|
Activity Number:
|
540
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Thursday, August 10, 2006 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Statistics and Marketing
|
| Abstract - #305635 |
|
Title:
|
Estimating Factor Effects with Data from Heterogeneous Experimental Units
|
|
Author(s):
|
Sam Weerahandi*+ and Martin Koschat
|
|
Companies:
|
Time Warner and Time Warner
|
|
Address:
|
135 W. 50th Street, New York, NY, 10020,
|
|
Keywords:
|
weights in WLSE ; repeated measures
|
|
Abstract:
|
We address a serious inference problem that arises in marketing experiments especially, a problem involving longitudinal data that has been overlooked by practitioners and researchers. The problem concerns the poor performance and bias of widely used methods of estimating parameters of linear models when the measurements from underlying experimental units could vary substantially from one unit to another. By means of a dataset from a marketing experiment, we demonstrate the drawbacks of conventional methods of making inferences and provide a simple approach to overcoming the problem. For the benefit of those practitioners who insist on conventional methods of modeling, we also will outline a Bayesian solution.
|
