The Effects of Anchor Length, Test Difficulty, Population Ability Differences, Mixture of Populations and Sample Size on the Psychometric Properties of Levine Observed Score Linear Equating Method for Different Assumptions

Abstract

The Non-Equivalent groups with Anchor Test equating (NEAT) design is a widely used equating design in large scale testing that involves two groups that do not have to be of equal ability. One group P gets form X and a group of items A and the other group Q gets form Y and the same group of items A. One of the most commonly used equating methods in the NEAT design is the Levine Observed Score method for linear equating. The purpose of this study was to compare two different assumptions for the Levine Observed Score method of linear equating and to establish how accurately these two assumptions recover the true equating function. These two assumptions were compared using simulated data at synthetic population level and at sample level by manipulating anchor length, differences in ability distribution for populations P and Q, differences in test difficulty, mixture of populations and sample size. The traditional assumption outperformed the alternative assumption in conditions with larger difference in standard deviation for the ability distribution and shorter anchor length.