Race and the Religious Contexts of Violence: Linking Religion and White, Black, and Latino Violent Crime

Abstract

Research has demonstrated that concentrated disadvantage and other measures are strongly associated with aggregate-level rates of violence, including across racial and ethnic groups. Less studied is the impact of cultural factors, including religious contextual measures. The current study addresses several key gaps in prior literature by utilizing race/ethnic-specific arrest data from California, New York, and Texas paired with religious contextual data from the Religious Congregations and Memberships Survey. Results suggest that, net of important controls, (1) religious contextual measures have significant crime-reducing associations with violence; (2) these associations are race/ethnic specific; and (3) religious contextual measures moderate the criminogenic association between disadvantage and violence for blacks. Implications for future research are discussed.

ACKNOWLEDGMENTS

This research made is possible by National Science Foundation Grant SES-0719648.

NOTES

Notes

¹ In our view, the religious context includes religious adherents, their beliefs, and prescriptions for behavior, as well as the institutions to which they belong. Regarding the latter, Scott (2008:48) defines institutions as “multifaceted, durable social structures made up of symbolic elements, social activities, and material resources” that include norms and behavioral regulation (social control) as central ingredients. As we elaborate throughout, our empirical examination focuses on the population of adherents and the relative diversity of the adherent population that is consistent with much prior empirical research (e.g., Lee and Bartkowski 2004; Lee et al. 2010) and has been shown to impact crime net of institutional/denominational structural measures (e.g., Beyerlein and Hipp 2005). Moreover, as we also detail, the theoretical frameworks utilized throughout the current study are not inconsistent with adherent-centered measures of the religious context in that adherents are those who act out the prescriptions of religious institutions on a day-to-day basis (or rather, put the words of institutions into action) and are, therefore, central to the link between religious contexts and crime. Indeed, even Stark's (1996:164) statement encourages social scientists to focus on “the proportion of persons in a given ecological setting who are actively religious” (see also Lee and Bartkowski 2004).

² While much prior research and theorizing suggest that more religious contexts may be associated with reduced rates of crime and violence, some scholars suggest that this may vary by denomination (e.g., Evangelical Protestant) and context (Ellison et al. 2003; Desmond et al. 2010; Lee et al. 2010; Shihadeh and Winters 2010). For instance, Ellison et al. (2003) find that Evangelical Protestant adherence is associated with greater violence in Southern (but not Northern) cities, while Beyerlein and Hipp (2005) find that evangelical adherence was positively associated with higher overall crime rates. These authors argue that while mainline and civically engaged adherent populations, along with perhaps Catholic adherents, foster “bridging” social capital, Evangelical Protestant adherents mostly foster “bonding” social capital that leads to insularity and comparatively less social control for the larger community. Similarly, Lee et al. (2010) find that Evangelical Protestant adherence was associated with more argument homicides for whites and blacks in both the North and South, although the effects were smaller for blacks. While it is beyond the scope of the current study to do so, such scholarship points to the need for more empirical research examining specific types of denominations in relation to violent crime, since there are good reasons to suspect that not all have the same effects on crime (as well as differing effects across racial and ethnic groups). We return to this point in our supplemental analyses and in discussing directions for future research that might include disentangling how various denominational groups foster violence in specific contexts.

³ Research on religious pluralism, particularly its impact on religious participation, is a growing body of literature in the sociology of religion (see Finke and Stark 2005). Yet, our review reveals virtually no research on the link between religious pluralism/homogeneity and social problems, such as crime or violence (for one exception, see Ellison et al. 1997, who examined religious homogeneity as it impacted suicide at the macro-level).

⁴ On the other hand, some scholars have argued that religious contexts should partially mediate the relationship between disadvantage and crime (Maume and Lee 2003). A full treatment of whether religious contexts mediate the effect of disadvantage on violence is beyond the scope of this article, but supplemental analyses (available from authors) find no significant mediation of the disadvantage–violence relationship through religious contexts.

⁵ For the year 2000, the overall racial/ethnic composition for the United States was roughly 69 percent white, 12 percent black, and 13 percent Latino. The racial/ethnic composition of the combination of California, New York, and Texas is comparable at roughly 52 percent white, 10 percent black, and 28 percent Latino.

