Building a Model to Predict Which Fourth Through Eighth Graders Will Drop Out in High School

Abstract

The classification accuracies of models for predicting later high school drop out from data available in grades 4 through 8 were examined. The objective was to construct an actuarial dropout screening model to serve as part of an overall dropout identification paradigm for the state of Florida. Subjects were 503 dropouts and 282 persisters from six geographically representative school districts in Florida. Separate superior prediction models were identified for each grade level and yielded cross-validated classification accuracies ranging from 73% to 88% for the dropout group and 73% to 86% for the persister group. All classification accuracies for each grade for both the dropout and persister groups and for the combination of the two were significantly greater than chance expectation and were considered of practical importance as well.