Investigation of the Performance of Multidimensional Equating Procedures for Common-Item Nonequivalent Groups Design

Abstract

In this study, the performance of the multidimensional extentions of Stocking-Lord, mean/mean, and mean/sigma equating procedures under common-item nonequivalent groups design was investigated. The performance of those three equating procedures was examined under the combination of various conditions including sample size, ability distribution, correlation between two dimensions, and percentage of anchor items in the test. Item parameter recovery was evaluated calculating RMSE (root man squared error) and BIAS values. It was found that Stocking-Lord procedure provided the smaller RMSE and BIAS values for both item discrimination and item difficulty parameter estimates across most conditions.