A new paradigm for modelling ordinal responses in sample surveys

A growing interest in the current surveys is focused on human and relational issues collected as ordinal variables. Standard approaches interpret them as manifest expressions of a continuous latent variable and the current methodology is based on the relationship between the cumulative function of the ratings and the subjects’ variables. A different class of models, called CUB, is based on the statement that ordinal responses are a weighted combination of a personal feeling and an inherent uncertainty surrounding the decisional process. In this paper, the novel paradigm is presented and applied to real data sets to show the advantages of this method for analyzing big data in the context of official statistics.