01744nas a2200145 4500008003900000245009400039210006900133300001200202490000700214520125900221100001901480700002501499700002101524856005301545 2012 d00aBalancing Flexible Constraints and Measurement Precision in Computerized Adaptive Testing0 aBalancing Flexible Constraints and Measurement Precision in Comp a629-6480 v723 a
Managing test specifications—both multiple nonstatistical constraints and flexibly defined constraints—has become an important part of designing item selection procedures for computerized adaptive tests (CATs) in achievement testing. This study compared the effectiveness of three procedures: constrained CAT, flexible modified constrained CAT, and the weighted penalty model in balancing multiple flexible constraints and maximizing measurement precision in a fixed-length CAT. The study also addressed the effect of two different test lengths—25 items and 50 items—and of including or excluding the randomesque item exposure control procedure with the three methods, all of which were found effective in selecting items that met flexible test constraints when used in the item selection process for longer tests. When the randomesque method was included to control for item exposure, the weighted penalty model and the flexible modified constrained CAT models performed better than did the constrained CAT procedure in maintaining measurement precision. When no item exposure control method was used in the item selection process, no practical difference was found in the measurement precision of each balancing method.
1 aMoyer, Eric, L1 aGalindo, Jennifer, L1 aDodd, Barbara, G uhttp://epm.sagepub.com/content/72/4/629.abstract