NCJ Number
88812
Date Published
1982
Length
26 pages
Annotation
This paper examines the National Crime Survey (NCS) methodology and outlines several longitudinal models for victimization that can be used to produce annual prevalence rates of the numbers of housing units touched by crime.
Abstract
Based on a stratified, multistage cluster sampling plan, the NCS uses a rotating panel of household locations that provide victimization information for the preceding 6 months. This means that complete longitudinal records are not available for any specific 12-month period and that data for 6-month intervals can be missing on individuals in households because every member may not have been interviewed. Moreover, it is necessary in the NCS to distinguish among housing unit (HU), the household or family living in that unit, and the individuals who compose the household. This study used a systematic random sample of 1,539 HU's and attempted to account for missing data and weighted analysis in its models. The paper describes the following models for estimating the percentage of crime-free HU's: an ad hoc estimator, the Bureau of Justice Statistics (BJS) estimator, the homogeneous Bernoulli model of victimization, a correlated Bernoulli model, and the Markov model. Calculated estimates of victimization prevalence for each model show that the proportion of crime-free HU's appears to remain fairly constant over the 1973-75 period. This exercise indicates that missing data represent far greater problems for rotating panel surveys than has been acknowledged and that more attention should be given to modeling 'missingness' and victimization simultaneously. Appendixes provide information on BJS estimators and likelihood plots for the correlated Bernoulli and Markov models. Graphs, tables, and 16 references are included.