Double-split cross-validation: An extension of Mosier's design, two undesirable alternatives, and some enigmatic results

Abstract

3 possible extensions of Mosier's double cross-validation (DCV) design are considered for the multiple-criterion case. 1 extension, double-split cross-validation (DSCV), is defended as preferable on grounds of unbiasedness, relative efficiency of estimation, and procedural simplicity. An empirical test of the rationale, employing self-report inventory predictors and peer-rating criterion measures for 5 personality variables, is reported. Results support the arguments in favor of the DSCV design but highlight the need for the development of a multiple-criterion extension of the theory and techniques of multiple linear regression analysis for use in studies employing this design. (PsycInfo Database Record (c) 2025 APA, all rights reserved)