Reliability of Attitude Scores Based on a Latent Trait Model

In this paper we present a coefficient for measuring the reliability of scores based on a fitting of a logit model with normal prior to k dichotomous items, and we consider methods of estimating the constituent terms on which that coefficient rests. We organize numerical examples around five topics: (1) the reliability of individual items, (2) reliability and goodness of fit, (3) scale differences in reliability within and between samples, (4) the influence of item slopes on reliability, and (5) variation in sample estimates of the reliability coefficient. We discuss why reliabilities differ within and between samples and present the argument for a random effects model in social research.