Econometrics and Criminology
MERGING DISCIPLINES: ECONOMETRICS AND CRIMINOLOGY
With this article, I hope to show you how broadly econometrics can be applied. In my research, I use econometric techniques to explain criminal behaviour using panel data. I start off by discussing a paper that is just accepted for publication in a criminological journal (Van de Werve, Blokland and Weerman, forthcoming) and thereafter I will discuss next steps.
Having access to crime data with offenses on the individual level is often challenging, because such data sets are not publicly available due to the sensitive nature of the data. Luckily, I have access to an interesting data set because, besides my research and teaching tasks at the VU Econometrics and Data Science department (EDS), I am also a fellow at The Netherlands Institute for the Study of Crime and Law Enforcement (NSCR). With my colleagues at NSCR, I analysed the impact of macroeconomic circumstances on individual-level crime and whether this effect is conditional on marital status and parenthood.
We are interested in three types of effects, as visualised in Figure 1. First, we consider the macro-effect measured by the parameter β, because we want to know whether economic decline has a direct impact on the probability that somebody commits a crime. We distinguish between two measures of macroeconomic circumstances. On the one hand we consider the unemployment rate, which can be seen as an objective and backward-looking indicator. However, people are more likely to respond to their own beliefs about the future. So, on the other hand, we also use the consumer confidence index. It reflects a level of optimism or pessimism towards the expected development of the economy in the next twelve months and therefore we consider it to be a subjective and forward-looking indicator. To formulate our hypothesis, we rely on several theories from economics and criminology. The rational choice theory by Becker (1968) implies that the expectation of maintaining a certain standard of living in a legal manner will be lower when the state of the economy is less favorable, so the “benefit” of committing a crime might be higher. Also the general strain theory by Agnew (1992) implies that the social environment can create incentives for criminal behavior. Therefore, our hypothesis for the macro-effect is that individuals are more likely to commit crimes in times of economic decline.
The second effect of interest is the micro-effect measured by γ. A lot of empirical research showed that married individuals are less likely to commit crimes, while parenthood has no impact on criminal behaviour (see, for example, the overviews by Craig et al., 2014, and Bersani and van Schellen, 2014). Our hypotheses are in line with these results.
Lastly, of main interest, is the interaction effect measured by δ. Analysing whether the effect of macroeconomic circumstances on criminal behaviour is conditional on marital status and parenthood has not been done before. Another challenge in formulating the hypotheses for this effect is that theories on which we can rely are not unequivocal: does marriage provide a financial safety-net or do married individuals feel responsible for being the breadwinner? We decided to hypothesise that the impact of macroeconomic decline on individual criminal behaviour is conditional on being married and on being a parent, but we leave room for the direction of the effect.
Figure 1: Simplified visual representation of study by Van de Werve, Blokland and Weerman (forthcoming).
Our data set originates from a Dutch large-scale longitudinal study and we enriched this with macro-level data from Statistics Netherlands. For 2,251 individuals we know how many crimes and which types of crime they committed between 1972 and 2006, along with some characteristics of the individuals. Since our outcome variable is binary (a dummy whether individual i committed a crime in year t), we set up a logit model. Introduce xt as the macro-level regressor (note the missing subscript i, as the regressor value is the same for all individuals in year t), zit as the micro-level regressors and wit as the control variables related to age and past criminal behaviour (because the past is often the best predictor of the future). Furthermore, let µi be the unobserved individual fixed effect and εit the idiosyncratic error. Our model is given by:
where the standard errors are clustered on individual level to account for serial correlation over time. We estimate this panel data model by within-estimation.
The results of the micro-effects are as expected: marriage reduces criminal behaviour and parenthood has no impact. So, let us focus on the macro- and interaction effects, which are the new contributions of this study. We find that individuals are more responsive in their criminal behaviour towards changes in the consumer confidence than changes in the unemployment rate. If the consumer confidence decreases (so the perception about the state of the economy in the future declines), we see that more crimes are committed. However, the positive effect of a decreasing consumer confidence is nullified by a negative interaction effect for married individuals. We think that the financial and/or mental safety-net of marital partners prevents criminal behavior when the consumer confidence decreases.
We also perform a couple of sensitivity analyses, of which one is worth highlighting here. We namely also find that unemployment rates above the trend might actually lead to less crime. So, for the unemployment rate, it seems that the “state” has more explanatory power than the “direction of change”. However, we also think that it might be more important what the own employment situation is rather than the nation’s average, but we do not have the data to test that further at this point.
Lastly, I would like to shortly touch upon next steps. The paper that I just discussed shows the potential of the research area, but is limited by a relatively simple model. For example, we assumed that the effect of macroeconomic circumstances on criminal behaviour is the same for all individuals and that there are no time-varying effects. Also, by using within estimation, we cannot estimate the individual-specific effects. In new research with colleagues from the EDS department, we will estimate individual-specific and time-varying effects using the approach by Mesters and Koopman (2014). The latter paper relies on performing importance sampling on two dimensions, so becomes computationally challenging for truly big data, while individuals can probably be grouped. Therefore, we will also use the clustering approach by Blasques et al. (2021) to measure to what extent individuals respond similarly to macroeconomic cycles.
References
Agnew, R. (1992). Foundation for a General Strain Theory of Crime and Delinquency. Criminology, 30(1), 47–88.
Becker, G. S. (1968). Crime and Punishment: An Economic Approach. Journal of Political Economy, 76(2), 169–217.
Bersani, B. E., & van Schellen, M. (2014). The Effectiveness of Marriage as an “Intervention in the Life Course: Evidence from the Netherlands. In J. A. Humphrey & P. Cordella (Eds.), Effective Interventions in the Lives of Criminal Offenders (pp. 101–119). Springer.
Blasques, F., Hoogerkamp, M. H., Koopman, S. J., & van de Werve, I. (2021). Dynamic Factor Models with Clustered Loadings: Forecasting Education Flows using Unemployment Data. International Journal of Forecasting, 37(4), 1426–1441.
Craig, J. M., Diamond, B., & Piquero, A. R. (2014). Marriage as an Intervention in the Lives of Criminal Offenders. In J. A. Humphrey & P. Cordella (Eds.), Effective Interventions in the Lives of Criminal Offenders (pp. 19–37). Springer.
Mesters, G., & Koopman, S. J. (2014). Generalized Dynamic Panel Data Models with Random Effects for Cross-section and Time. Journal of Econometrics, 180(2), 127-140.
Van de Werve, I., Blokland, A.A.J., & Weerman, F.M. (Forthcoming). Crime, Families and the Economy: Micro Conditions as Moderator of Macro Effects. Journal of Developmental and Life Course Criminology.