⁶ One limitation of our data is that we are unable to capture communities with the largest concentration of black residents (e.g., the southern “Black Belt”) and, thus, may be unable to fully examine the link between the religious context and violence for blacks. However, the central goal of the current study is to compare the impact of religious contextual measures on broader violence (versus lethal crime only) and across three racial/ethnic groups (whites, blacks, and Latinos). Restricting our sample to counties in the South with larger black populations would not provide enough Latinos for comparison and, moreover, there are significant problems with the race/ethnic coding of the data used in much prior research (e.g., Uniform Crime Reports) that provide information on criminal offending in the South (as well as many other areas of the United States) (for a review, see Steffensmeier et al. 2011). As such, our California, New York, and Texas data are instrumental in (1) broadening the criminal landscape beyond homicide while (2) allowing for a comparison across whites, blacks, and Latinos without confounding these racial/ethnic groups, and (c) still capturing sizable populations of all three race/ethnic groups (which is necessary to compare across each). In short, our data may not be ideal for the analysis of any one racial/ethnic group (a point which we note is a direction for future research), but are perhaps the most valuable source of crime data for addressing the shortcomings of prior research we have identified and for answering our specific research questions.

⁷ Even utilizing the revised estimates, these data are known to undercount black Baptists and Latino and nondenominational congregations in the Latino Pentecostal realm that are not part of the Glenmary process (see, for example, Finke and Scheitle 2005). Some attempts have been made to correct these problems, but minorities in marginal denominations or those which are nondenominational often get overlooked. Again, we know of no comparable alternative sources of data, and while we are not focusing specifically on denominational effects, it is important to keep in mind that many religious adherents (particularly minorities) may still be missing despite efforts to capture them. We thank a reviewer for drawing our attention to this cautionary issue and note that the undercount of certain minority and small denominational members may actually work to conservatively bias our results—by missing certain minority and smaller denominational adherents, the relationship between total adherence, civically engaged adherence, and religious homogeneity may more strongly reduce violence for Blacks and Latinos than we observe here.

⁸ An analysis of influential cases suggests that our results are not sensitive to the sample selection criteria. Cook's D values are all well below 1.0 (highest is .421), indicating no consistently influential cases (Agresti and Finlay 1997). We also note here that, as a result of our selection criteria, our sample of counties is overwhelmingly urban. As such, we do not control for percent urban (or other similarly invariant macrostructural traits associated with an almost exclusively urban sample) as there is so little variation in this measure. Because our sample selection criteria eliminate many smaller, rural counties from the analysis, we utilized three-year averaged violence rates rather than five-year (or longer) averages because (1) Diagnostic analyses revealed no problematic outliers or influential cases in our sample (and, hence, no counties whose size appeared to skew the crime data). and (2) the authors' examination of violence rates using additional years of data (five years) revealed virtually identical rates to those employed in the current study. Moreover, much prior macrostructural criminological research has relied upon three-year averaged rates to the point that this has now become somewhat standard as long as reasonable population criteria are imposed (see, for example, Steffensmeier et al. 2010). Regarding the temporal period covered (1999–2001), we recognize that more recent data would be ideal; however, crime data are often delayed for several years before release to academics and the general public and, as such, we utilized data from the most recent time point at this writing for which religion and crime data were available for all three states.

⁹ We follow prior research in measuring the civically engaged religious population. However, we also recognize that civic engagement measured nationally (via the General Social Survey) may miss important regional heterogeneity within a denomination. That is, denominations deemed to be civically engaged may not be so in all places, just as non-civically engaged denominations in some communities may indeed exhibit significant community outreach. Unfortunately, capturing such heterogeneity (to our knowledge) is impossible with the current religious data and the extent of such patterns (and subsequently, their implications for the findings presented here and in prior research) remains unsettled. Nevertheless, we utilize this measure as (1) it has been argued to be the central way that religious adherents and their institutions leverage social capital and impact crime in communities (as argued by the civil society literature—see Lee and Bartkowski 2004), and (2) it has been the primary focal variable in prior research against which we are able to compare findings. We also emphasize here that we cannot control for denominations and adherent populations simultaneously because of collinearity. Diagnostics reveal that because our models are race/ethnic specific (with small crime counts in some instances), we cannot control for strongly correlated measures like congregational presence (i.e., per capita prevalence of specific congregations) and rates of adherence in the same models. Moreover, we reiterate that it is not our goal here to disentangle the impact of congregations relative to adherents (see, for example, Beyerlein and Hipp 2005), but rather demonstrate the salience of religious contextual measures (here, at the adherent level) across racial and ethnic groups on rates of violence.

