Exploring alternative screening models for identifying intern administrators

Predictive discriminant analysis model testing was used to test alternative screening models for selecting administrative interns for the School Board of Broward County, Florida. The current screening process includes 43 scores based on job dimensions that are gathered from performance ratings and evaluations of written behavioral examples. The job dimensions include the 19 Florida Principal Competencies. The 273 subjects (171 females and 102 males) included all applicants for the administrative intern program at the elementary (121), middle (81) and high (71) school levels. Minorities comprised 36.6% of the sample. Clusters of scores were examined to determine which, if any, could be eliminated without significantly reducing the classification accuracy of the model for elementary, middle and high school intern administrator candidates. McNemar's test statistic was used to compare the difference in classification accuracy between the full and various reduced models for both calibration and leave-one-out cross-validation accuracy estimates as recommended by Morris and Huberty (1991). Model performance relative to proportional and maximum chance expectations also was examined. Analyses revealed that at some school levels as many as three of the five score clusters could be eliminated without significantly (p <.05) reducing classification accuracy. These analyses, as well as parallel analyses of reduction possibilities of other selection strategies, could save significant resources devoted to collecting statistically redundant, therefore unnecessary, information. Estimates of resulting cost reductions are included.