Preliminary Investigation of Variance Issues Related to Generalized Regression Estimation Used for American Community Survey Five-Year Estimates

• Activity Number: 690
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Richard Griffin*
• Companies: U.S. Census Bureau
• Keywords: Generalized Regression Model ; Successive Difference Variance Estimator

Abstract

The American Community Survey (ACS) multiyear housing unit (HU) weighting process includes a model-assisted step that uses a generalized regression model (GREG) to calibrate the HU weights using covariates derived from administrative records. The Census Bureau discovered that the variance estimates of some 5-year estimates at the national level may be larger than the corresponding 1-year estimates. The ACS standard error estimates are calculated using the successive difference variance estimator, which measures both sampling error and ratio estimation bias. One initial tool considered is calculating an alternative variance estimator targeting only sampling error. Some evidence exists that using this alternative variance estimator provides a possible explanation for this concern. This paper evaluates this concern using 2006-2010 ACS 5-year data.