¹⁰ Our multigroup religious homogeneity (RH) measure can be expressed formally as follows:

where, π_m is the proportion of people adhering to religious group m (e.g., proportion Catholic) and M is the total number of religious groups (here, five). E has a maximum value of 1 when a county has no diversity and is composed entirely of one racial/ethnic group and a minimum value of 0 when each religious group is equally represented. Religious homogeneity scores were divided by their maximum values to impose a range of 0 to 1. Our religious homogeneity measure is virtually identical to the more common measure of religious pluralism (see Finke and Stark 1988; Finke, Guest, and Stark 1996). In our data, the correlation between the two measures is approximately .85 and models using a pluralism measure produced results paralleling our findings for the religious homogeneity measure (available upon request). Thus, we present the homogeneity index for two reasons. First, it is standardized to have a defined range between 0 and 1 and is easy to interpret (see endnote 8). Second, this measure is not only easy to interpret, but it is also easy to compare with other key macrostructural variables, such as racial/ethnic diversity (entropy). Indeed, just as Finke and Stark (1988:44) note that the more common pluralism index “is based on a probability equation commonly used to measure linguistic diversity,” our homogeneity measure is developed very simply employing the same formula used for computing racial/ethnic diversity with population proportions (rather than raw population counts).

¹¹ More specifically, for each race/ethnic group, we ran principal component analyses for poverty, unemployment, female headship, and low education with factors retained at a cutoff Eigenvalue of 1. In all cases, poverty, unemployment, and female headship loaded on a single factor. This process, in sum, yielded three new factor score variables (one for each race/ethnic group) where each set of scores had a standardized sample mean of zero and standard deviation of one, which we then saved and used in our subsequent regression models as our indexes of concentrated disadvantage (see, for example, Land et al. 1990). After creating our three disadvantage indexes, all variables in our models are correlated at .41 or below and all variance inflation factors are below 2.0 for all variables in each model (well below the threshold of 10 that is frequently used—see Kutner, Nachtsheim, and Neter 2004), indicating that there is little evidence of problematic multicollinearity in our final models.

¹² Our entropy measure (E) for racial/ethnic heterogeneity is derived from the same manner as our religious homogeneity measure and is expressed as follows:

where, π_m is the proportion of the population belonging to race/ethnic group m (e.g., whites) and M is the total number of race/ethnic groups. E has a minimum value of 0 when a census place has no diversity and is composed entirely of one racial/ethnic group and a maximum value of 1 when blacks, whites, and Latinos are equally represented. Entropy scores were divided by their maximum values to impose a range of 0 to 1.

¹³ We also ran supplemental models examining the effects of the separate RELTRAD groupings (Steensland et al. 2000) on race/ethnic-specific violence. For white violence, the results for Mainline Protestant and Jewish adherence paralleled the strong, negative effects of civically engaged denomination adherence in Table 3. For black violence, Evangelical Protestant adherence and Catholic adherence both showed marginally significant negative effects. For Latino violence, Mainline and Evangelical Protestant adherence exhibited negative but marginally significant effects.

¹⁴ We note that the R-squared values observed in Table 3 exhibit small changes from those observed in Table 2. Although it is beyond the scope of the current study to fully expound on this issue, we emphasize, first, that in most cases, the changes in the R-squared values are trivial. Second, we are hesitant to place too much emphasis on the R-squared values and are cautious in drawing conclusions from these statistics given that some scholars (e.g., McElroy 1977) question their validity in SUR models and treat them primarily as goodness-of-fit measures that depend on the correlation of errors across equations. Third, and related to our second point, adding additional variables (like interaction terms) in SUR models will necessarily yield a new error variance-covariance matrix with correspondingly unique explained variance values that are not comparable with other other SUR models with their own unique matrix.