Loss Function Adjustment Accounting for Synthetic Bias to Evaluate Coverage Measurement for Census 2000

• Activity Number: 348
• Type: Topic Contributed
• Date/Time: Wednesday, August 14, 2002 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods*
• Author(s): Richard Griffin*+
• Affiliation(s): U.S. Census Bureau
• Keywords: Loss Function ; Synthetic Bias

Abstract

Loss function analysis for the 1990 adjustment, was generally based on the assumption that synthetic estimates included only biases due to the dual system estimator. The differential losses under comparison were only valid if the synthetic assumption held. In 1990, some evaluations of the synthetic assumption were made, using artificial populations. The approach advocated in this paper isolates the effect of synthetic bias on a loss function analysis based on the assumption of no synthetic bias. The reason for this approach is to expeditiously build on the national level loss function work. Artificial population analysis is employed only for estimates of the synthetic bias, other estimates are made from relevant samples. An unweighted squared error loss is employed, since this is the loss function analysis used in 1990. The synthetic bias that effects the loss function analysis is a complicated composite of the Census counts, the expected synthetic population counts and the true population. A simple model introducing local area variation within post-stratum is used to provide insight into the degree of synthetic bias on the loss functions which ignore synthetic bias.