Do inequality, unemployment and deterrence affect crime over the long run?

ABSTRACT

Do inequality, unemployment and deterrence affect crime over the long run? Regional Studies. This paper investigates the long-run relationship between crime, inequality, unemployment and deterrence using US state-level data from 1978 to 2013. The novelty is to use non-stationary panels with a factor structure. The results show that: (1) a crime-theoretical model fits the long-run relationship well; (2) income inequality and unemployment have a positive impact on crime, whereas that of deterrence is negative; (3) the effect of income inequality on crime is larger when inequality is measured on a wider population proportion; and (4) property crime is generally highly sensitive to the deterrence effect of police.

摘要

不均、失业与吓阻长期而言是否会影响犯罪？ Regional Studies. 本文运用 1978 年至 2013 年美国州层级的数据，探讨犯罪、不均、失业与吓阻之间的长期关联性。其新颖之处在于运用具有因子结构的非平稳面板。研究结果显示：（1）犯罪—理论模型非常符合长期的关联性；（2）所得不均与失业对犯罪具有正向影响，而吓阻则具有负向影响；（3）在更广泛的人口比例中进行衡量时，所得不均对犯罪的影响较大；以及（4）一般而言，财产犯罪对于警方的吓阻效应具有高度敏感性。

RÉSUMÉ

Est-ce que l’inégalité, le chômage et les mesures de dissuasion ont un impact sur la criminalité à long terme? Regional Studies. À partir des données rassemblées entre 1978 et 2013 au niveau des états aux E-U, cet article examine le rapport à long terme entre la criminalité, l’inégalité, le chômage et les mesures de dissuasion. La nouveauté est d’utiliser des panels non-stationnaires dotés d’une structure factorielle. Les résultats laissent voir que: (1) un modèle théorique de la criminalité correspond bien au rapport à long terme; (2) l’inégalité des revenus et le chômage ont un effet positif sur la criminalité, tandis que l’impact des mesures de dissuasion s’avère négatif; (3) l’incidence de l’inégalité des revenus est d’autant plus importante que l’on mesure l’inégalité auprès d’une plus grande proportion de la population; et (4) la criminalité contre les biens est en règle générale particulièrement sensible à l’effet dissuasif des forces de l’ordre.

ZUSAMMENFASSUNG

Wirken sich Ungleichheit, Arbeitslosigkeit und Abschreckung langfristig auf die Kriminalität aus? Regional Studies. In diesem Beitrag untersuchen wir die langfristige Beziehung zwischen Kriminalität, Ungleichheit, Arbeitslosigkeit und Abschreckung anhand von Daten auf der Ebene von US-Bundesstaaten im Zeitraum von 1978 bis 2013. Das Neuartige an diesem Ansatz ist die Verwendung von nichtstationären Panels mit Faktorenstruktur. Aus den Ergebnissen geht Folgendes hervor: (1) Ein theoretisches Kriminalitätsmodell passt gut zur langfristigen Beziehung, (2) Einkommensungleichheit und Arbeitslosigkeit haben eine positive und Abschreckung eine negative Auswirkung auf die Kriminalität, (3) die Auswirkung der Einkommensungleichheit auf die Kriminalität ist stärker, wenn die Ungleichheit in einem größeren Bevölkerungsanteil gemessen wird, und (4) Eigentumsdelikte sind generell hochgradig sensibel gegenüber dem abschreckenden Effekt der Polizei.

RESUMEN

¿Afectan a largo plazo la desigualdad, el desempleo y la disuasión en la delincuencia? Regional Studies. En este artículo investigamos la relación a largo plazo entre delincuencia, desigualdad, desempleo y disuasión a partir de datos estatales de Estados Unidos de 1978 a 2013. La novedad en este planteamiento es el uso de paneles no estacionarios con una estructura de factores. Los resultados muestran que: (1) un modelo teórico de la delincuencia encaja bien en una relación a largo plazo; (2) las desigualdades de ingresos y el desempleo tienen un efecto positivo en la delincuencia, mientras que el de la disuasión es negativo; (3) el efecto de la desigualdad de ingresos en la delincuencia es mayor cuando se mide la desigualdad en un porcentaje más amplio de la población; y (4) los delitos contra la propiedad son en general altamente sensibles al efecto de disuasión por parte de la policía.

KEYWORDS:

• US crime
• deterrence
• inequality
• unemployment
• panel co-integration
• cross-section dependence

关键词:

• 美国犯罪
• 吓阻
• 不均
• 失业
• 面板共同整合
• 跨部门依赖

MOTS-CLÉS:

• criminalité aux E-U
• dissuasion
• inégalité
• chômage
• co-intégration des panels
• dépendance transversale

SCHLÜSSELWÖRTER:

• Kriminalität in den USA
• Abschreckung
• Ungleichheit
• Arbeitslosigkeit
• Panel-Kointegration
• Querschnittsdependenz

PALABRAS CLAVES:

• delincuencia en Estados Unidos
• disuasión
• desigualdad
• desempleo
• cointegración de panel
• dependencia transversal

JEL:

• C33
• E20
• K40

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

SUPPLEMENTAL DATA

Supplemental data for this article can be accessed at https://doi.org/10.1080/00343404.2017.1341626

ORCiD

Mauro Costantini http://orcid.org/0000-0002-6991-5159

Antonio Paradiso http://orcid.org/0000-0003-0083-1341

Notes

1. Generally, an increase in GDP improves the well-being of citizens, reduces the unemployment rate and affects other economic factors that may have an impact on crime (Malby & Davis, 2012).

