Abstract

Presenting questionnaire items in comparative formats can reduce common response sets and biases. These formats require respondents select most attractive out of two stimuli, or arrange a larger number of stimuli in order of attractiveness, either fully or partially. Until recently, basic classical scoring methods were applied to this kind of data, generally leading to scores relative to the person’s mean (ipsative scores). Recent advances in estimation methods have enabled rapid development of IRT models for forced-choice questionnaires that overcome problems of ipsative data. In this talk, a unified framework for modelling comparative data is provided. It is shown that different preference mechanisms, and different models for absolute judgement (i.e. dominance or ideal-point models) can be accomodated under this framework. Finally, the Thurstonian IRT model (Brown & Maydeu-Olivares, 2011) is described that allows flexible modelling and scoring of forced-choice questionnaires in practice.