Brzezińska, Justyna, Sagan, Adam and Rybicka, Aneta - Archives of Data Science, Series A

Article Details

Title IRT Measurement Models for Conjoint Analysis
Authors Brzezińska, Justyna, Sagan, Adam and Rybicka, Aneta
Year 2018
Volume 4(1)
Abstract Conjoint analysis and discrete choice models are widely accepted methods for preference measurement in marketing research. However, in all of these methods, the measurement of overall consumer preferences is based on binary, nominal or ordinal scales without implying any measurement model of these overall preferences. The aim of the paper is to propose Item Response Theory (IRT) latent variable models of overall preference measurement model for conjoint analysis. The model–based overall preference index (as a factor or ability scores) may be introduced into traditional conjoint analysis, instead of ordinal or choice-based preferences measured on weak scales without evidence of measure reliability. Two classes of models, Rasch-conjoint and nominal response-conjoint models, are developed and compared in the paper. The advantage of model-based preferences is to control for error of measurement and reliability (via standard error of measurement and test information function) of preference measurement model and the size of potential distortions related to preference scale unreliability and part-worth parameters bias. The comparative analysis based on the banking products described with 5 binary attributes was done on the sample of 542 respondents from 172 households in the southern part of Poland.