Abstract
Objectives
We seek evidence for economic and social mechanisms that aim to explain the relationship between employment and crime. We use the distinctive features of social welfare for identification.
Methods
We consider a sample of disadvantaged males from The Netherlands who are observed between ages 18 and 32 on a monthly time scale. We simultaneously model the offending, employment and social welfare variables using a dynamic discrete choice model, where we allow for state dependence, reciprocal effects and timevarying unobserved heterogeneity.
Results
We find significant negative bidirectional structural effects between employment and property crime. Robustness checks show that only regular employment is able to significantly reduce the offending probability. Further, a significant unidirectional effect is found for the public assistance category of social welfare on property offending.
Conclusion
The results highlight the importance of economic incentives for explaining the relationship between employment and crime for disadvantaged individuals. For these individuals the crime reducing effects from the public assistance category of social welfare are statistically equivalent to those from employment, which suggests the importance of financial gains. Further, the results suggest that stigmatizing effects from offending severely reduce future employment probabilities.
Similar content being viewed by others
Notes
The omitted variables problem, or selection problem, arises when variables are omitted from the offendingemployment/welfare regression that are not randomly correlated with the outcome variable. The corresponding OLS regression parameter estimates become biased whenever this occurs, see Davidson and MacKinnon (2004, Chapter 8). The simultaneity problem, or reverse causality problem as discussed by Ehrlich (1973), arises when offending, employment and welfare outcomes have mutual causal effects on each other. The corresponding OLS regression parameter estimates, resulting from a oneway regression of employment and welfare outcomes on offending outcomes, become biased whenever this occurs.
The other category that is included in the 1.4 million is maternity leave (42,800), which is not included in the current study.
Modeling the actual probabilities \(\pi _{i,t}\) is difficult as these are restricted to lie between zero and one. The logistic transformation ensures that the transformed probabilities \(\theta _{i,t}\) are unrestricted. Note that the \(3\times 1\) vector \(\theta _{i,t}\) has elements \(\log \left[ \pi _{C,i,t} / (1+ \pi _{C,i,t})\right] , \log \left[ \pi _{E,i,t} / (1+ \pi _{E,i,t})\right]\) and \(\log \left[ \pi _{W,i,t} / (1+ \pi _{W,i,t})\right]\).
The functional form of this specification can be relaxed to allow for different agedependent correlations
$$\delta = \delta _{0} + \sum _{t=1}^T X_{i,t} \delta _{t} + \lambda Y_{i,0},$$which is the specification adopted in Chamberlain (1980). Here the coefficients \(\delta _{t}\) capture the correlation between the random part of the factor structure and the explanatory variables at each monthly period. We experimented with such extensions and the results did not qualitatively change.
In our empirical study we experimented with different ways of including the detention variable. No qualitative changes in the structural effects were found for different constructions of the detention variable.
References
Agnew R (1992) Foundation for a general strain theory of crime and delinquency. Criminology 30:47–87
Ahn SG, Lee YH, Schmidt P (2013) Panel data models with multiple timevarying individual effects. J Econom 174:1–14
Alessie R, Hochguertel S, van Soest A (2004) Ownership of stocks and mutual funds: a panel data analysis. Rev Econ Stat 86:783–796
Apel R, Sweeten G (2010) The impact of incarceration on employment during the transition to adulthood. Soc Probl 57:448–479
Bai J, Ng S (2009) Panel data models with interactive fixed effects. Econometrica 77:1229–1279
