Abstract
Despite the importance of relative inequality in studies on race, place and crime, existing literature has paid little attention to the possibility of racial differences in the effect of relative inequality at the neighborhood level. I argue the anomie and social disorganization frameworks make competing predictions regarding whether relative inequality aligns with the racial invariance thesis, and assess hypotheses derived from each perspective using data from the 2000 National Neighborhood Crime Study. A comparison of marginal effects derived from multilevel negative binomial regression models indicate relative inequality effects on homicide are larger in neighborhoods comprised primarily by Blacks and Latinos, while effects on robbery and burglary are greater in White, Latino and Integrated areas. My observations add to a growing body of work demonstrating the importance of income inequality for neighborhood crime but suggest the magnitude of the effect varies considerably by crime type and neighborhood ethnoracial composition.
Acknowledgements
The author would like to thank Ruth Peterson and Lauren Krivo for the original NNCS data collection; Reuben “Jack” Thomas for statistical advice; Karen Parker for helpful comments following a roundtable presentation; and Christopher Lyons, Noah Painter-Davis, Maria Vélez, Sharon Nepstad, and Machienvee Lammey for support and feedback throughout the manuscript’s completion. An earlier version of this article was presented by the author at the 2018 Annual Meeting of the American Society of Criminology.
Disclosure Statement
No potential conflict of interest was reported by the author.
Notes
1 In the current study, I use the phrase “relative inequality” to refer to the same concept as does the more common phrase “relative deprivation.” I distinguish between relative and absolute inequality to draw attention to the way the term “inequality” has typically been employed in neighborhood crime research, which has been to refer to socioeconomic disparity across, rather than within, local communities (Hipp, Citation2007).
2 Of the 29 dropped cases, 20 were designated in the NNCS as partial tracts (i.e. subsections of whole tracts that straddle census place boundaries and thus appear in more than one city), although the NHGIS provides data on most partial tracts. The other 9 represented combined tracts located in Seattle, Milwaukee, and Detroit. In these and several other cities, police departments provided crime counts using 1980 or 1990 tract boundaries, so some of these tracts were combined in the NNCS so that data are comparable with 2000 census tracts.
3 In the census data for 2000, “Hispanic” and “Latino” refer to the same ethnic category, and thus in this paper I use the terms interchangeably. Latino-identified persons can be of any census racial identification.
4 Pearson’s r correlation coefficients between the Gini coefficient and the structural disadvantage index at the census tract level are .153 in White neighborhoods, .455 in Black neighborhoods, .364 in Latino neighborhoods, .443 in Minority neighborhoods, and .360 in Integrated neighborhoods. Refer also to Chamberlain and Hipp (Citation2015) for an example of separate application of the Gini coefficient and the structural disadvantage index across census tracts.
5 My dependent variable spatial lags are potentially endogenous regressors. Estimating their effects while controlling for simultaneity with the dependent variables would require, for example, a two-stage instrumental variable analysis or a simultaneous autoregressive (SAR) model specification. However, because I only employ the spatial lags as controls, conditional autoregressive (CAR) models that assume exogenous effects of spatial lags on the dependent variable are appropriate. I therefore utilize CAR models in the current study, recognizing their inherent limitations for explaining the complete spatial pattern of between-neighborhood crime. For an example of this approach when using spatially lagged dependent variables as controls see Mears and Bhati (Citation2006); for a detailed discussion of the relative advantages of CAR vs. SAR models, see Anselin (Citation2003).
6 Even when non-centered variables are used, however, there is limited evidence of collinearity problems in the models I estimate. Regardless of outcome, variation inflation factor (VIF) scores were consistently less than 3.0.
7 For a discussion comparing the relative advantages of different summary measures including the average marginal effect (AME), marginal effect at the mean (MEM), and marginal effect at representative values (MER), see Long and Freese (Citation2014:244-246).
8 I estimated the multilevel generalized structural equation models with the same control variables and exposure by tract population specification as in the earlier models. However, these models were fit using variable values as observed rather than grand-mean centered because convergence of the structural equation models necessitated use of the non-centered variables. In the multilevel negative binomial regression models, coefficients and standard errors were substantively similar regardless of whether grand-mean centered variables were used.
Additional information
Notes on contributors
Samuel A. Torres
Samuel A. Torres is a doctoral student of sociology at the University of New Mexico. His work focuses on how socioeconomic inequalities influence racial differences in crime at the macro-level. His research interests include absolute and relative deprivation factors of crime, political contexts of crime patterns, and links between drug markets and crime.