|
Activity Number:
|
17
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Sunday, August 3, 2003 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistics and Marketing
|
| Abstract - #301656 |
|
Title:
|
Use Holdout Respondents to Validate Models of Customer Behavior
|
|
Author(s):
|
Terry Elrod*+
|
|
Companies:
|
University of Alberta
|
|
Address:
|
3-23 School of Business, Edmonton, AB, T6G 2R6, Canada
|
|
Keywords:
|
model validation ; choice models ; conjoint analysis ; customer behavior
|
|
Abstract:
|
It is common practice in marketing to validate a model of customer behavior by assessing its ability to predict ratings or choices for new stimuli or occasions but for the same respondents as were used in estimation. Conceptually, this model validation procedure is incorrect in most cases. Normally, our model is intended to characterize the population of potential customers from which our sample of respondents was taken. Therefore the respondents used in model validation should also differ from those used in model estimation. The consequence of using the same respondents in both estimation and validation is predictable: the performance of methods that overfit customer heterogeneity (such as individual-level estimation) will be overstated relative to those methods that do not (such as hierarchical models). This presentation describes the recommended model validation procedure and contrasts its findings to those obtained with the conventional approach. Although the presentation will utilize the important context of conjoint analysis, the procedure is applicable to virtually any type of model and data that allow for differences among respondents.
|
- The address information is for the authors that have a + after their name.
- Authors who are presenting talks have a * after their name.
Back to the full JSM 2003 program |
