Citation

Nelson JF. Soc. Sci. Res. 1984; 13(4): 352-372.

Copyright

(Copyright © 1984, Elsevier Publishing)

DOI

10.1016/0049-089X(84)90009-7

PMID

unavailable

Abstract

A set of gamma-Poisson models are used to (1) find a parsimonious description of the univariate distributions of events recorded for several aggregates of persons and (2) interpret differences between individual rates for persons both within and across the aggregates. The selection and interpretation techniques are applied to the criminal victimization data reported in the National Crime Survey. Models were fitted to the distributions of individual victimizations reported in 16 aggregates. The most parsimonious model suggests that sex, age, and interview status were far more predictive of differences between aggregate rates than they were of differences between individual rates because the aggregate rates did not represent typical rates for individuals within any aggregate. The model selection and interpretation techniques appear to be applicable to a wide variety of problems.