The colors of fear: A multilevel analysis of fear of crime across Houston area neighborhoods

ABSTRACT

Research on fear of crime has revealed the impact that neighborhoods' racial composition has on individuals' perceptions of fear and safety. Yet, few studies have examined how between-neighborhood variation in diversity, as distinct from minority concentration, influences residents' perceptions. Using multilevel regression and drawing on data gathered from a sample of residents in the Greater Houston Area, we examine the between-neighborhood effects of diversity on perceived risk of victimization. Our findings suggest that both individual and community factors have utility for predicting fear of crime and that racial and ethnic heterogeneity influence fear differently for members of different racial groups.

KEYWORDS:

• Ethnicity
• race
• minorities
• theories

Acknowledgments

The authors would like to express appreciation to Larry Lyon, Elizabeth Embry, Wes Hines, Christopher Bader, and Kevin Dougherty for providing feedback on earlier versions of this article.

Notes

1. For a thorough discussion of the differences in demographic effect between group size and heterogeneity, see Blau (1977).

2. The HAS has asked Houston residents about their perceptions of neighborhood safety since the mid-1990s. While the question asked differs slightly from the GSS question and the fear of crime question that we use, it is reasonable to assume that it taps similarly into residents' perceptions about their risk of becoming a victim in their neighborhoods (Klineberg, 2005).

3. The geographic level used to represent Houston Area neighborhoods in this study is a Community Service Area (CSA). This is a geographical area constructed by researchers at the Center for Community Research and Development at Baylor University to approximate recognizable neighborhoods in the Houston area. CSAs are roughly equivalent to zip code areas in Houston, TX.

4. We restrict analyses to 1,862 respondents in 44 different neighborhoods. This smaller data set is the result of excluding cases that were missing data on dependent or independent variables by listwise deletion. Additionally, all of the cases from one Houston-area CSA were removed from analyses due to missing CSA-level data.

5. In the case of a standard two-level hierarchical linear model, it would be customary to present the variance components derived from the null model, a model with no predictors included at either level, in order to compute an interclass correlation. This would reveal the magnitude of level-2 variation relative to the model's total variation. However, computing the interclass correlation for a nonlinear hierarchical model is problematic because the level-1 variance is heteroscedastic (for a thorough discussion, see Raudenbush & Bryk, 2002).

Additional information

Notes on contributors

Eric Y. Liu

Eric Y. Liu earned his Ph.D. in sociology from Baylor University. His current research interests include advanced quantitative methods, customer behavior, and demographic impact on business and society. Dr. Liu's research findings, supported by prestigious research institutions such as NIH and China National Science Foundation, have been well recognized and published in numerous peer-reviewed journals.

Edward C. Polson

Edward C. Polson is an assistant professor in the Diana R. Garland School of Social Work at Baylor University. His current research interests include community engagement, civil society, the intersection of religion and civic life, and the work of nonprofit organizations and voluntary associations.