Reliability and Validity of Adaptive and Conventional Tests in a Military Recruit Population.
Abstract
A conventional verbal ability test and a Bayesian adaptive verbal ability test were compared using a variety of psychometric criteria. Tests were administered to 550 marine recruits, half of whom received two 30-item alternate forms of a conventional test and half of whom received two 30-item alternate forms a Bayesian adaptive test. Both types of tests were computer administered and were followed by a 50-item conventional verbal ability criterion test. The alternate forms of the adaptive test resulted in scores that were much more similar in means and variances than were the conventional tests for which most means and variance for various test lengths were significantly different. Analysis showed some differences in test duration between the two testing strategies; were they occurred, they were explained by the ability level of the examinees, i.e., higher ability examinees who were administered adaptive tests received more difficult items and therefore had significantly longer testing times. Combined with reduced test length for the adaptive test to obtain similar reliabilties and validities to the conventional test, however, the slight increases observed in adaptive testing time were negligible. The data support the feasibility of adaptive testing with military recruit populations and support theoretical predictions of the psychometric superiority of adaptive tests in comparison with number-correct scored conventional tests.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1983
- Accession Number
- ADA129324
Entities
People
- David J. Weiss
- James R McBride
- John T. Martin
Organizations
- University of Minnesota