Open AccessTHE IMPACT OF ANCHOR TEST LENGTH ON EQUATING RESULTS IN A NONEQUIVALENT GROUPS DESIGN
Author(s) -
Ricker Kathryn L.,
Davier Alina A.
Publication year - 2007
Publication title -
ets research report series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.235
H-Index - 5
ISSN - 2330-8516
DOI - 10.1002/j.2333-8504.2007.tb02086.x
Subject(s) - equating , statistics , mathematics , mean squared error , test (biology) , paleontology , rasch model , biology
ABSTRACT This study explored the effects of external anchor test length on final equating results of several equating methods, including equipercentile (frequency estimation), chained equipercentile, kernel equating (KE) poststratification PSE with optimal bandwidths, and KE PSE linear (large bandwidths) when using the nonequivalent groups anchor test (NEAT) design. This study used pseudotests constructed of item responses from a real operational test. The equating methods were evaluated using an equating criterion. Conditional differences between the criterion scores and equated scores, and root mean square error of the differences (RMSE) were used as measures to compare the methods to the criterion equating, which in this study is an equivalent groups (EG) equipercentile equating function. The results indicate that bias tended to increase in the conversions as the anchor test length decreased, but the KE PSE with optimal bandwidths and equipercentile (frequency estimation) methods were less sensitive to this change than the other methods. The KE PSE linear method with large bandwidths performed poorly compared to the criterion across all anchor test lengths.