Controlling Item Exposure for Response Time-Informed Item Selection in Computerized Adaptive Testing

• In Event: Test Form Development and Item Exposure

Abstract

Inversely weighting item selection criteria in CAT by the expected response time for each
item was shown to effectively reduce the average completion time for a test of fixed length
without appreciable detriment to ability estimation. However, this also resulted in greater
distributional skew of item exposure rates. To address this issue, the current study
proposes exponentiating expected time weights and beta-partitioning items into multiple
stages of item selection, where beta is the time intensity parameter of an item assuming a
lognormal model of response time. Simulations are conducted to investigate the efficacy of
these techniques in balancing item exposure rates while preserving the reduction in
average test time.

Author

• Edison Choe, University of Illinois at Urbana-Champaign