PROBLEM OF OVERLAP IN EXPERIENCE TABLE CONSTRUCTION SUPPLEMENTAL REPORT 3

• NCJ Number: 13238
• Author(s): L T WILKINS
• Date Published: 1973
• Length: 22 pages
• Grant Number(s): NI 72-017-G; NI-71-080
• Sponsoring Agency: US Dept of Justice; US Dept of Justice NIJ Pub
• Corporate Author: National Council on Crime and Delinquency
• Sale Source: National Institute of Justice/
• Language: English
• Country: United States of America

Annotation

REVIEW OF METHODS AND PROBLEMS IN THE EFFECTIVE USE OF STATISTICAL INFORMATION BASED ON PAST BEHAVIOR TO PREDICT FUTURE PAROLE BEHAVIOR.

Abstract

MANY INFORMATIONAL ITEMS CONCERNING OFFENDERS, TAKEN FROM CASE FILES, OVERLAP AND ARE CORRELATED WITH ONE ANOTHER. FOR EXAMPLE, AUTO THIEVES TEND TO BE YOUNGER THAN OFFENDERS IN GENERAL, AND PERSONS WITH MORE PRIOR CONVICTIONS TEND TO HAVE MORE PRIOR ARRESTS AND SENTENCES. STATISTICIANS HAVE INVENTED PROCEDURES WHICH TAKE SUCH OVERLAPPING INTO ACCOUNT. WHEN THIS IS DONE IT IS TYPICALLY FOUND THAT ONLY A FEW ITEMS, APPROPRIATELY WEIGHTED, MAY BE EXPECTED TO DO THE WORK, IN PREDICTION, OF A MUCH LARGER NUMBER. VARIOUS STUDIES INDICATE THAT LESS SOPHISTICATED METHODS OF COMBINING THE INFORMATION - SUCH AS SIMPLY ADDING FAVORABLE ITEMS TOGETHER WITHOUT WEIGHTING - MAY END UP, IN PRACTICE, AS BETTER THAN THE MORE SOPHISTICATED TECHNIQUES. THIS CURIOUS RESULT SUGGESTS NOT THAT THE STATISTICAL THEORY IS WRONG BUT THAT THE NATURE OF THE DATA DOES NOT SATISFY THE ASSUMPTIONS WHICH ARE MADE IN STATISTICAL THEORY.