Online Program Home
My Program

Abstract Details

Activity Number: 339 - Official Statistics and Small Area Estimation
Type: Topic Contributed
Date/Time: Tuesday, July 31, 2018 : 10:30 AM to 12:20 PM
Sponsor: Survey Research Methods Section
Abstract #330550
Title: Mitigating Standard Errors of County-Level Survey Estimates When Data are Sparse
Author(s): Valbona Bejleri* and Habtamu Benecha and Andreea Erciulescu and Nathan Cruze and Balgobin Nandram
Companies: USDA National Agricultural Statistics Service and USDA National Agricultural Statistics Service and National Institute of Statistical Sciences and USDA National Agricultural Statistics Service and Worcester Polytechnic Institute
Keywords: Agricultural Survey; Bootstrap; Official Estimates; Small Area Estimation; Zero Variances
Abstract:

The USDA National Agricultural Statistics Service's (NASS's) official statistics at the county level are composites of survey and non-survey data that are manually benchmarked to state and national official estimates. NASS is currently developing Bayesian hierarchical models as an alternative to produce county official statistics using survey summaries and auxiliary data as covariates. The modeled county estimates are linear combinations of survey summaries and auxiliary data, with coefficients depending on the standard errors of direct survey estimates. With this approach, the auxiliary data are not used to produce the final model estimate when the standard error of the direct survey estimate is zero. In this paper, it is shown how to mitigate estimated standard errors of zero. The relationship between the direct survey estimates and their standard errors is modeled, if a relationship between the two is present. Exploratory data analysis is conducted and a data driven distribution-based technique using bootstrapping is proposed for cases where the relationship between estimates and their standard errors cannot be modeled well. An illustration of the method using NASS's County Agricultural Production Survey data is presented.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2018 program