HIERARCHICAL LOGISTIC-REGRESSION MODELS FOR IMPUTATION OF UNRESOLVED ENUMERATION STATUS IN UNDERCOUNT ESTIMATION

In the process of collecting Post-Enumeration Survey (PES) data to eva luate census coverage, it is inevitable that there will be some indivi duals whose enumeration status (outcome in the census-PES match) remai ns unresolved even after extensive field follow-up operations. Earlier work developed a logistic regression framework for imputing the proba bility that unresolved individuals were enumerated in the census, so t hat the probability of having been enumerated is allowed to depend on covariates. The covariates may include demographic characteristics, ge ographic information, and census codes that summarize information on t he characteristics of the match (e.g., the before-follow-up match code assigned by clerks to describe the type of match between PES and cens us records). In the production of 1990 undercount estimates, the basic logistic regression model was expanded into a mixed hierarchical mode l to allow for the presence of group-specific effects, where groups ar e characterized by common before-follow-up match code. Parameter estim ates for individual match-code groups thus ''borrow strength'' across groups by making use of observed relationships between group-specific parameter estimates in the various groups and the characteristics of t he groups. This allows predictions to be made for groups for which the re are few or no resolved cases to which to fit the model. The model w as fitted by an approximate expectation-conditional-maximization (ECM) algorithm, using a large-sample approximation to the posterior distri butions of group parameters. Uncertainty in estimation of model parame ters was evaluated using a resampling procedure and became part of the evaluation of total error in PES estimates of population. Results fro m fitting the model in the 1990 Census and PES are described.