The modelling of the latent class structure of multiple Likert items is reviewd.
The standard latent class approach is to model the absolute Likert ratings. Commonly, an ordinal latent class model is used where the logits of the profile probabilities for each item have an adjacent category formulation (DeSantis et al., 2008). an alternative developed in this paper is to model the relative orderings, using a mixture model of the relative differences between pairs of Likert items. This produces a paired comparison adjacent category log-linear model (Dittrich et al., 2007; Francis and Dittrich, 2017), with item estimates placed on a (0,1) “worth” scale for each latent class. The two approaches are compared using data
on environmental risk from the International Social Survey Programme, and conclusions are presented.