Cross-validation and multinomial prediction

Abstract

The method of cross-validatory choice and assessment is applied to prediction of a multinomial indicator. The resulting predictor is compared with analogous expressions due to Good and Fienberg & Holland. Some numerical comparisons are made.