The Goodness of Fit of Latent Trait Models in Attitude Measurement

The logit latent trait model offers a promising means of constructing attitude scales in sociology. The adequacy of such scales depends on the appropriateness of the model on which they are based. This article shows that the standard goodness-of-fit tests, based on chi-square distributions, are often invalid. It proposes an alternative approach based on Monte Carlo methods and residuals that has the additional merit of showing how the set of items may be refined to improve the fit and hence the quality of the scale.