A Comparison of Procedures for Constructing Large Item Pools

摘要

The purpose of this study was to investigate the properties of available procedures for linking item parameter estimates. The properties investigated included sample size requirements, overlap requirements, and drift. From the analyses of these properties reported, the following conclusions were reached. When large sample sizes were employed, the use of LOGIST a-values with any of the procedures appeared to yield adequate results. The same was true for the linking of b-values using the 3PL model. The best combination for linking b-values was the use of LOGIST b-values with any of the linking procedures. When small sample sizes were used for the linking of a-values the best results were obtained using the major axis and least square procedures with either the ANCILLES or LOGIST a-values. Again, the results were the same for small sample size linking of b-values. The linking results using the small sample sizes were not as satisfactory as the results obtained using large sample sizes. The small sample sizes appeared to be inadequate for both a-value and b-value linking using the 3PL model. For the linking of 1PL b-values a sample size of 100 appeared adequate. Level of overlap did not seriously affect the results of linking for any of the procedures. However, it did appear that five items probably was not adequate. For the 15 and 25 item overlaps the results were quite similar, indicating that 15 items is probably sufficient overlap.