Alternative Weight Trimming Methods to Estimate Totals in Single-Stage Sample Designs

• Activity Number: 568
• Type: Topic Contributed
• Date/Time: Wednesday, August 4, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Kimberly Henry*+
• Companies: IRS
• Keywords: administrative data ; survey weights ; simulation

Abstract

Extremely large survey weights, resulting from the inverse of the differential probabilities of selection, can often produce unreasonable point estimates of population means and totals without some form of adjustment. Considerable variation in the weights, which in some cases is dominated by a few extremely large weights, leads to less efficient estimators of totals. Trimming or truncating these weights can substantially reduce the overall variability, leading to increased precision of the survey-based estimates, at the expense of introducing bias. This study uses the Statistics of Income (SOI) Division's 2007 Form 1065 Partnership data of 3.5 million 1065 returns that were processed by the U.S. Internal Revenue Service between January and December 2008 to compare alternative model-based, design-based, and generalized design-based methods of weight trimming in simulations.