Serious Crime in Urban Neighborhoods: Is There a Race Effect?

Abstract

Most studies of the relationship between race and crime employ data aggregated at the city or SMSA level to show that high proportions of Blacks indicate high levels of crime. However, theoretical accounts of underlying criminogenic processes do not imply an effect for race, but rather involve explicit or implicit reference to the neighborhood as the unit of analysis. We combine 1990 census data with arrest records for 1989–1991 to generate complete block group information (n = 276) for a medium size southern city. Contrary to studies showing a relationship between racial composition and crime, we show that structural factors account for this effect completely. This finding challenges accounts of crime based on subculture and is consistent with accounts based on the social and economic structure of communities.

The authors wish to thank Charles Tolbert, Bill Bankston, Michael Irwin, Ryken Grattet, Mary Gautier, Michael Maume, and the Baton Rouge Police Department for assistance.

Notes

¹Land et al. mention one relatively consistent effect on homicide rates: percentage of children under age 18 not living with both parents. However, this was employed in only five of the studies reviewed and then only as a component in Loftin and Hill's (1974) Structural Poverty Index.

²One study employed percentage Black in one of two models estimated (Rosenfeld 1986). In four models the effect of percentage nonwhite was nonsignificant, but in all of these the state was the unit of analysis.

³ Blau and Blau (1982) are also careful to note that “racial composition primarily influences violence between persons who know each other, which does not at all correspond to the stereotype held by many whites, fearful of being attacked by a black stranger” (p. 125) though their aggregate data do not permit any such inference.

⁴The coefficient for percentage Black is positive in all cases and statistically significant in all cases except the state level analysis for 1980.

⁵The index consists of percentage of the population that is Black, percentage of children under 18 not living with both parents, median family income, percentage of families living below the official poverty line, and the Gini index of family income inequality.

⁶The “subculture of violence” was originally proposed as a thesis about southern violence in classic papers by Hackney (1969) and Gastil (1971). With few exceptions, analysts have been more interested in discussing the thesis as it applies to regional differences than to racial differences (Curtis 1975; Messner 1983).

⁷ Blau and Blau (1982) begin by allowing that the relationship between race and crime “has been interpreted as reflecting a subculture of violence in ghettos” (p. 114). Because of their emphasis on social inequality they do not pursue this line of interpretation further.

⁸ Bursik and Grasmick (1993) discuss the race and delinquency debate that has simmered since the 1940s (pp. 25–7).

⁹This is true of all studies reviewed by Land et al. (1990).

¹⁰ Bursik, Grasmick, and Chamlin (1990) discuss this issue in reference to ecological deterrence theory.

¹¹ Smith and Jarjoura (1989) find that racial composition of neighborhood is one of the most powerful predictors of burglary while race of household makes no difference (pp. 633–5). They use the risk of burglary victimization rather than the burglary rate as a dependent variable.

¹²Harlem is the only area in the study that has a majority of Blacks (90%+), and this is an aggregate of 40 census tracts and nearly 135,000 people (U.S. Bureau of the Census 1983; pp. 512–4). While we agree with their assertion that SMSAs and central cities are too large to appropriately examine neighborhood-level processes, we are hesitant about any inferences of race-crime links at the neighborhood level that are based on such highly populated and racially heterogeneous units.

¹³However, see Taylor et al.'s (1995) demonstration of the importance of street blocks in the study of land use patterns and physical deterioration.

¹⁴Offenses reported to the BRPD that did not take place within city limits were referred to the appropriate police jurisdiction and are not included in these data. Offenses occurring within city limits that are incorrectly reported to the sheriff's office are referred back to BRPD and, thus, are included in these data.

¹⁵This measure can underestimate the joblessness among Blacks because they are more likely to leave the labor force as discouraged workers. As an alternative measure of unemployment we use the proportion of male working age adults that are either unemployed or out of the labor force. The findings using this alternative measure do not alter the substantive conclusions of the article.

¹⁶Income inequality is typically measured using the Gini Coefficient of income distribution within the spatial unit. However, it is unlikely that all subjective evaluations of one's economic standing relative to some external standard are bounded by the neighborhood. Thus, our measure takes into account the economic standing of the neighborhood relative to its external environment.

*Offense rates per 100,000 population.

*P < .10

**P < .05.

¹⁷Because multicollinearity is often a problem in macro-level studies on crime, we examine the intercorrelations between predictor variables and the variance inflation factors of the parameter estimates. None of the intercorrelations were greater than .75 and virtually all were far below that. VIFs were all well below 4. Indeed, an added advantage of neighborhood level analyses over those at the city level is greater heterogeneity across spatial units, reducing the potential for multicollinearity. In regard to appropriate transformations, Land et al. (1990) note the inconsistency in past studies on whether and how variables should be transformed. Some do not transform variables, others use natural or common logs, or some other transformation. Some studies transform only the crime rates, others transform some of the independent variables, while some transform all variables. We follow the practice of Land et al. and report results for untransformed variables. However, to assess the sensitivity of our results to transformation, we reestimated models after transforming all variables with skew factors greater than 1.0 using log10, log to base e, or the inverse of the variables. Such transformations, whether methodologically justified or not, do not change our substantive conclusions.

*P < .10

**P < .05.

¹⁸Crime rates, being rare events, may take on a distribution that violates the normality assumptions of OLS. In response to this concern we reestimated all our models using negative binomial estimation. The results do not change our basic conclusions. As with the OLS results, these new results show that in the reduced models race is overwhelmingly predictive of crime. In the full models, once again, as with the OLS estimates, race is disconnected from crime once economic deprivation and other structural disadvantage is accounted for. This additional analysis further strengthens our inferences with regard to race and crime. See also Osgood (2000).

¹⁹This program, along with other street-level and community policing programs was widely credited with a drop in violent crime that began in 1993. Over 7,500 arrests were made and over 1,200 guns were confiscated. However, we have no direct evidence linking this program with a drop in crime that is clearly a nationwide trend.