Tackling the Overabundance of Options in Survey Estimation

• Activity Number: 549 - Integration of Design and Estimation Approaches in the Use of Auxiliary Data with Sample Surveys
• Type: Topic Contributed
• Date/Time: Thursday, August 11, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Survey Research Methods Section
• Author(s): Kelly McConville* and Grayson White and Olek Wojcik and Samuel Olson and Paul Nguyen and Gretchen Moisen and Tracey Frescino
• Companies: Harvard University and RedCastle Resources, Inc. and Reed College and Reed College and Reed College and USDA Forest Service and USDA Forest Service
• Keywords: model-assisted estimation; domain estimation; small area estimation; forest inventory; generalized regression estimator; hierarchical bayes

Abstract

With the increased availability of non-survey data and the plethora of exciting, new estimators, survey practitioners may start feeling paralyzed by all these choices. How should we combine the survey and non-survey data? Should we use a model- or design-based approach? Bayesian or frequentist? Do we borrow strength outside the domain or leverage more auxiliary data within the domain? If borrowing from outside, what similar domains should we utilize? In this talk, I will tackle these questions using examples from my work with the US Forest Inventory and Analysis Program, and, in true statistical fashion, we will find that the answers vary depending on the context.