再犯予測に於けるカテゴリー数量化の比較

抄録

The aims of this study are to find out significant factors of the repetition of an offense when we predict it concerning the delinquent juveniles under the probation, to compare empirically the variation in the accuracy of the prediction which are due to the different techniques in the quantification of categories of those factors, and thereafter, to get acquainted with the useful technique for making the prediction table.
The surveyed factors include not only those before they came under probation such as the life history, the personality, the environments, but also those observed, recorded by the probation officers during their probation. 136 successful cases and 59 cases a second offense are the subjects of this survey.
x² and critical ratio technique are used for the selection of the significant factors and category. The seized significant factors are fifteen in number, and the significant categories involved in those factors were twenty two in number. Four quantification techinques are compared with one another: (1) the technique which Mr. Hayashi used for parole prediction,(2) the technique employed by Dr. Glueck to investigate the delinquent juveniles,(3) the technique which gives-1 to each of the categories which are closely related to a second offense, and gives +1 to those related to a success,(4) the way to apply Dr. R. J. Wherry's formula of weighting to the percentage of a second offense and a success, concerning each of those twenty two categories.
When the score of each of those 195 subjects is calculated according to the abovementioned techniques, the accuracy of prediction can be theoretically calculated as follows: the technique (1) has an accuracy of 88%, the technique (2) 74%, the technique (3) 80%, and the technique (4) 80%. The difference between the technique (1) and the other three techniques is significant (P=0.05), and there can be found no significant differnce between the other three techniques.
The technique (1) quantifies the categories byweighting both the factor and category. Moreover, it uses such a way of weighting that will maximize, with each of the factors, the difference of the mean in both groups,“Success” and “Second Offense”. In case of the technique (4), the figures attached to each category is optional. The technique (2) and (3) are less valid because the weight given to the categories is decided only by the percentage of “Success” or “Second Offense”. It is presumable that the difference in the way of weighting categories is the main reason why we get different accuracy percentage among the four techniques.
The distribution of scores of both groups,“Success” and “Second Offense”, which is on the prediction table, shows us such accuracy percentages as follows: t.(1) 84%, t.(2) 79%, t.(3) 82%, t.(4) 80%. This means that there is no significant difference among the four techniques, and that the figures do not coincide with the theoretically calculated results mentioned above. The latter is due to the fact, I presume, that the distribution of scores of both groups does not show the normal probability curve as the surveyed subjects are rather few in number. In order to make a strict comparison of the accuracy percentage, it will be necessary to apply the quantified categories to the new samples which have not been used in their quantification.