Problems in Application of Latent Trait Models to Tailored Testing.
MISSOURI UNIV-COLUMBIA TAILORED TESTING RESEARCH LAB
Pagination or Media Count:
Computerized tailored testing procedures have been successfully applied in the past to the measurement of aptitude or ability. The latent trait models employed in these procedures make the basic assumption that the underlying latent trait being measured is unidimentional. However, achievement tests are commonly found to measure several factors. The purpose of the present research was to study the effects of using tailored tests for achievement measurement, knowing that the unidimensionality assumption would be violated. Of equal importance to the study was a comparison of the one- and three-parameter logistic models to each other as well as to a traditional paper-and-pencil achievement test. A total of 110 undergraduate students enrolled in an introductory educational psychology and measurement course at the University of Missouri-Columbia served as examinees for the study. A counterbalanced test-retest design was employed in which there were two separate test sessions one week apart for each examinee, with both the one- and three-parameter tests administered at each session. The results of the study indicated that neither tailored test procedure performed as well as the traditional course exam in terms of reliability. However, the three-parameter procedure had higher test information and better fit of observed responses to the model than the one-parameter procedure. Neither the one-parameter nor the three-parameter tailored tests yielded satisfactory content validity. The attitude scale results indicated generally favorable student attitudes toward tailored testing.