Increasing test validity by considering interitem correlations

Abstract

Several investigators have proposed item-selection methods which construct a 1st-stage test consisting of the most valid items then a 2nd-stage test by adding to the 1st-stage test items which are moderately valid yet which correlate low with the 1st-stage test. Several proposed indices for selecting 2nd-stage items were compared, and some found noticeably better than others; a 3rd-stage test was noticeably better than a 2nd-stage test, but a 4th-stage test was no better than the 3rd-stage test. A method which adds several items to form each new stage was found superior to a method which adds only 1 item. The best method constructed tests substantially better on cross-validation than methods which ignore interitem correlations. (19 ref.) (PsycInfo Database Record (c) 2025 APA, all rights reserved)