How Item Response Theory can solve problems of ipsative data

Abstract

Multidimensional forced-choice questionnaires can reduce the impact of numerous response biases typically associated with Likert scales. However, if scored with traditional methodology these instruments produce ipsative data, which has psychometric problems, such as constrained total test score and negative average scale inter-correlation. Ipsative scores distort scale relationships and reliability estimates, and make interpretation of scores problematic. This research demonstrates how Item Response Theory (IRT) modeling may be applied to overcome these problems. A multidimensional IRT model for forced-choice questionnaires is introduced, which is suitable for use with any forced-choice instrument composed of items fitting the dominance response model, with any number of measured traits, and any block sizes (i.e. pairs, triplets, quads etc.). The proposed model is based on Thurstone's framework for comparative data. Thurstonian IRT models are normal ogive models with structured factor loadings, structured uniquenesses, and structured local dependencies. These models can be straightforwardly estimated using structural equation modeling (SEM) software Mplus. Simulation studies show how the latent traits are recovered from the comparative binary data under different conditions. The Thurstonian IRT model is also tested with real participants in both research and occupational assessment settings. It is concluded that when the recommended design guidelines are met, scores estimated from forced-choice questionnaires with the proposed methodology reproduce the latent traits well.