Reliability of Factors Used in Predicting Success or Failure in Parole

Abstract

Attempts to predict the conduct of prisoners following release from penal institutions have developed quite rapidly since the appearance of Professor E. W. Burgess' study of parolees in Illinois.' In 1930 Sheldon and Eleanor Glueck2 presented a study of 500 criminal careers in Massachusetts. This was about the time that George B. Vold3 was completing a study of parolees from the Minnesota State Prison. These studies have all followed the general plan of basing predictions upon the relationship of various factors affecting the individual, such as the Nature of the Offense, Previous Criminal Record, Age, etc., to success or failure on parole. Each of these studies is evidence of an increasing interest on the part of the investigators in basing the predictions, not upon a multiplicity of factors, but rather, on a group of selected factors. Selections have been suggested or made on the basis of degree of association between particular factors and parole outcome, on interrelationship among factors, and on reliability of classification, as well as on a priori considerations. Thus Professor Burgess collected data on nearly every available social factor, omitting from final consideration two which showed virtually no relation to success or failure on parole. The present writer in a study of 3,000 parolees from the Illinois State Reformatory employed a scheme of prediction which neglected every category under the several factors which did not show a definite relationship to parole outcome.4 The Gluecks assembled data on fifty different items characterizing each individual. By means of the coefficient of contingency the several factors were ranked according to the extent of association with parole and post-parole conduct. The different types of predic-