NCJ Number
196622
Journal
Criminology Volume: 40 Issue: 3 Dated: August 2002 Pages: 711-736
Date Published
August 2002
Length
26 pages
Annotation
The potential of nonlinearity to account for changes in the United States’ homicide time series is presented in this article.
Abstract
While temporal patterns in United States’ homicide data are most often explained using linear processes to account for variation, this study presents a test of nonlinear processes to account for homicide series dynamics. After presenting the historical fluctuation of United States homicide rates and counts, the author discusses the nature of linear and nonlinear time series process models. In order to test the possibility that nonlinearity accounts for dynamics in homicide rates, the author analyzed two homicide rate series from the National Vital Statistics Systems. The first series presented annual homicide rates from 1925 to 1998, and the second series were a monthly homicide rates from 1952 to 1996. Applying Keenan, Tsay, and Luukkonen’s, White’s Artificial Neural Network, Brock, Dechert, and Scheinkman’s, and ARCH/GARCH tests to the dataset provided little evidence of a nonlinear structure in regards to dynamics in homicide rates. The author maintains that changes in the homicide time series appear to be generated by and accounted for by linear processes. While supporting other, previous research, the author concludes that this study raises the question of why homicide trends follow a linear time path. Figures, tables references, appendix