Wharton-SMU Research Center

In-House Seminar

Guest Speaker:

Teck H. Ho
Associate Professor of Marketing, The Wharton School, University of Pennsylvania

Topic:

A Learning-based Model for Imputing Missing Levels in Partial Conjoint Profiles

Venue:

Business Block, Level 2, Seminar Room 6
Singapore Management University
469 Bukit Timah Road, Singapore 259756

Date:

Wednesday, 31 July 2002, at 10.30am

Reservation:

This seminar is free. Places are limited. Please confirm your attendance by Tuesday, 30 July 2002, 12 noon with Ms. Lim Lih Yeng at lylim@smu.edu.sg or telephone: 6822-0197.

About the Seminar:

Rating and pair-wise comparison are the two most popular preference elucidation methods for designing in conjoint analysis. A drawback, however, of these methods is that respondents often find the resulting task lengthy and cognitively difficult when they are asked to rate or choose complex product profiles with many attributes. One solution is to present respondents with partial product profiles with one or more missing attributes.

This research has two goals. First, we model how respondents infer missing information about product attributes in a partial conjoint profile by developing a learning-based imputation model that nests several extant models as special cases. The advantage of our approach over previous research is that the imputation model infers missing levels of an attribute not only from prior levels of the same attribute levels of the current product profile). A second goal is to use the proposed model to examine whether the part-worth utilities derived using various degrees of missingness in partial profiles are systematically different than those obtained from full profile and self-explicated data (standard benchmarks).

Our empirical results yield four important findings. (1) Our general model is superior to current extant models (in terms of various fit criteria), such as the averaging and the recency models. (2) We further demonstrate that individuals do focus on attributes other than the one under consideration when imputing missing attribute levels. (3) Our model results show that the relative importance of an attribute could change depending on whether one or two attributes are missing in the conjoint profiles. (4) Via a prior manipulation on the correlation between two attributes, we are able to significantly manipulate the degree to which one attribute is used for imputation when the other is missing.