NCJ Number
85369
Date Published
1981
Length
24 pages
Annotation
This study develops and applies a methodology for estimating the size of the criminal population, including those offenders not arrested, from arrest history records.
Abstract
Three parametric models were selected for testing, with all three based on the Poisson distribution. The mathematical theory used is drawn from the theory of truncated probability distributions as found in Sanathanan (1977). Samples are said to be truncated when some values of the associated random variable are not observed, as in the circumstance being considered where the sample has no information on the number of offenders not arrested. Data for the testing were drawn from the 1974-75 PROMIS public access tapes. These tapes contain the universe of adult arrests for felonies and misdemeanors for Washington, D.C. The heterogeneous Poisson model was selected as the prefered methodology. The heterogeneous Poisson fits much better than the homogeneous Poisson because of the additional parameters that can be used to change the shape of the model, so the model can be more closely tuned to the data. The results indicate that the bulk of the offender population as modeled seems to have such a low arrest rate that the probability of at least one arrest in 5 years is .63, roughly 6 chances in 10. Because the heterogeneous Poisson fits the data so well, it seemed likely that the offender population was heterogeneous and that the data represented underlying arrest heterogeneity much more than openness or behavior effects. The success of the heterogeneous Poisson model made the testing of the third model unnecessary. Twenty-eight references are provided.