Baltagi BH (2005) Econometric analysis of panel data. Wiley, New York
Becker GS (1968) Crime and punishment: an economic analysis. J Polit Econ 78:169–217
Becker GS, Mulligan CB (1997) The endogenous determination of time preferences. Q J Econ 112:729–758
Bersani BE, Laub JH, Nieuwbeerta P (2009) Marriage and desistance from crime in the netherlands: Do gender and sociohistorical context matter? J Quant Criminol 25:3–24
Block M, Heineke J (1975) A labour theoretical analysis of criminal choice. Am Econ Rev 65:314–325
Cappé O, Moulines E, Rydén T (2005) Inference in hidden Markov models. Springer, New York
CBS (2010) Centraal bureau voor de statistiek: standaardclassificatie misdrijven. http://www.cbs.nl
CBS (2012) Centraal bureau voor de statistiek: crime rates database, The Netherlands. http://statline.cbs.nl
Chalfin AJ, Raphael S (2011) Work and crime. In: Tonry M (ed) Oxford handbook of crime and criminal justice. Oxford University Press, Oxford
Chamberlain G (1980) Analysis of covariance with qualitative data. Rev Econ Stud 47:225–238
Cohen MA (1998) The monetary value of saving a highrisk youth. J Quant Criminol 14:5–33
Cook PJ (1980) Research in criminal deterrence: laying the groundwork for the second decade. In: Morris N (ed) Crime and justice: an annual review of research. University of Chicago Press, Chicago
Cook PJ (1986) The demand and supply of criminal opportunities. In: Tonry M, Morris N (eds) Crime and justice: an annual review of research. University of Chicago Press, Chicago
Cook PJ, Kang S, Braga AA, Ludwig J, O’Brien ME (2014) An experimental evaluation of a comprehensive employmentoriented prisoner reentry program. J Quant Criminol 43:1–28
Davidson R, MacKinnon JG (2004) Econometric theory and methods. Oxford University Press, New York
de Mooij R (2006) Reinventing the welfare state (Publication number 60 of the CPB, Netherlands Bureau for Economic Policy Analysis, available from http://www.cpb.nl/node/10226)
Dobkin C, Puller S (2007) The effects of government transfers on monthly cycles in drugs abuse, hospitalization, and mortality. J Public Econ 91:2137–2157
Durbin J, Koopman SJ (2012) Time series analysis by state space methods, 2nd edn. Oxford University Press, Oxford
Durlauf SN, Navarro S, Rivers DA (2010) Understanding aggregate crime regressions. J Econom 158:306–317
Ehrlich I (1973) Participation in illegal activities: a theoretical and empirical investigation. J Polit Econ 81:521–565
Fishback PV, Johnson RS, Kantor S (2010) Striking at the roots of crime: the impact of welfare spending on crime during the great depression. J Law Econ 53:715–740
Foley CF (2011) Welfare payments and crime. Rev Econ Stat 93:97–112
Freeman RB (1999) The economics of crime. In: Ashenfelter O, Card DE (eds) Handbook of labor economics, vol 3c. Elsevier Science, Amsterdam, pp 3529–3571
Goodman P (1956) Growing up absurd. Random House, New York
Gottfredson M, Hirschi T (1990) A general theory of crime. Stanford University Press, Stanford
Gould ED, Weinberg BA, Mustard DB (2002) Crime rates and local labor market opportunities in the United States: 1977–1997. Rev Econ Stat 84:45–61
Grogger J (1995) The effect of arrest on the employment and earnings of young men. Q J Econ 110:51–71
Grogger J (1998) Market wages and youth crime. J Labor Econ 16:756–791
Heckman JJ (1981a) Heterogeneity and state dependence. In: Rosen S (ed) Studies of labor markets. The National Bureau of Economic Research, The University of Chicago Press, Chicago, pp 91–140
Heckman JJ (1981b) The initial parameters problem of initial conditions in estimating a discrete timediscrete data stochastic process. In: Manski CF, McFadden DL (eds) Structural analysis of discrete data and econometric applications. The MIT Press, Cambridge, pp 179–195
Heckman JJ (1981c) Statistical models for discrete panel data. In: Manski CF, McFadden DL (eds) Structural analysis of discrete data and econometric applications. The MIT Press, Cambridge, pp 113–177
