Citizens' perceptions of police in rural US communities: a multilevel analysis of contextual, organisational and individual predictors

Abstract

There has been little research on citizens' attitudes about policing in rural communities in the USA, and existing studies do not examine the effects of community characteristics on these attitudes. We extend this work theoretically and analytically by considering the effects of community context, police organisation and individual characteristics on attitudes about police. Using data for a large sample of citizens residing in 98 small towns in Iowa, we employed multilevel ordered logistic regression techniques to model citizens' rating of police protection and degree of trust in the police. At the community level, social disorganisation was negatively associated with both outcome variables, and social integration was positively related to trust in the police. Town police departments were viewed more favourably than county sheriff's offices for both police protection and trust. Individual-level perceptions of social integration and community safety were positively related to both outcome variables. Respondents' sociodemographic characteristics had relatively few significant effects. A statistical interaction between social disorganisation and individual perceptions of social integration was observed for trust in the police, with higher levels of perceived social integration attenuating the negative effect of social disorganisation. In sum, contextual, organisational and individual predictors all had important effects on attitudes about police in this study. These findings demonstrate that theories emphasising community context are essential to a more complete understanding of crime-related attitudes in rural communities.

Keywords:

• policing
• citizens' attitudes
• rural communities
• social disorganisation

Acknowledgements

We thank Professor Terry Besser and the Rural Development Initiative at Iowa State University for sharing the ICS Data, and Chris Holmes for comments on a draft of this paper. The authors remain solely responsible for the analyses and interpretations presented herein.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1. There is good reason to think that the incorporated town is a more appropriate geographic unit than the neighbourhood (the commonly used unit in studies of urban areas) for understanding the influences of rural social disorganisation. Rural populations are more geographically dispersed than those in urban areas, and patterns of social interaction and organisation may extend well beyond the narrower neighbourhood boundaries typical of urban locales (Osgood and Chambers 2000; Roussell et al. 2013).

2. If a respondent of the pre-designated gender did not reside in the household, the survey requested that the recipient fill out the survey (Besser and Ryan 2000).

3. For agencies reporting 3–11 months of data, a weighting scheme of 12 divided by the number of months reported was used. If an agency only reported 1–2 months of offence data, then estimates were based on agencies reporting all 12 months of data that were located in the same state and population-based geographic stratum (USDOJ 2006).

4. The demographic composition of the communities selected for inclusion in the ICSs was very similar to the overall population of communities with 10,000 or less population in Iowa (Rural Development Initiative, n.d.). We also compared the structural variables for the towns in the data-set to all towns of 10,000 or fewer population in Iowa, and found no statistically significant differences. Given these observations, we are confident in the representativeness of the sample analysed in this study.

5. The reliability of the social disorganisation index was acceptable but not optimal. Therefore, we tested whether individually dropping items from or adding items to the index increased the reliability estimate. Of particular significance, we added percent non-white as a proxy for racial/ethnic heterogeneity, and we also included population density. The latter is social disorganisation variable that may have less relevance in rural (compared to urban) settings and has produced inconsistent findings in studies of rural crime (e.g., Osgood and Chambers 2000; Roussell et al. 2013). Each of the modifications to the scale resulted in appreciably lower alpha values. In addition, we factor analysed the index items and weighted them by factor scores, which also did not improve the reliability estimate. Given the consistency of findings for the index reported below, we are confident that it adequately captures the overall degree of structural disorganisation in the rural communities under study here.

6. Potentially the aggregation of social integration to the community level could be biased by sample size per community and sampling variation. This problem was ameliorated in part by the sampling design, which included 150 residents of each community in the sample. To further address community-level heterogeneity, we used several strategies. First, we controlled for a number of community-level factors identified by the literature that may be associated with attitudes about rural policing. Second, we controlled for the size of the community. Third, we used a random intercepts model, which allows for a different intercept for each community (see the analytical method section for more detail).

7. VIFs are not available for the maximum-likelihood techniques used here. The values of VIFs (calculated as 1 + [1−R²]) are not affected by the metric of the outcome variable, and the use of OLS to obtain them is statistically appropriate even though that technique may not be the best estimator for ordinal variables.

8. We estimated the models using the SAS procedure Glimmix (SAS 2006). We fitted them using a residual pseudo-likelihood (RMPL) estimator. We specified a multinomial conditional probability distribution and a cumulative logit link function. For the denominator degrees of freedom, we used the between-/within-subject specification. We employed an unstructured specification for the covariance matrix, where Proc Glimmix models a different variance component for each community.

9. The equation used to calculate the predicted values for Figure 1 was: Trust in police = −2.860 + (−.180 × Social Disorganisation) + (.097 × Social Integration) + (.002 × Social Disorganisation × Social Integration). As noted, these results were from a logistic regression of high trust in police (1 = ‘just about always’/’most of the time’; 0 = ‘some of the time’/‘hardly ever’) that included all of the variables described in the text and displayed in Table 3, Model 2. For social disorganisation, we used the mean (0 since the variable is centred) and ±1 (−2.74/2.74) SD. For social integration, we used the mean (56.78) to depict the average or ‘medium’ social integration and used the SD (10.71) to represent low social integration (−1 SD or 46.07) and high social integration (+1 SD or 67.48).