable for employment and gang affiliation and police enforcement (specifically, police contact) as instrumental variables for arrest. Each of these studies found that the effect of (un)employment on arrest was stronger than the contemporaneous feedback effect of arrest on (un)employment. In fact, both studies discovered that the contemporaneous effect of arrest on (un)employment was not statistically significant, although Thornberry and Christenson discovered that the influence of arrest was lagged one period, and Good and colleagues noted that the total number of prior police contacts was more salient. These studies thus suggest that the influence of criminality on employment operates through the accumulation of an arrest record that impedes the acquisition of stable employment.