Hirschi T (1969) Causes of delinquency. University of California Press, Berkeley
Hyslop DR (1999) State dependence, serial correlation and heterogeneity in intertemporal labor force participation of married women. Econometrica 67:1255–1294
Jungbacker B, Koopman SJ (2007) Monte Carlo estimation for nonlinear nonGaussian state space models. Biometrika 94:827–839
Jungbacker B, Koopman SJ, van der Wel M (2014) Smooth dynamic factor analysis with application to the US term structure of interest rates. J Appl Econom 29:65–90
Kelly M (2000) Inequality and crime. Rev Econ Stat 82:530–539
Lageson S, Uggen C (2013) How work affects crime—and crime affects work—over the life course. In: Gibson CL, Krohn MD (eds) Handbook of lifecourse criminology: emerging trends and directions for future research. Springer, New York
Laub JH, Sampson RJ (2003) Shared beginnings, divergent lives: delinquent boys to age 70. Harvard University Press, Cambridge
Lemert E (1967) Human deviance, social problems, and social control. PrenticeHall, Englewood Cliffs
Levitt SD, Venkatesh SA (2001) An economic analysis of a drugselling gang’s finances. Q J Econ 39:755–789
Lin MJ (2008) Does uemployment increase crime? J Hum Resour 43:413–436
Machin S, Meghir C (2004) Crime and economic incentives. J Hum Resour 39:958–979
Maruna S (2001) Making good: How exconvicts reform and rebuild their lives. American Psychological Association Books, Washington, DC
Merton RK (1938) Social structure and anomie. Am Sociol Rev 3:672–682
Mesters G, Koopman SJ (2014) Generalized dynamic panel data models with random effects for crosssection and time. J Econom 180:127–140
Mundlak Y (1978) On the pooling of time series and cross section data. Econometrica 46:69–85
Mustard DB (2010) How do labor markets affect crime? New evidence on an old puzzle. In: Benson BL, Zimmerman PR (eds) Handbook on the economics of crime. Edward Elgar Publishing, New York
Nagin D, Paternoster R (2000) Population heterogeneity and state dependence: future research. J Quant Criminol 16:117–145
Nocedal J, Wright SJ (1999) Numerical optimization. Springer, New York
Pager D (2003) The mark of a criminal record. Am J Sociol 108:937–975
Pager D (2007) Marked: race, crime and findings work in an era of mass incarceration. University of Chicago Press, Chicago
Pager D, Western B, Bonikowski B (2009) Discrimination in a lowwage market: a field experiment. Am Sociol Rev 74:777–799
Pesaran HM (2006) Estimation and inference in large heterogeneous panels with a multifactor error structure. Econometrica 74:967–1012
Phelps ES (1994) Lowwage employment subsidies versus the welfare state. Am Econ Rev Papers Proc 84:54–58
Poirier DJ (1976) The econometrics of structural change: with special emphasis on spline functions. NorthHolland, Amsterdam
Raphael S (2011) Improving employment prospects for former prison inmates: challenges and policy. In: Cook PJ, Ludwig J, McCrary J (eds) Controlling crime: strategies and tradeoffs. University of Chicago Press, Chicago
Raphael S, WinterEbmer R (2001) Identifying the effect of unemployment on crime. J Law Econ 44:259–283
Ripley BD (1987) Stochastic simulation. Wiley, New York
Sampson RJ, Laub JH (1993) Crime in the making: pathways and turning points through life. Harvard University Press, Cambridge
Sampson RJ, Laub JH (2005) A lifecourse view of the development of crime. Ann Am Acad Polit Soc Sci 602:12–45
Shanahan MJ (2000) Pathways to adulthood in changing societies: variability and mechanisms in life course perspective. Annu Rev Sociol 26:667–692
Shapiro J (2005) Is there a daily discount rate? Evidence from the food stamp nutrition cycle. J Public Econ 89:303–325
Steinberg L, Albert D, Cauffman E, Banich M, Graham S, Woolard J (2007) Age differences in sensation seeking and impulsivity as indexed by behavior and selfreport: evidence for a dual systems model. Dev Psychol 44:1764–1777
Stolzenberg LJ, D’Alessio SJ (2007) The effect of divorce on domestic crime. Crime Delinq 53:281–302
Tannenbaum F (1938) Crime Commun. Columbia University Press, London
