Doubly robust integration of nonprobability and probability survey data
By: Shaun R Seaman, Tommy Nyberg, Anne M Presanis
Potential Business Impact:
Combines survey data for more accurate results.
Doubly robust estimators combine an inverse probability weighting estimator and a mass imputation estimator. Several doubly robust estimators for estimating the population mean (or prevalence) of an outcome have been proposed for integrating outcome and covariate data from a nonprobability survey with covariate data from an auxiliary probability survey. However, the question of how to combine a doubly robust estimate with a corresponding estimate based on outcome data from the auxiliary probability survey alone has only received limited attention. In this paper, we (i) review previously proposed doubly robust estimators, (ii) provide formulae for the variance of doubly robust estimators and the covariance between doubly robust and probability survey estimates, (iii) propose a framework for how to combine efficiently a doubly robust estimate from a nonprobability sample with an estimate based on the auxiliary probability sample alone, and (iv) provide formulae for the variance of such combined estimates.
Similar Papers
Debiased machine learning for combining probability and non-probability survey data
Methodology
Improves survey answers using two data sources.
Why Is the Double-Robust Estimator for Causal Inference Not Doubly Robust for Variance Estimation?
Methodology
Fixes math for fairer study results.
Survey Data Integration for Distribution Function Estimation
Statistics Theory
Combines different data to get better answers.