How to Deal with Missing Categorical Data: Test of a Simple Bayesian Method

Organizational Research Methods, Vol. 6, No. 3, pp. 309-327, 2003

Posted: 20 Aug 2004

See all articles by Gongyue Chen

Gongyue Chen

University of Waterloo - Department of Management Sciences

Thomas B. Astebro

HEC Paris - Economics and Decision Sciences

Abstract

We analyze the efficiency of six missing data techniques for categorical item non-response under the assumption that data are missing at random or missing completely at random. With efficiency we mean a procedure that produces an unbiased estimate of true sample properties that, as well, is easy to implement. The investigated techniques include list-wise deletion, mode substitution, random imputation, two regression imputations and a Bayesian model-based procedure. We analyze efficiency under six experimental conditions for a survey-based data set. We find that list-wise deletion is efficient for the data analyzed. If data loss due to list-wise deletion is an issue, the analysis points to the Bayesian method. Regression imputation is also efficient, but the result is conditioned on the specific data structure and may not hold in general. Additional problems arise when using regression imputation making it less appropriate.

Keywords: Missing Categorical Data, Imputation, Bayesian conjugate analysis

JEL Classification: C12, C19, C53, C82

Suggested Citation

Chen, Gongyue and Astebro, Thomas B., How to Deal with Missing Categorical Data: Test of a Simple Bayesian Method. Organizational Research Methods, Vol. 6, No. 3, pp. 309-327, 2003. Available at SSRN: https://ssrn.com/abstract=578523

Gongyue Chen

University of Waterloo - Department of Management Sciences ( email )

Waterloo, Ontario N2L 3G1
Canada

Thomas B. Astebro (Contact Author)

HEC Paris - Economics and Decision Sciences ( email )

Jouy-en-Josas Cedex, 78351
France

HOME PAGE: http://www.hec.edu/Faculty-Research/Faculty-Directory/ASTEBRO-Thomas

Register to save articles to
your library

Register

Paper statistics

Abstract Views
1,783
PlumX Metrics