![Sample our Law journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5942/TOC-Subject-Sample-2021-Law.jpg"})
ABSTRACT
Scholarly attention directed toward hate crime, especially across communities, has grown in the past two decades. Rural communities, however, have been neglected in such empirical inquiry, driving issues on the ability to draw reliable conclusions on ecological variations of such offenses. Thus, we examined structural and demographic predictors of hate crime across rural and urban counties, focusing our attention on whether patterns varied in predicting anti-Black, Hispanic, and White crimes from 2012 through 2016. Using social disorganization and the defended communities perspective as explanatory frameworks, we find important differences and similarities between these settings, with implications for theory and research.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1. Both Alaska (Boroughs) and Louisiana (Parishes) do not feature traditional counties. However, these entities are similar in many regards. Furthermore, identical data for the independent and dependent measures were able to be gathered for each. A total of 41 independent cities were included within the analysis (38 in Virginia; St. Louis, Missouri; Baltimore, Maryland; District of Columbia (D.C.)), as they are considered county-equivalents for purposes of data collection by the American Community Survey (utilized to create the independent measures). Identical data were also gathered for these units.
2. No additional steps were required to assign incidents to independent cities.
3. The majority of the independent predictors were operationalized through data available in the 2016 ACS. However, the five-year estimates from the 2011 iteration were required to compute measures related to population group change over time.
4. There was quite a bit of difference in the population structures between rural and urban counties. While rural counties had much lower average population densities, rural counties were also slightly less racially diverse, older, and poorer.
5. Though other categories were considered (e.g., anti-Asian hate crimes), these outcome variables were selected due to the fact that they represented the majority of racially-motivated incidents in the U.S. during the time period under analysis.
6. The measure for heterogeneity was only explored in the models for total hate crimes. Inclusion of other predictors related to population unit size (e.g., percent Black, percent Hispanic) and population change presented potential issues with multicollinearity. Since these measures are integral to the defended communities perspective, their inclusion was selected over the heterogeneity measure.
7. Cronbach’s alpha was used to assess the reliability of the index for concentrated disadvantage. The resulting value (α = .83) indicated a suitable measure. Because each of the other indices were comprised of only two items, relying on the α value was not appropriate. As recommended by Eisinga et al. (Citation2013), Spearman-Brown reliability estimates were computed. The results for residential instability (.65) and heterogeneity (.60) provided sufficient support for the included items.
8. The defended communities perspective was not addressed in models assessing overall hate crime, as these counts cannot conceptually be argued to result from any one “group threat.”
9. Potential issues with multicollinearity were explored for population density and rural/urban county designation. Analysis of the bivariate coefficient, and tolerance and VIF values revealed that including population density as a variable in the models was not problematic.
10. Zero-inflated negative binomial models were considered due to the high number of zero values for the dependent measures. Estimates and standard errors were similar between these and the negative binomial models employed in the current study. Mandate, officer density, and South were selected as the inflation variables for these models in line with previous research on the topic. Since similarities were observed, negative binomial regression was selected with these measures as additional controls.
11. All offsets (total population, Black population, Hispanic population, White population) featured a fixed coefficient of one (1).
12. For the standardized social disorganization measures, changes in rates can be interpreted based upon a one standard deviation increase in the predictor. For all other measures, changes relate to a one unit increase in the predictor.
13. Since some counties featured significantly higher numbers of reported hate crimes, the potential impact of outliers on the findings was explored. Models with these outliers removed featured only minor differences.