Non–Fisher-Information Item Selection Criteria 1 Running head: Non–Fisher-Information Item Selection Criteria Comparison of Non–Fisher-Information Item Selection Criteria in Fixed-Length Computerized Adaptive Testing

The maximized Fisher information (MFI) criterion has been the mainstream of the item selection algorithm in many computerized adaptive test (CAT) programs because of its effectiveness and simplicity. There are still several issues with the MFI criterion, however, that need to be resolved in the field, specifically regarding estimation accuracy at the beginning of CAT administration and item pool utilization. Several non-MFI criteria (or approaches) to item selection have been developed, but practitioners still lack insight into the consequential differences among those nonMFI approaches in terms of proficiency estimation and item pool utilization. This study compared five non-MFI item selection approaches (a-stratification, interval information, likelihood weighted information, global information, and gradual maximum information ratio) and attempted to find the item selection criteria that struck a good balance between performance and efficiency in pool utilization. The study found that the gradual maximum information ratio criterion resulted in the most efficient pool use while achieving proficiency estimation as good as the MFI with various test lengths.