Estimation of the Squared Population Cross-Validity Under Conditions of Predictor Selection

• Main Author: Kircher, Andrew J.
• Format: Others
• Published: TopSCHOLAR® 2015
• Subjects: Predictor; Criterion; Formula; Based; Applied Behavior Analysis; Psychology
• Online Access: http://digitalcommons.wku.edu/theses/1472; http://digitalcommons.wku.edu/cgi/viewcontent.cgi?article=2475&context=theses

The current study employed a Monte Carlo design to examine whether samplebased and formula-based estimates of cross-validated R2 differ in accuracy when predictor selection is and is not performed. Analyses were conducted on three datasets with 5, 10, or 15 predictors and different predictor-criterion relationships. Results demonstrated that, in most cases, a formula-based estimate of the cross-validated R2 was as accurate as a sample-based estimate. The one exception was the five predictor case wherein the formula-based estimate exhibited substantially greater bias than the estimate from a sample-based cross validation study. Thus, formula-based estimates, which have an enormous practical advantage over a two sample cross validation study, can be used in most cases without fear of greater error.