A comparison of kernel equating and IRT true score equating methods

UNCG Author/Contributor (non-UNCG co-authors, if there are any, appear on document)
Kelly Elizabeth Godfrey (Creator)
The University of North Carolina at Greensboro (UNCG )
Web Site: http://library.uncg.edu/
Terry Ackerman

Abstract: "This two-part study investigates 1) the impact of loglinear model selection in pre-smoothing observed score distributions on the kernel method of test equating and 2) the differences between kernel equating, chained equipercentile equating, and true score methods of concurrent calibration and Stocking and Lord's transformation method. Data were simulated to emulate realistic situations in which test difficulty differed, sample sizes varied, anchor test lengths were of varying lengths, and test lengths ranged from 20 items to 100 items. Difficulty of anchor tests were held constant. Because data were simulated in a single group (SG) format, traditional unsmoothed equipercentile equating was used as a criterion by which all other methods, which use the non-equivalent groups with an anchor test design (NEAT), were compared. Data were simulated using IceDog (ETS, 2007) and analyzed using KE software (ETS, 2007), MULTILOG (Thissen, 2003), IceDog (ETS, 2007), PARSCALE (Muraki & Bock, 2003) and Fortran programming code developed by the author. Results indicate the impact of equating technique chosen on examinees' test scores in a variety of realistic situations, and have further recommendations for further study."--Abstract from author supplied metadata.

Additional Information

Language: English
Date: 2007
impact, loglinear model, selection, pre-smoothing, score distributions, kernel method, test equating, kernel equating, chained equipercentile equating, score methods, concurrent calibration, Stocking and Lord, transformation method
Examinations--Design and construction
Educational tests and measurements--Standards

Email this document to