COMPARISONS OF THE KERNEL EQUATING METHOD WITH THE TRADITIONAL EQUATING METHODS ON PRAXIS™ DATA

Abstract

Kernel equating (KE) is a new approach to observed-score equating and is described in detail in von Davier, Holland, and Thayer (2004b). Over the past months, several evaluation studies of KE have been designed and carried out. In this part of the overall evaluation study, we compared the KE method with other equating methods using real data from the program of Praxis Series: Professional Assessments for Beginning Teachers®. The goal of this study was to check how closely the KE results agree with the results from other observed-score equating methods in different operational settings. The equating designs involved include an equivalent-groups (EG) design and two nonequivalent groups with anchor test (NEAT) designs with different sample sizes and different characteristics. We compared KE with the equating methods that were used operationally. The results showed that the differences between KE and the traditional equating methods are very small in the EG design, especially in the linear case. In the two NEAT designs, the KE version of poststratification equating with optimal bandwidths produced close results to its analogue, frequency estimation equipercentile equating, except at the lower score range. The KE linear method yielded very similar results to the Tucker method.