Research Reports

Imputing Person Age for the 2000 Census Short Form: A Model-Based Approach

Todd R. Williams

KEY WORDS: Census Item Imputation, Multiple Regression, Variance Estimation

ABSTRACT

We present a model-based approach for imputing missing person age for the 2000 Census short form. We use a series of easy to implement multiple regression models, first proposed at the 1997 ASA Joint Meetings, to predict the missing age of a person. This paper extends our work by comparing our results with that of the hot-deck method used for the 1990 Census. By using a comprehensive set of information to determine the most important predictors of age, our modeling approach shows an improvement in finding key characteristics such as householder ages. In our paper, we also discuss a method of estimating the variance associated with replacing the missing ages with imputed ages for various demographic characteristics. These estimates of variance due to imputation are not available from past censuses.