PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

An Optimization Framework for Adaptive Questionnaire Design
Jacob Abernethy, Theodoros Evgeniou and Jean-Philippe Vert
(2004) Working Paper. INSEAD, France.

Abstract

We propose a general approach for adaptively designing questionnaires for conjoint analysis customized at the individual level. At each step the next question presented to an individual is designed on the fly and computationally fast based on the responses the individual has given to all previous choice questions. Our framework also encompasses recent polyhedral adaptive conjoint methods as a special case. Within our framework we develop a novel conjoint analysis method that is in the spirit of recently proposed conjoint estimation methods. We test the proposed method on widely used simulation data and compare the effectiveness of the designed questionnaires with a standard orthogonal design, a random design, and a polyhedral adaptive conjoint questionnaire under varying conditions. The results show that the proposed method leads to individual-specific questionnaires and estimations of individual utilities that are significantly more accurate than what is estimated with the other methods and questionnaires when there is high response error. We finally show that further significant improvements are achieved when we use a hybrid individual-specific and aggregate customization method that we also develop within our general framework for conjoint analysis.

PDF - PASCAL Members only - Requires Adobe Acrobat Reader or other PDF viewer.
EPrint Type:Monograph (Working Paper)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Computational, Information-Theoretic Learning with Statistics
User Modelling for Computer Human Interaction
Learning/Statistics & Optimisation
Theory & Algorithms
ID Code:826
Deposited By:Theodoros Evgeniou
Deposited On:01 January 2005