Restricted Multinomial Regression for a Triple-System Estimation with List Dependence

• Activity Number: 509 - Methodological Innovations and Applications in Government Statistics
• Type: Contributed
• Date/Time: Wednesday, August 2, 2017 : 10:30 AM to 12:20 PM
• Sponsor: Government Statistics Section
• Author(s): Luca Sartore* and Habtamu Benecha and Kelly Toppin and Clifford Spiegelman
• Companies: National Institute of Statistical Sciences and USDA National Agricultural Statistics Service and National Agricultural Statistics Service and Texas A&M University
• Keywords: Triple-System ; Estimation ; Weights ; Capture ; Probability ; Dependence

Abstract

The National Agricultural Statistics Service (NASS) conducts the U.S.\ Census of Agriculture every five years. In 2012, NASS began using a capture-recapture approach to adjust the Census estimates for under-coverage, non-response, and misclassification. This requires two independent samples. NASS has kept its Census Mailing List (CML) independent from its area frame, which is used for the June Area Survey (JAS) every June. NASS is exploring the use of web-scraping to develop a third list-frame (TL) that would be independent of the CML and the area frame. In this paper, a Triple-System Estimation (TSE) methodology based on regularized multinomial regression is proposed to investigate for possible dependence between the CML and the TF. A simulation study is performed to compare the performance of the estimator based on the proposed methodology, which can take into account the frame dependence with others already presented in the literature.