SOME INHERENT DIFFICULTIES IN THE METHOD OF PREDICTION BY CLASSIFICATION.

The article presents comments of the author on the article "Is Prediction Feasible in Social Work? An Inquiry Based Upon a Sociological Study of Parole Records," by Ernest W. Burgess. I am to discuss Burgess' paper with respect to its general methodology. A personal reference, which will disclose my standpoint of interest, will perhaps be pardoned. In a publication of 1914, I endeavored to find a basis for predicting the future political behavior of farmers and working men with respect to each other. For this purpose I developed certain indices based on the previously recorded votes of individuals of both classes in state legislatures.' In what I conceive to be their essentials, my method and that which Burgess has used in the paper of today are the same. We both have taken account of simple alternatives in behavior, an aye or no vote in one instance; conformity with parole or its violation in the other. In neither case, that is, has the behavior measured been regarded as variable. We both have employed the percentage of individuals in a general group who comply with one of the alternatives as a normal expectancy, or probability ratio by which to compare the corresponding behavior tendencies of subgroups, as expressed by similar percentages within the latter.