2. Levitt (2004) shows that the growth rate of the US economy over last few years has played a marginal role in explaining the trend of crime.

3. The regional categorization follows the US Bureau classification. We are interested in understanding whether a common factor may lead to a co-movement in the US states’ crime rates. One simple way of identifying this factor is to find a linear combination of crime rates in the specific US regions (i.e., Northeast, South, West and Midwest). The first-principal component is calculated such that it accounts for the greatest possible variance in crime data. Therefore, the first component, denoted as , is computed as a linear combination of constituent variables, , such as , where denotes the weights. For more details, see Joliffe (2002).

4. Spelman (2006) concludes that a 10% increase in imprisonment rates produces, on average, a 2–4% decrease in crime rates.

5. Nagin (2013, p. 89) summarizes that ‘studies of police presence consistently find that putting more police officers on the street has a substantial deterrent effect on serious crime’. Becsi (1999) and Doyle et al. (1999) are the exceptions in the literature since in their analysis police increases crime.

6. In addition to the top 10% and Gini inequality measures used by Chintrakarn and Herzer (2012), this paper also considered the top 5%.

7. The model presented here is a static model, as in Edmark (2005), and is sufficient to represent the argumentations of the empirical framework.

8. In more general terms, may also capture costs associated with the recovery of stolen goods by the authorities.

9. It is plausible to assume that an unemployment shock (e.g., due to a technology innovation) will have a big impact on low-skilled workers (e.g., Brynjolfsson & McAfee, 2014).

10. In the strain theory of Merton (1938), it is stressed that individuals in the lowest scale of social structure tend to feel disadvantaged and alienated. In response to this, they are more inclined to commit violent crime.

11. In the log–log model, the estimated parameters represent the elasticity of the explanatory variables with respect to crime rate.

12. The average of unemployment benefits () and the psychological cost of crime () are not included in specification (9) due to lack of data (see also Edmark, 2005). In addition, we investigate the two forms of deterrence by comparing the coefficients of and in different equations. For further details, see note 20 below.

13. Both measures of deterrence may suffer from simultaneity bias in crime equations. This issue is addressed here by using the CUP estimator which accounts for endogeneity (and serial correlation). For details, see Appendix C in the supplemental data online. In addition, prison admission may suffer from ratio bias (Fisher & Nagin, 1978), especially when crime equations are estimated in first difference. In general, there is very little evidence of ratio bias for the US data (Levitt, 1998).

14. Data on crimes and prison admissions are taken from the Bureau of Justice Statistics, whereas data on police defence expenditures are from http://www.usgovernmentspending.com/. Data for unemployment rate are taken from US Bureau of Labour Statistics. Income inequality data are from Frank (2009) (available at http://www.shsu.edu/~eco_mwf). GDP per capita is taken from http://www.usgovernmentspending.com/. Resident population data are from the US Bureau of the Census. Total land area is from http://www.census.gov/geo/reference/state-area.html. Descriptive statistics of data are reported in Appendix A in the supplemental data online.

15. See Bai and Ng (2004) for the model with constant and trend.

16. For details, see Bai and Ng (2004).

17. As for the assumptions in the data-generating process (17) to (18), see Westerlund (2008).

18. Equation (9) is also estimated using standard panel one- and two-way fixed-effects estimators, as suggested by an anonymous referee. The results are available from the authors upon request. They contrast with the theoretical expectations, confirming previous findings in the crime literature, as shown by Neal (2015).

19. While a different model specification is proposed by Ehrlich (1973, 1975) that considers measures of probability of arrest, probability of conviction given arrest and probability of incarceration given conviction separately, this paper follows Edmark (2005) and Wu and Wu (2012) and assumes that when a criminal is caught they are incarcerated. The reason for this is that there are no available data for the three measures used by Ehrlich (1973, 1975) over a long time span for all the US states.

20. In order to test for the equality of the estimated coefficients for deterrence measures across different specifications, we run the Z-test suggested by Paternoster, Brame, Mazerolle, and Piquero (1998). In particular, we find that the coefficients of and are statistically different for property crime, whereas for violent crime those are statistically different only when the Gini inequality measure is considered. The results are available from the authors upon request.

21. Auto theft crime regression shows no co-integration for the top 5%, unemployment and , while it is instead statistically significant with the top 10%, unemployment and Gini measures (see Table 4).

22. Using the life-satisfaction approach to non-market valuation, Manning, Fleming, and Ambrey (2016) reveal that the intangible costs of crime can be significant when considered alongside what a state may spend on policing to achieve a reduction in crime.

23. A reinforcement of imprisonment policies may also produce a rise in wage inequality with an increasing impact on crime (Western, Kling, & Weiman, 2001; Western & Pettit, 2002). Individuals who are released from prison may suffer from low earnings and irregular employment. This may cause a deterioration in job skills and undermine potential connection with job opportunities. All this may produce an increase in crime (Hagan, 1993).

24. The choice of these two variables is dictated by the availability of data.

25. Levine, Grengs, Shen, and Shen (2012) also suggest that large urban density increases job opportunities and facilitates education access with an impact on crime.

26. Similar results are valid for different formulations of crime (i.e., robbery, burglary, auto theft and larceny theft) and the Gini index, but are not reported here due to space concerns. These results are available from the authors by request.