Why small is better: Advancing the study of the role of behavioral contexts in crime causation

Abstract

In this chapter we argue, both from a theoretical (Situational Action Theory) and methodological(homogeneity of environmental conditions) point of view, that smallenvironmental units are preferable to large in the study of environmentaleffects on crime. Most empirical research in the field of communities and crimeutilizes fairly large spatial units of several thousand residents,such as U.S. census tracts or even clusters of census tracts, thus evoking doubts about internal homogeneity. If geographical areas are heterogeneous in their environmental conditions, associations between structural conditions, social organization, and outcomes such as crime may be clouded or rendered insignificant. On the other hand, due to common financial restrictions, choosing more units often (but not necessarily) imply fewer subjects per units which may cause a 'small number problem', that is, that the prediction of events as rare as crime will lose precision (compared to the use of larger units with more subjects). The question then is how small can you go before this potential problem outweighs the benefits of more homogeneous areas? This chapter assesses the added value of using very small area units in a community survey on environmental influences on crime. This survey was carried out in 2005 as part of the Peterborough Adolescent and Young Adult Development Study (PADS+) and covers the UK city of Peterborough and some rural surroundings. For the purpose of this study, we used the smallest administrative unit which subdivides the city, isolating 550 areas with about 300 residents each. We sampled an average of 13 respondents per unit for a total sample of 6,600 respondents. Multilevel analyses and Sampson's (1999a) ecometric approach are applied to compare the aggregate-level reliability of survey scales on this very small geographical level to the larger spatial level conventionally used for geographical analysis. The results show a considerable increase in between-neighborhood variance, reflecting a higher degree of homogeneity and statistical power for detecting particularly moderate to weak area-level effects. We use the collective efficacy scale and its subscales to illustrate these results. © 2009 Springer New York.