PROBLEM OF OVERLAP IN EXPERIENCE TABLE CONSTRUCTION (FROM PROBATION, PAROLE, AND COMMUNITY CORRECTIONS. 2D ED., 1976 BY ROBERT M CARTER AND LESLIE T WILKINS - SEE NCJ-35412)

• NCJ Number: 35421
• Author(s): L T WILKINS
• Date Published: 1976
• Length: 8 pages
• Corporate Author: John Wiley and Sons
• Publication Format: Document
• Publication Type: Statistics
• Language: English
• Country: United States of America

Annotation

THE MORE POWERFUL AND EFFICIENT THE STATISTICAL PROCEDURES FOR THE ADDITION OF INFORMATION INTO A PREDICTION SCORE, THE BETTER THE SCORE FITS THE 'CONSTRUCTION' SAMPLE.

Abstract

HOWEVER, WHEN A VARIETY OF POSSIBLE METHODS ARE USED ON ONE SET OF DATA AND TESTED ON VALIDATION SAMPLES, THE LESS POWERFUL METHODS SHRINK LESS AND MAY (AND USUALLY DO) END UP IN PRACTICE BETTER THAN THE SOPHISTICATED TECHNIQUES. THE MORE POWERFUL, EFFICIENT PROCEDURES TAKE THE EFFECT OF INFORMATION 'OVERLAP' (THE CONTRIBUTORY PREDICTIVE POWER OF A SECOND OR THIRD PIECE OF INFORMATION AFTER THE POWER OF A FIRST ITEM IS TAKEN OUT) INTO CONSIDERATION. SIMPLE PREDICTION PROCEDURES MERELY GIVE DIFFERENT WEIGHTS TO ITEMS OF INFORMATION MAKING NO ALLOWANCE FOR THE OVERLAP FACTOR. THE APPARENT AND ALSO CONTRADICTORY CONCLUSION THAT INEFFICIENT STATISTICS ARE BEST FAILS TO CONSIDER THE FACT THAT STATISTICAL EFFICIENCY ASSUMES QUALITY DATA, A TERM WHICH DOES NOT DESCRIBE THE MAJORITY OF INFORMATION RECORDED REGARDING OFFENDERS IN THE PENAL SYSTEM. THEREFORE, THE CONFLICT OF STATISTICAL THEORY WITH EXPERIENCE IN THE PRACTICAL WORLD OF DECISION MAKING IN CRIMINAL JUSTICE ONLY REVEALS A FUNDAMENTAL PROBLEM OF THE QUALITY OF THE RAW MATERIAL AND THE OBVIOUS IMPORTANCE OF SAME.