A0573
Title: A mixture model for ordinal variables measured on semantic differential scales
Authors: Marica Manisera - Universita' degli Studi di Brescia (Italy) [presenting]
Paola Zuccolotto - University of Brescia (Italy)
Abstract: In surveys aimed at measuring subjective perceptions toward latent traits, questions asking to rate opinions on ordered response scales are usually exploited. The most popular ordered response scale is the Likert-type, but semantic differential scales (SMD) are also commonly used. In SMD, the respondent is asked to rate his/her position between two opposite adjectives. We present the CUM model, a mixture of a linearly transformed Multinomial and a Uniform random variable, suited to fit rating data expressed on SMD. The formulation of the model, belonging to the CUB class, is derived from specific hypotheses about the decision process in the mind of the respondent, who is assumed to start their reasoning from the center of the SMD and move upward/downward according to two specific feeling parameters, measuring the attitudes toward the two opposite adjectives. A specific representation of the parameter space is proposed with a triangular plot. Parameter estimation is carried out via the EM algorithm and a case study is examined, also with the comparison to the results obtained using other models of the CUB class.