|
Activity Number:
|
443
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2008 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Statistics and Marketing
|
| Abstract - #302649 |
|
Title:
|
Retaining Incomplete Data Records for Market Research Estimation: CART Decision Tree Imputation Techniques for Data Sets with Missing Values
|
|
Author(s):
|
Ingo Bentrott*+
|
|
Companies:
|
University of Technology, Sydney
|
|
Address:
|
School of Marketing, Broadway , International, 2007, Australia
|
|
Keywords:
|
Missing Data ; CART ; Imputation ; Marketing ; Decision Trees
|
|
Abstract:
|
Survey nonresponse is increasing in both traditional and web based marketing surveys. Missing data will increase variance of the estimates from parametric marketing research analysis. Although parametric missing data recovery methods such as MI and EM can preserve the variance in the data, these techniques work best when the data is missing completely at random (MCAR). The MCAR assumption is rarely met in practice as there is a multivariate nature to most all marketing data sets. The problem with parametric methods of missing data imputation is exacerbated when the data set contains a mixture of data types (i.e., nominal and ratio) and the data space contains a high degree of interactivity. This paper looks at a classification tree nonparametric missing data imputation technique using CART that preserves the natural variance in the data while minimizing the bias of the imputed values.
|
