Substantively motivated extensions of the traditional latent trait model

In this paper we advocate the use of latent trait models to make inferences about psychological construct as measured by psychological tests and questionnaires. Latent trait models have the advantage that measurement error is isolated, that items are weighted according to how well they measure the construct, and that explicit tests concerning the underlying construct are feasible. However, latent trait models come with the requirement of distributional assumptions concerning the item scores. We show in this paper that these assumptions may conflict with specific psychological phenomena. We discuss a substantively motivated latent trait model that can accommodate these phenomena.