Thornberry TP, Christensen RL (1984) Unemployment and criminalinvolvement: an investigation of reciprocal causal structures. Am Sociol Rev 49:398–411
van der Geest V, Blokland A, Bijleveld C (2009) Delinquent development in a sample of highrisk youth: shape, content and predictors of delinquent trajectories from age 12 to 32. J Res Crime Delinq 46:111–143
van der Geest V, Blokland A, Bijleveld CCJH (2011) The effects of employment on longitudinal trajectories of offending: a follow up in high risk youths from ages 18–32. Criminology 49:1195–1234
van Oorschot W (2006) The Dutch welfare state: recent trends and challenges in historical perspective. Eur J Soc Secur 8:57–76
Wijkman M, van der Geest V, Bijleveld C (2006) Delinquentie van op civielrechtelijke en strafrechtelijke titel behandelde jongeren (delinquency after treatment in a juvenile justice institution under civil or criminal law). Tijdschrift voor Familie en Jeugdrecht 1:18–27
Wilkinson R, Pickett K (2009) The spirit level: Why more equal societies almost always do better. Allen Lane, London
Wooldridge JM (2005) Simple solutions to the initial conditions problem in dynamic, nonlinear panel data models with unobserved heterogeneity. J Appl Econom 20:39–54
Zhang J (1997) The effect of welfare programs on criminal behavior: a theoretical and empirical analysis. Econ Inq 35:120–137
Zweig J, Yahner J, Redcross C (2011) For whom does a transitional jobs program work? Examining the recidivism effects of the center for employment opportunities program on former prisoners at high, medium, and low risk of reoffending. Criminol Public Policy 10:945–972
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix 1: Estimation Method
In this appendix we discuss the Monte Carlo maximum likelihood method that is used to estimate the parameters of the logistic trivariate panel data model that is discussed in “Statistical Model” section. The methodology is discussed for a more general panel data model in Mesters and Koopman (2014). All parameters are placed in the vector \(\psi\), which includes the unrestricted elements of \(\Gamma , \beta , \delta _0, \delta _1, \lambda , \Sigma _v\) and the knots of the splines.
We summarize the vector of dependent variables for individual i in time period t by \(Y_{i,t} = (C{i,t},E_{i,t},W_{i,t})^{\prime}\), which is thus a \(3\times 1\) vector of binary variables. The loglikelihood for the observations \(Y=\{Y_{i,t}\}_{i=1,\ldots ,N \ t=1,\ldots ,T}\) is defined as \(\ell (\psi ;Y) = \log p(Y;\psi )\), where \(p(Y;\psi )\) is the joint density of all observations. In the presence of the random effects \(\mu _i\), defined in (3), we can express the joint density as a high dimensional integral as follows
where \(\mu = \{\mu _i\}_{i=1,\ldots ,N}\) and \(p(\mu ;\psi )\) is defined in Eq. (3). The conditional density \(p(Y\mu ;\psi )\) for the trivariate model can be written as
where
where \(\theta _{i,t} = (\theta _{C,i,t},\theta _{E,i,t},\theta _{W,i,t})\) is given in (2) and \(Y_{j,i,t}\) corresponds to the outcome variables \(Y_{C,i,t} = C{i,t}, Y_{E,i,t} = E_{i,t}\) and \(Y_{W,i,t} = W_{i,t}\) (Durbin and Koopman 2012, e.g., Section 10.3).
As \(p(Y\mu ;\psi )\) corresponds to a logistic binary density no closed form solution exists for the high dimensional integral in (9). Instead we follow the conventional literature and solve the integral using Monte Carlo methods. We refer to Cappé et al. (2005) and Durbin and Koopman (2012, Part 2) for general introductions into these methods. A simple Monte Carlo estimate is obtained by drawing S samples of \(\mu\) from \(p(\mu ;\psi )\) and computing the average
where \(\mu ^{(s)}\) denotes the sth sample from \(p(\mu ;\psi )\). From the law of large numbers it follows that \(\hat{p}(Y;\psi ) \rightarrow p(Y;\psi )\) as \(S \rightarrow \infty\). However, the simple estimate requires many draws S before convergence is achieved. This follows as the density \(p(\mu ;\psi )\) does not account for the observations Y.
