409 – Evaluating Alternative Imputation Methods for Economic Census Products: The Cook-Off
Implementation of Ratio Imputation and Sequential Regression Multivariate Imputation on Economic Census Products
Maria M. Garcia
U.S. Census Bureau
Darcy Steeg Morris
U.S. Census Bureau
L. Kaili Diamond
U.S. Census Bureau
The Economic Census collects general items from business establishments such as total receipts, as well as detailed items such as product data. Although product data are an essential component of the Economic Census, they vary by establishment and across trade areas, can be difficult to collect, and are characterized by low item response rates. Beginning in 2017, the Census Bureau will begin using the North American Product Classification System (NAPCS) for economy-wide product tabulations. Under NAPCS, products are no longer linked to industry, so we seek a single imputation method for all products. We present two regression models for these data: Ratio imputation and sequential regression multivariate imputation (SRMI). The ratio estimator uses a simple linear regression model with total receipts as the single predictor and product receipts as the estimated value. The SRMI method proposed by Raghunathan et al. (2001) imputes missing values consecutively by fitting a sequence of regression models to estimate product receipts conditioning on observed and imputed variables. We present the methodologies, implementation, and application to missing product data imputation.