ERIC Number: ED392843
Record Type: RIE
Publication Date: 1995-Apr
Reference Count: N/A
Assuming Equal vs. Unequal Prior Probabilities of Group Membership in Discriminant Analysis: Effect on Predictive Accuracy.
Meshbane, Alice; Morris, John D.
Cross-validated classification accuracies were compared under assumptions of equal and varying degrees of unequal prior probabilities of group membership for 24 bootstrap and 48 simulated data sets. The data sets varied in sample size, number of predictors, relative group size, and degree of group separation. Total-group hit rates were used to compare the relative accuracies across six assumptions about prior probabilities. Contrary to expectations, use of population priors did not always yield the highest hit rate. When group sizes were similar, equal priors yielded greater classification accuracy than sample estimated priors. Results suggest that, when group sizes are similar, use of unequal priors may lead to a decrement in classification accuracy, even with knowledge of population priors. (Contains 5 tables and 13 references.) (Author)
Publication Type: Reports - Evaluative; Speeches/Meeting Papers
Education Level: N/A
Authoring Institution: N/A
Identifiers: Accuracy; Bootstrap Methods
Note: Paper presented at the Annual Meeting of the American Educational Research Association (San Francisco, CA, April 18-22, 1995).