ERIC Number: ED207998
Record Type: Non-Journal
Publication Date: 1981-Aug
Reference Count: N/A
A Comparison of Procedures for Constructing Large Item Pools.
McKinley, Robert L.; Reckase, Mark D.
This study was conducted in order to evaluate available linking techniques for forming large item pools and to make recommendations as to which techniques should be used under various circumstances. Variables of interest included calibration model and procedure, sample size, overlap level, and linking procedure. The calibration models considered were one- and three-parameter logistic models (1PL and 3PL). The MAX calibration program for 1PL and the LOGIST and ANCILLES calibration programs for 3PL were examined. Response data used for this study were for a sample of 4000 examinees from an administration of the Iowa Tests of Educational Development during the 1975-76 school year. The following conclusions were reached. For the best results an overlap of 15 items appeared to be best. At the 15 item overlap level a sample size of 2000 appeared to be necessary for stable linking of the 3PL model parameters, although when LOGIST was used 1000 seemed to be a sufficient sample size for linking item discrimination estimates. For the 3PL model the LOGIST program yielded best overall results. With a sample size of 2000 all linking procedures yielded adequate results. A sample size of 100-300 gave adequate results for the 1PL procedure. (Author/AEF)
Publication Type: Reports - Research; Reports - Evaluative; Numerical/Quantitative Data
Education Level: N/A
Sponsor: Office of Naval Research, Arlington, VA. Personnel and Training Research Programs Office.
Authoring Institution: Missouri Univ., Columbia. Tailored Testing Research Lab.
Identifiers - Assessments and Surveys: Iowa Tests of Educational Development