NCJ Number
153368
Journal
Journal of Quantitative Criminology Volume: 10 Issue: 4 Dated: (December 1994) Pages: 361-373
Date Published
1994
Length
13 pages
Annotation
This study identifies methodological problems that have arisen in fitting nonlinear regressions to a dataset in the context of research on the age-crime relationship, based on an analysis of Britt's (1992) data.
Abstract
In an effort to determine whether the age distribution of crime in the United States has been historically invariant and is the same for different offenses, Chester Britt, III has fit the cross-sectional distribution of arrests for the seven Index offenses in the years 1952, 1957, 1962, 1967, 1972, 1977, 1982, and 1987 to two parametric forms. Earlier analyses of the historical invariance of the age-crime distribution relied on analysis of graphs (Greenberg, 1977; Hirschi and Gottfredson, 1983, 1985) or tests for change in the mean, dispersion, skewness, and kurtosis of the distribution, rather than on parametric fits. Britt found that the gamma function fit the observed distributions better than the log-normal distribution. Britt concluded that his data analysis supported the contention of Hirschi and Gottfredson that the age-crime distribution has the same historically invariant form or shape regardless of offense. The current study by the author undertook a further analysis of Britt's data in the hope of improving the fits and then extended the analysis to all the Index offenses for the 8 years Britt studied. The analysis found some problems with Britt's fitting procedures. The current study concluded that over the years 1952-57, the age distribution of involvement in crime, as measured by arrests, showed no clear historical trend for the four acquisitive Index offenses. This invariance broke down, however, for the three assaultive offenses and for other crimes as well. These results highlight problems that must be faced in conducting nonlinear regressions with standard computational packages. The author briefly discusses the value of parametric fits such as were used by Britt. 2 tables, 2 figures, and 13 references