|
Activity Number:
|
364
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 4, 2009 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistics and Marketing
|
| Abstract - #303874 |
|
Title:
|
A Bias-Correction Procedure for Using Aggregate Data to Proxy Individual Characteristics
|
|
Author(s):
|
Jason Duan*+
|
|
Companies:
|
The University of Texas at Austin
|
|
Address:
|
One University Station, B6700, Department of Marketing, Austin, TX, 78712,
|
|
Keywords:
|
aggregate variable ; empirical Bayes ; mixed effect model ; correlation estimation
|
|
Abstract:
|
Researchers in socioeconomic analysis, marketing and public health often face a common problem of limited data on individual characteristics. A common practice is to augment the limited individual data using the average characteristics of the group to which the individual belongs. We show that this standard practice of using group averages as a proxy for individual information leads to bias in estimation. We propose the use of group-level conditional mean characteristics given the available individual information to proxy the missing individual characteristics. In practice, group-level joint distributions of characteristics are generally unknown. We develop flexible methods to obtain group-level joint distributions using secondary data on individual characteristics. We illustrate the bias correction procedure using simulated examples and the data from a commercial bank customer data.
|
