Assessing Variability Due to Race Bridging

• Activity Number: 1
• Type: Invited
• Date/Time: Sunday, August 8, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Social Statistics Section
• Author(s): Nathaniel Schenker*+
• Companies: National Center for Health Statistics
• Keywords: Bayesian ; missing data ; multiple imputation ; Taylor series ; vital statistics

Abstract

Whereas census counts are traditionally treated as nonrandom population quantities in many applications, bridged census counts have random variability, because they are estimates. Techniques for assessing variability due to race bridging are discussed. Methods developed by Schafer and Schenker for inference with imputed conditional means, which can be considered a first-order approximation to multiple imputation, are adapted to the bridging problem and applied to bridged 2000 census counts as well as to selected vital rates for 2000 computed using bridged census counts as denominators. The relative standard errors of the bridged census counts by race under the 1977 standards tend to be higher for finer geographic levels and lower for coarser geographic levels. For each state (or the District of Columbia), the relative standard error of the count for a given race is no greater than .05. For birth and death rates by age group and 1977 race at the national level, on an absolute basis, use of bridged counts in the denominators does not add substantially to the relative standard errors.