NCJ Number
148103
Journal
Law and Human Behavior Volume: 17 Issue: 6 Dated: (December 1993) Pages: 695-700
Date Published
1993
Length
6 pages
Annotation
The accuracy of dichotomous violence predictions is discussed.
Abstract
Two major reviews of the accuracy of dichotomous (yes/no) violence predictions which reported rates of false positive errors that were calculated using conceptually and mathematically dissimilar methods are examined. Potential problems that result from analyzing data in the form of 2 X 2 contingency tables are outlined and recommendations for future research on the prediction of violence are made in the text and also are presented in table form. These recommendations include the advice that investigators who continue to report the results of studies based on dichotomous predictors should stipulate clearly what they mean by "false positive errors" and relate the number of errors to base rates. To help prevent future misinterpretations and to facilitate inspection and re- analysis by readers, journal editors should require authors to report in their manuscript the raw data for any 2 X 2 analyses. When the focus of analyses is predictive accuracy, positive predictive power, negative predictive power, and overall chance-corrected predictive accuracy statistics should be reported. References