Using Sample Data to Reduce Nonsampling Error in Unit-Level Tax Administrative Data

• Activity Number: 161
• Type: Topic Contributed
• Date/Time: Monday, August 1, 2016 : 10:30 AM to 12:20 PM
• Sponsor: Government Statistics Section
• Author(s): Tracy Haines* and Victoria Bryant and Kimberly Henry
• Companies: IRS/SOI and IRS/SOI and IRS/SOI
• Keywords: Survey sampling ; Administrative Tax Records ; Small area Estimation

Abstract

Tax policy research relies heavily on the availability of accurate and timely income-related statistics on the filing population of the United States. With computing advancements, more researchers are seeking quality income information for smaller subsets of the filing population. While the Statistics of Income (SOI) Division of the Internal Revenue Service's Individual sample has had a long history of providing quality data at the national level, estimates for small domains are subject to sampling variability. Henry et al. (2007) used area-level small area models to produce improved state-level estimates. We extend this research, using rich auxiliary data in unit-level models to reduce error in zip code-level estimates. We apply some alternative estimators to data from SOI's Tax Year 2012 sample and associated population frame data to gauge the estimates' accuracy.