Appropriate roles for statistical decision theory and hypothesis testing in model selection: An exposition

Statistical studies in the social sciences often analyze a cross section of communities that are assumed to be homogeneous for certain characteristics. However, the researcher might be interested in investigating the general behavior of all communities as well as particular differences between groups of communities that are spatially separated. This paper suggests, and empirically examines, a procedure which utilizes both statistical decision theory and hypothesis testing and enables us to determine the best model for estimating behavioral relationships for the full sample and for subgroups. The paper examines property crime occurrences in 94 New Jersey suburban communities of Philadelphia utilizing the full sample as well as subgroups within it. The purpose of grouping is to further understand the possible interjurisdictional mobility of urban criminals. The study suggests possible factors that induce urban criminals to commit crimes in the suburbs.