Is victimization chronic? a Bayesian analysis of multinomial missing data

This paper analyses, using Bayesian methods, a simple data set concerning successive criminal victimization drawn from the National Crime Survey, to see what difference missing data makes to the conclusions.

Abstract

This paper analyses, using Bayesian methods, a simple data set concerning successive criminal victimization drawn from the National Crime Survey. As in many surveys, not all the intended interviews could be conducted. Unlike many analyses, however, this paper takes that missing data explicitly into account, to see what difference it makes to the conclusions. This leads to very substantial (and apparently novel) computational difficulties. Hence this paper is partly addressed to victimization, and partly to statistical computation. However, the latter is addressed only to the extent necessary to support the former. Whether the computational methods used here are wise general strategies for such problems is left to further investigation.