Using SRMI for Nonresponse Adjustments in IRS' Taxpayer Compliance Studies

• Activity Number: 483
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Karen Masken+ and Wei Liu* and Getaneh Yismaw
• Companies: IRS and IRS and IRS
• Keywords: Missing at random ; Nonresponse bias ; Multiple imputation ; Sequential regression ; Complex survey

Abstract

The Internal Revenue Service has begun conducting annual studies of individual taxpayer compliance. For these studies a stratified random sample of taxpayers is selected and audited. However, not all audits are completed for a variety of reasons (i.e. the taxpayer was not locatable or did not show up for the audit) resulting in non-response. Earlier work (Masken, JSM 2004) argued for treating non-response in these studies as missing at random, but did not come to a conclusion on the exact method that should be implemented. Continuing that effort, we first explore various weighting and imputation techniques that we considered and then detail the sequential regression multivariate imputation (SRMI) method that we implemented. Finally, we expect to demonstrate that the nonresponse bias is well adjusted and the standard errors of the estimates are reduced.