A Social-Psychological Process of “Fear of Crime” for Men and Women: Revisiting Gender Differences from a New Perspective

ABSTRACT

Prior research has identified gender as a significant predictor of crime fear. Specifically, women are typically more fearful of crime than men, despite being relatively less likely to be victimized. The current study examines different ways men and women may think about crime and victimization within their neighborhoods, using contemporary social-psychological models of victimization worry. Data were collected from a sample of community residents (N = 713) living on the Gold Coast of Queensland, Australia. Results suggest that men’s and women’s fear of crime and perceptions of victimization threat are dependent on crime type and can be represented by a number of relationships among different social-psychological dimensions of victimization worry. The study concludes with practical implications for researchers seeking to examine the complex associations between gender and fear of crime.

KEYWORDS:

• atitudes toward crime
• gender
• social-psychological dimensions of crime fear
• gender-fear-victimization paradox

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. In light of Reid and Konrad’s (2004) findings and existing research related to socially desirable responses to self-reported questions (Sutton & Farrall, 2005), the “hegemonic masculine ideal” (Goodey, 1997), and the “shadow sexual assault hypothesis” (Ferraro, 1995, 1996), the models of gender difference in fear of crime that follow do not include rape or sexual assault as a type of crime that is analyzed.

2. In the current manuscript the term general fear of crime encompasses fear of personal and property victimization. However, it is important to note that other researchers have drawn distinctions between “general” or “pragmatic” fear (e.g., generalized anxiety) and fear of crime (see Chadee & Ng Ying, 2013).

3. The Gold Coast is located in Southeast Queensland, Australia. It is the second most populous city in the state, with approximately 537,844 residents (Australian Bureau of Statistics, 2010). The Gold Coast covers about 1,379km².

4. Demographic characteristics of the sample are somewhat consistent with the general population living on the Gold Coast. Participants within the current sample differ on the following characteristics of age, housing tenure, and marital status with reference to the general population according to data published by the Australian Bureau of Statistics (2010). The current sample has somewhat similar demographic characteristics with reference to place of birth and Aboriginal descent to the general population living on the Gold Coast.

5. We did not collect data on ethnicity because official Australian statistics to which we compare our sample to does not either. Furthermore, a very small percentage of our sample, for example, identified as Aboriginal/Torres Strait Islander (7%). Therefore, there is not enough variability in these demographic variables to do any substantial multilevel analyses or control tests. This is the same for variables such as marital status that have little variability.

6. Assessing these crime indicators separately using first-generation regression analyses (i.e., linear regressions) would limit interpretations of (Jackson’s, 2005) model dimensions and proposed factors of personal and property crime. Therefore, it is logical to test this model with more robust statistical analyses that allow for simultaneous assessment of model coefficients, such as SEM, which accounts for the degree of correlation that may be shared between personal and property crime indicators.

7. For the analyses that follow, the incivility and the cohesion measures were reverse coded to more accurately reflect the hypothesized relationships between these and other constructs in model shown in Figure 1.

8. Due to the measurement of the variables of interest to the study, factor scores were created to enable model estimation with SEM. Factor scores were created using maximum likelihood estimation (MLE) and the regression method in IBM SPSS 22.0.

9. We assess model fit before testing mean differences by gender because “as demonstrated in our review of literature” many gender-fear studies continue to use single-item measures, despite many scholars cautioning against their reliability and validity. Due to this, it is logical to first test whether these more sophisticated measures of victimization worry fit the theoretical model proposed, before comparing mean differences in gender models.

10. We also fit (Jackson’s, 2005) model of victimisation worry, irrespective of gender. The fit statistics for the personal crime model were: (GFI = .98, CFI = .93, NFI = .92, TLI = .84). Non-significant paths in this model are consequences-control and control-worry. The fit statistics for the property crime model were: (GFI = .98, CFI = .93, NFI =  .92, TLI = .83). Nonsignificant paths in this model are the same as the personal crime model. Despite some nonsignificant relationships, both full models have satisfactory to reasonable fit when assessing them irrespective of individual differences. Interested readers can refer to the following manuscript testing the reliability and validity of the model (Chataway & Hart, 2016).

11. Because Kenny, Kaniskan, and McCoach (2011) caution that indices such as the root mean square error of approximation (RMSEA) may be problematic and misleading when estimated models have small degrees of freedom, such as the current model, fit statistics exclude the RMSEA. Instead alternative indices of absolute and approximate/relative fit (e.g., GFI, CFI, NFI, IFI) are presented (Kenny et al., 2011). The CFI in this case is reported instead of the TLI, because these two indices are highly correlated. Hu and Bentler (2009) suggest that for models using MLE, the desired cutoff for the CFI be close to .95.

12. In addition to this we also examined the Standard Errors for each relationship implied by the model tested (see Figure 1). All Standard Errors were less than .06 and therefore were in acceptable ranges.

13. This is a 90% confidence interval (CI) threshold. We note the limitations of this but also caution readers relying too heavily on p values in overall interpretations of analyses. We refer readers to recent commentary in Nature around the fickle value of p values and concerns that have been raised about current research practice relying predominantly on the interpretation of this statistic (Halsey, Curran-Everett, Vowler, & Drummond, 2015).