(UM05-06) - Enhancing Data Quality through Longitudinal Imputation: Studies of the Income and Asset Data from HRS and AHEAD
F. Thomas Juster, Daniel H. Hill, Honggao Cao and Michael Perry
This project continues work done over the past several years on the assessment of data quality in the HRS and AHEAD surveys, eventually intended to result in a set of longitudinally consistent data files for these surveys. Much of the work is motivated by the proposition that enhancing the quality of survey data, based on changes in the design of the survey, will often introduce biases into the measurement of change over time. We argue that the longitudinal consistency of the survey data will clearly be enhanced if the imputation procedure employed, for example, to assign missing data on amounts of financial assets, or of income from those assets, is able to make use of observations or relationships that appear in prior or subsequent waves of the survey data. That is, using information that is only available in a longitudinal panel survey will make the imputed values less biased and often less noisy. The resulting set of longitudinally consistent files will be of enormous value to the scientific and policy communities.