The Predictive-Mean Method of Imputation for Preserving Coupling Between Assets and Liabilities

October 2006
Yves Thibaudeau, Alfred Gottschalck, and Thomas Palumbo

In the context of Interagency Agreement (IAA) BC-04-03 between the Social Security Administration and the U.S. Bureau of the Census, the Census Bureau researched and developed an alternative to the present imputation strategy of using a univariate hot-deck methodology to impute assets and liabilities in the Wealth Topical Module for the Survey of Income and Program Participation (SIPP). Unlike the present strategy, the alternative imputation strategy (hereafter referred to as the "alternative strategy") explicitly accounts for the correlation between assets and liabilities at the household level by supplementing the univariate hot-deck with a joint hot-deck methodology and a model-based methodology referred to as predictive-mean imputation.

By blending of both statistical-match and model-based imputation methodology, the alternative strategy represents an intermediate step toward one of the goals of the IAA: developing a purely model-based approach to imputing assets and liabilities in SIPP.