Online Program Home
My Program

Abstract Details

Activity Number: 365 - SPEED: Innovations in Survey Sampling Designs: Administrative Data, Record Linkage, Non-Probability Samples, and More
Type: Contributed
Date/Time: Tuesday, July 31, 2018 : 10:30 AM to 11:15 AM
Sponsor: Survey Research Methods Section
Abstract #332830
Title: Variance Estimation Under Model-Implied Randomization of Nonrandom Samples
Author(s): Vladislav Beresovsky*
Companies: National Center for Health Statistics
Keywords: nonrandom samples; conditional randomization; double robustness; Taylor linearization; adjusted jackknife; hot-deck
Abstract:

The growing costs of randomized surveys along with proliferation of inexpensive datasets from web surveys stimulate interest in statistical techniques for valid inferences from nonrandom samples. A nonrandom sample contains the variable of interest and shares some covariates with an auxiliary randomized sample, which is used for adjusting information from the nonrandom sample. Using response propensity (RP) and outcome prediction (OP) models, we create conditionally randomized blocks within which data may be considered independent of its source. If both models are used for block-wise randomization, estimates are robust to misspecification of either one of the models. Point estimates may be obtained by imputation within randomization blocks, using either random hot-deck or deterministic imputation, corrected with weighted residuals from the nonrandom sample. Adjusted jackknife and Taylor linearization methods for variance estimation are formulated and tested in simulations. The proposed estimation methodology is applied to conventional random and nonrandom NHIS samples. The nonrandom sample was collected using web survey from a panel of web responders by a commercial vendor.


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

Back to the full JSM 2018 program