IRT-based conjoint analysis in the optimization of banking products

Abstract

Conjoint measurement and analysis have common underlying psychometric and statistical assumption concerning axioms of additivity and two-way frame of reference in preference measurement. However, whereas the former concept is widely used in fundamental measurement of subject $\times$ object  dominance structures  as in IRT and Rasch measurement models, the latter is utilized in a broad familiy of object $\times$ object dominance structures in both compositional (i.e. Thurestone case III and V) as well as decompositional (classical conjoint experiments and BTL/alpha simulation) preference measurement models. These two traditions are rarely combined in one measurement model and research design that integrates subject $\times$ object $\times$ object measurement (Neubauer 2001). The aim of the paper is to adopt and compare three types of preference measurement models in the area of banking products in Poland:1/ paired-comparisons and rating scale conjoint experiment,2/ IRT-based conjoint (Rasch and Birnbaum politomous models) and 3/ compositional Thurstone III/V models (Bockenholt 2006). Part-worth utilities are used for product optimization and comparison across the estimated models.