A one dimension latent trait model to infer attitude from nonresponse for nominal data

Abstract

This paper discusses the problem of missing values in attitude scales with categorical items. A simple method is proposed for obtaining information about attitude from nonresponse that is based on a latent variable model for nominal responses, The analysis treats nonresponse as a separate response category and then fits a latent variable model to the set of attitudinal items. The model coefficients provide information about the relationships between attitude and the probability of not responding to an item in the scale. Graphical methods based on response probabilities and posterior probabilities are used to reveal any relationship between attitude and missing values.