Empirical and Simulation Studies of Flexilevel Ability Testing. Research Report No. 75-3.

A 40-item flexilevel test and a 40-item conventional test were compared using data obtained through (1) computer-administration of the two tests to three groups of college students, and (2) monte carlo simulation of test response patterns. Results indicated the flexilevel score distribution better reflected the underlying normal distribution of ability, and that the flexilevel test had a higher paralleled-forms reliability and a higher relationship to underlying ability level than did the conventional test. The overall test-retest stability of the two tests was equivalent, but there was evidence indicating that memory effects inflated the stability of the flexilevel test scores less than that of conventional test scores. The flexilevel provided more accurate measurement at almost all ability levels, although its information function was similar in shape to that of the conventional test. However, the interpretation of differences in the level of information provided were confounded by differences in the average discriminating power of the items in the two tests. The flexilevel test also appeared to reduce random guessing behavior in comparison to the conventional test. (Author)