A MACHINE SEARCH FOR MODERATOR VARIABLES IN MASSIVE DATA
EDUCATIONAL TESTING SERVICE PRINCETON N J
Pagination or Media Count:
Searching for moderator variables in massive data was carried out with the IBM 650. Indices to represent validities for all possible predictors based on subjects standing low, middle, and high on all possible moderator variables were computed. Predictors and criterion scores were paired after both had been trichotomized this gave 3 x 3 contingency tables from which the indices were derived. The method permitted the speed necessary for the search, but allowed distortion due to skewness in the distributions of the variables and of correlation between the moderator variable and either the predictor or the criterion. Three groups of subjects totaling 1297 students were used with 42 predictors. The indices resulted in a small number of situations where moderator variables were operating effectively. The relationship between the indices and product-moment correlations was unsatisfactory due to distortion. Either a larger machine is required or a more time-taking, expensive computational procedure is necessary. One clear moderator variable was revealed visual spatial ability interferes with the correlation between interest in English and grades in English.