178 – Bayesian Methods in Early-Phase Clinical Trials
Calculating Standard Error Estimates on American Community Survey Data with Variables Imputed from Outside Sources
Daniel Scheer
NYC Center for Economic Opportunity
Mark Levitan
NYC Center for Economic Opportunity
Christine D'Onofrio
NYC Center for Economic Opportunity
John Krampner
NYC Center for Economic Opportunity
Todd Seidel
NYC Center for Economic Opportunity
The NYC Center for Economic Opportunity (CEO) is engaged in developing an alternative poverty measure, based on the National Academy of Sciences' (NAS) recommendations, which employs the American Community Survey (ACS). While the ACS is a rich data source for measuring pre-tax cash income, it lacks data on several important components of family resources required by the NAS proposal. In order to estimate a NAS-style poverty rate for NYC, CEO must impute these variables into the ACS from outside data sets. By incorporating outside data into the ACS, however, we introduce two additional sources of error into our estimates: (1) sampling error from the outside datasets; and (2) error from the model parameters used for imputation.
This paper describes a modification to the Census-recommended ACS variance estimator, which is designed to capture the additional error introduced through imputing variables from outside data sources. It will compare the results of this modified variance estimator with variance estimates derived from the Census-recommended approach. It will also consider the implications of this new estimator on poverty rate estimates for NYC.