Inequality and crime: evidence from Russia's regions

Abstract

We examine the effect of inequality on crime in Russia's 88 regional entities from 2000 to 2005. Using dynamic Generalized Method of Moments (GMM) estimation techniques, we consider both violent (murder and robberies) and nonviolent crime (thefts, economic crimes), and drug crimes and crimes committed by juveniles. The results indicate that inequality has a significant, positive effect on murders, robberies, thefts and juvenile crimes.

Keywords:

• inequality
• crime

JEL Classification:

• D31
• D63
• D71
• D81

Notes

¹ See, for example, The Spirit Level: Why More Equal Societies Almost Always Do Better (Wilkinson and Pickett, 2009), in which the authors show a cross-country correlation between inequality and a cadre of social and health ills.

² See, for example, Witt et al. (1998), Virén (2001) and Wu and Wu (2012).

³ See Fajnzylber et al. (2002a, b) and Lederman et al. (2002).

⁴ Gaviria (2000) and Scorzafave and Soares (2009) analyse crime in Colombia and Brazil, respectively, while Demombynes and Özler (2005) and Meera and Maliyakal (1995) focus on South Africa and Malaysia, respectively.

⁵ Misreporting and the resulting bias for most of the data should largely be similar to any statistical surveys in any other country. The methodology for most studies complies with international requirements. There is a case for a relatively higher misrepresentation of crimes data, as the sampling is based on data from a potentially biased party. However, the bias should be visible only on cross-country comparison, while deviations on a cross-regional level should be roughly consistent with the actual distribution.

⁶ For example, Khanty-Mansiyskii AO and Yamalo-Nenetskiy AO are reported as part of Tyumen region; Evenkiysky AO as part of Krasnoyarsk region; Ust-Ordynskiy AO and Aginsky-Buryatsky AO as part of Irkutsks region; Koryakskiy AO as part of Komchatsky region; and so on. Ingushetiya and Chechnya were a single region in 1990, and Chechnya was outside of statistical coverage for most of the 1990s.

⁷ Roodman (2006) shows that using OLS will cause upward bias on the lag of the crime rate, while using fixed effects will cause downward bias. Thus, the magnitude of the estimated coefficient on the lag of the crime rate resulting from an improved estimation strategy should fall between those of the OLS and fixed effects. In this article, this general result holds for the regressions for all crime rate types.

⁸ We also fail to reject both the null hypothesis for serial correlation of the idiosyncratic error and the joint validity of the instruments for all the regressions for the varying types of crime rates.