ERIC Number: ED395964
Record Type: Non-Journal
Publication Date: 1989-Feb
Reference Count: N/A
The Kernel Method of Equating Score Distributions. Program Statistics Research Technical Report No. 89-84.
Holland, Paul W.; Thayer, Dorothy T.
A new and unified approach to test equating is described that is based on log-linear models for smoothing score distributions and on the kernel method of nonparametric density estimation. The new method contains both linear and standard equipercentile methods as special cases and can handle several important equating data collection designs. An example is used to illustrate the new method for the random groups and external anchor-test designs. The kernel method of equating, when coupled with estimated score distributions using log-linear models, has a number of advantages over other observed-score equating methods. The three phases of estimation, continuization, and equating form a unified approach to many equating problems. The kernel method contains linear and traditional equipercentile methods as special cases and can exploit the best features of both methods. An appendix presents a proof of the paper's second theorem. (Contains 7 tables, 9 figures, and 12 references.) (Author/SLD)
Publication Type: Reports - Evaluative
Education Level: N/A
Authoring Institution: Educational Testing Service, Princeton, NJ.