A comparison of item exposure control procedures with the generalized partial credit model

textTo enhance test security of high stakes tests, it is vital to understand the way various exposure control strategies function under various IRT models. To that end the present dissertation focused on the performance of several exposure control strategies under the generalized partial credit model with an item pool of 100 and 200 items. These procedures are relatively easy to implement and have shown promise as an alternative to more complex exposure control strategies. Through unique algorithms these procedures select an item for administration from a subset of items in the item pool. The five procedures examined for efficacy were the modified within .10 logits, restricted modified within .10 logits, randomesque, restricted randomesque, and progressive restricted procedures. The modified within .10 logits, restricted modified within .10 logits, and randomesque, and restricted randomesque procedures select an item for administration from a subset of optimal items. To test the effect of the number of items available for selection in this subset, 3, 6, and 9 items were made available for selection in these procedures. Maximum information item selection was used as a base line, no exposure control, condition. The progressive restricted, restricted randomesque, and restricted modified within .10 logits procedures were found to optimally protect test security while not significantly degrading measurement precision. The restricted forms of the randomesque and modified within .10 logits procedures proved superior to their base procedures, particularly in controlling average maximum exposure rate. The incrementation of item group size in the modified within .10 logits, restricted modified within .10 logits, and randomesque, and restricted randomesque procedures demonstrated that increasing the item group size provided better test security while not significantly degrading measurement precision. Additionally, in general, the increase of the item pool size from 100 to 200 improved measurement precision and test security. Implications towards practical application are discussed and directions for future research are suggested.Educational Psycholog