More efficiency can be obtained by sampling sequences for \(\mu\) from an appropriate importance density (Ripley 1987). For the construction of an adequate importance density we follow Jungbacker and Koopman (2007) and Mesters and Koopman (2014). The general importance sampling representation for the trivariate model is given by
where \(g(\mu Y)\) is the importance density. When applying Bayes rule to the right hand side we obtain
where we have imposed \(g(\mu ) = p(\mu )\). A Monte Carlo estimate for the importance sampling representation is given by
where samples \(\mu ^{(s)}\) are drawn independently from importance density \(g(\mu Y)\).
We choose \(g(\mu Y)\) to follow a Gaussian density with mean equal to the mode of \(p(\mu Y)\) and variance equal to the curvature around the mode. An instrumental basis for \(g(\mu Y)\) that allows us to obtain the mode is given by
where \(z_i\) and \(D_i\) are obtained by the following Gauss–Newton algorithm.
Algorithm

1.
Initialize \(\mu = \mu ^*\);

2.
Given \(\mu ^*\); compute
$$D_i =  \left[ \sum _{t=1}^T \frac{\partial ^2 \log p(Y_{i,t}\mu _i^*;\psi )}{\partial \mu _i^* \partial \mu ^{*\prime}_i} \right] ^{1},$$and
$$z_i = \mu _i^* + D_i \sum _{t=1}^T \frac{\partial \log p(Y_{i,t}\mu _i^*;\psi )}{\partial \mu _i^*},$$for \(i=1,\ldots ,N\);

3.
Update \(\mu ^*\) by computing \({\mathrm{E}}_{g}(\mu  z)\) based on \(z_i = \mu _i + u_i\) and \(u_i \sim NID(0,D_i)\);

4.
Iterate between (2) and (3) until convergence.
Convergence of the algorithm is typically quick (4–5 iterations). The derivatives in step (2) are given in Durbin and Koopman (2012, Part 2). After convergence we have obtained the mode of \(p(\mu Y;\psi )\) and we can sample S times from the importance density \(g(\mu Y) \equiv g(\mu z)\), where \(g(\mu z)\) is a Gaussian density where the mean and variance are implied by \(z_i = \mu _i + u_i\) and the distribution of \(\mu _i\) given in (3). Using these samples we construct the Monte Carlo likelihood. The resulting likelihood estimate \(\hat{p}(y;\psi )\) is optimized with respect to parameters \(\psi\) by numerical methods (Nocedal and Wright 1999). This is done while using the same random numbers and the same number of draws S in each iteration.
Appendix 2: Factor Splines
In this appendix we provide the details for the construction of the cubic splines that we use to model the factors. More details for methods using splines can be found in Poirier (1976). In principal, it is possible to treat all the factors \(f_{j,t}\) as deterministic parameters and estimate them along with the other parameters. However, since the time series dimension is \(T=168\) this would lead to difficulties in optimizing the likelihood using numerical methods.
To avoid this problem, we make the assumption that the individual preferences and abilities vary smoothly with age. This allows us to fit cubic splines for the factors, which rely on a smaller number of parameters. In particular, we seek a subset of K knots denoted by \(\bar{f}_{R(l),t}\), for \(l=1\ldots ,K\), where \(R(l) \in \{1,\ldots ,T\}\). The locations R(l) of the knots are increasing with age; i.e., \(R(1) < R(2) < \cdots < R(K)\). Between these knots we fit cubic polynomial functions to approximate the factors that lie between the knots. The knots \(\bar{f}_{R(l),t}\) are estimated along with the other parameters. The location of the knots can be determined in a variety of ways (Jungbacker et al. 2014). In this paper we set the locations equal to the first month of every age year. Thus, we take in total 15 knots with are placed at age 18 month 1, age 19 month 1 etc. The final knot is for age 31 month 12.
Rights and permissions
About this article
Cite this article
Mesters, G., van der Geest, V. & Bijleveld, C. Crime, Employment and Social Welfare: An IndividualLevel Study on Disadvantaged Males. J Quant Criminol 32, 159–190 (2016). https://doi.org/10.1007/s1094001592585
Published:
Issue Date:
DOI: https://doi.org/10.1007/s1094001592585