Bayesian Population Interpolation and Lasso-Based Target Selection in Survey Weighting

20 Pages Posted: 13 Dec 2019

See all articles by Devin Caughey

Devin Caughey

Massachusetts Institute of Technology (MIT) - Department of Political Science

Mallory Wang

affiliation not provided to SSRN

Date Written: July 15, 2014

Abstract

We propose solutions to two important problems that have received relatively little attention in the field of survey weighting: the construction of population targets in the face of irregularly missing data, and the optimal selection of weighting targets from the set of possible auxiliary variables. Our solution to the first problem relies on a dynamic Bayesian population-interpolation model that allows subpopulation estimates in a given year to be informed by data from other years. To address the second, we formulate the problem of target selection as one of variable subset selection, for which we propose a lasso-based solution. We demonstrate the usefulness of these techniques by using them to generate weights for quota-sampled opinion polls from the early days of survey research. Given the declining response rates, rising use of non-probability samples, and growth in potential sources of auxiliary information in modern-day polling, these methods have wide potential application in contemporary survey research as well.

Suggested Citation

Caughey, Devin and Wang, Mallory, Bayesian Population Interpolation and Lasso-Based Target Selection in Survey Weighting (July 15, 2014). Available at SSRN: https://ssrn.com/abstract=3494430 or http://dx.doi.org/10.2139/ssrn.3494430

Devin Caughey (Contact Author)

Massachusetts Institute of Technology (MIT) - Department of Political Science ( email )

77 Massachusetts Ave.
E53-470
Cambridge, MA 02139-4301
United States
6173244085 (Phone)

HOME PAGE: http://www.devincaughey.com

Mallory Wang

affiliation not provided to SSRN

Do you have negative results from your research you’d like to share?

Paper statistics

Downloads
51
Abstract Views
408
Rank
693,765
PlumX Metrics