Social Structure and Family Homicides

One reviewer asked whether we employed an average population estimate for the years of 2000–2007 or if we only relied on 2000 census data. As indicated in the text we employed 2000 Census population data. In order to ensure the validity of this approach, we conducted two tests. First, we calculated the average population for each city for the years of 2000–2007. This examination confirmed the cities included in the analyses. It is important to note, however, that ten of the cities averaged slightly under one-hundred thousand residents (i.e., some cities averaged from ninety-eight to one-hundred thousand). Due to slight difference we decided against excluding these cities. Second, we calculated population density with the average population, and reran the regression analyses with this measure. In both sets of tests, the results were consistent with those presented here-in, and thus we present our OLS regression results on the various homicide rates for parsimony and consistency with the other Census variables employed in the study.

It is important to note that the bulk of social disorganization research has been conducted with smaller units of analyses (i.e., census tracts and neighborhoods). There have been studies; however, that test this construct in larger units such as cities, counties, and standard statistical areas (i.e., Beaulieu and Messner 2010 [examined cities]; McCall and Nieuwbeerta 2007 [examined cities]; Land et al. 1990 [examined cities, counties, and standard statistical areas]; Melde 2006 [examined counties]; Oh 2005 [examined cities]; Osgood and Chambers 2000 [examined counties]. This study models its methods on the aforementioned studies. The reason for this is twofold. First, the Supplementary Homicide Report (SHR) does not report homicide data for geographic units smaller than cities. Second, family homicides are extremely rare. Therefore, employing smaller units would not yield the necessary cases to conduct an analysis of this type. Nonetheless, it important to note that employing cities and other bigger geographic units in the testing of social disorganization constructs poses the threat of not remaining theoretically “true” to the original assumptions posed by Shaw and McKay (1942).

A multi-step approach was utilized when creating the components in order to ensure they are independent from each other. First, a factor analysis of all the theoretical relevant independent variables measuring social structure (see Table 1) included in the component matrix was employed. The results are available in appendix. The second step, involved selecting the variables in each component that had a factor score of .700 or more (see Kim and Mueller 1978). As seen in Appendix seven variables fit this criterion. In order to confirm the findings of the first component we reran the factor analyses with these variables. We also included in this analysis the % homeowner variable because it almost met the criteria with a factor loading of −.640. The results of this analyses yielded similar results from those presented in the table above (i.e., all of the variables with the exception of % home owner had factor loading above .700—% home owner had a factor loading of −.599). As result, we excluded % homeowner from the component and based our economic deprivation measure on the seven variables presented in Table 2. We then reran the factor analyses with only the three variables not included in the first component (i.e., % homeowner, % foreign born, and population density). When examined separately these variables yielded factor loading over .700 (See Table 2). As result, we used these variables to create our second component—social disorganization.

Shaw and McKay (1942) originally posited that communities with high number of immigrants from different countries could be related to high crime rates due to culture conflict through the inability of immigrants to communicate with one another in order to deal with community problems due to language barriers. This notion has recently been debunked with new research that shows a negative relationship between immigration and crime (Stowell et al. 2009).

One reviewer appropriately pointed out that some of the outcome measures examined here had more ‘zero incidents’ (i.e., no specific types of homicides), which has the potential to skew the outcome variables modeled. This would make OLS regression susceptible to the skewed distribution and would have the potential to bias our estimates (see Long 1997). Consistent with the approach presented by Osgood (2000) dealing with the same situation (i.e., aggregate crime rates), we performed a two-step process to test for this issue. First, we re-estimated our OLS models using logged dependent variables. Second, we conducted count regression models, which specifically tests for an overdispersed outcome with a probability chi-square distribution. In four of the six models, the p value >.05, which indicated the outcome variable did not suffer from overdispersion (Long and Freese 2003: 269–70). When the distribution was not skewed, we used Poisson regression, and when it was skewed we used Negative Binomial estimation. In both sets of tests, the results were consistent with those presented here-in, and thus we present our OLS regression results on the various homicide rates for parsimony and consistency with prior research.

In order to shed some light into this unexpected finding we re-estimated the models testing each of the social disorganization variables independently from each other. The independent regression analyses suggest that the negative effect with social disorganization is being driven by the percentage foreign born variable. Percent foreign born had a significant negative relationship in all, but one, of the models we estimated (i.e., total homicides, total family homicides, IPH, filicide, and siblicide). Population density, on the other hand did not reach significance in of the models, and percent homeowners was only significant in the total homicide model.