Properties of a Latent Trait Reliability Coefficient

Abstract

This article takes as its point of departure a coefficient for gauging the reliability of composites based on a fitting of a logit model with normal prior to dichotomous items pertaining to social attitudes (abortion) and/or life feelings (alienation). The effect of item locations on that coefficient relative to that of item slopes is analyzed by means of simulated data. The sampling variation to which that coefficient is subject is investigated by Monte Carlo methods and by sampling theory; procedures for taking that variation into account are presented and illustrated. Discussion includes remarks on the following topics: sample size, sample design, missing responses, number of response categories, time-dependent reliability, group differences in reliability, random versus fixed effects.