Assessing the Accuracy of Estimates from a Blended Sample Where the Nonprobability Subsample Has Been Calibrated to the Probability Subsample

• Activity Number: 544 - Assessing Different Approaches to Combining Probability and Nonprobability Samples via Simulations
• Type: Topic Contributed
• Date/Time: Thursday, August 11, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Social Statistics Section
• Author(s): Jamie Ridenhour and Phillip S. Kott*
• Companies: RTI International and RTI International
• Keywords: selection model; logit function; WTADJX; selection bias; empirical bias; empirical mean squared error

Abstract

A simulated population of 113,549 (N) individuals was created from the Culture and Community in a Time of Crisis (CCTC) survey of behavior and attitudes before and during the Covid-19 crisis. Both 10,000 stratified simple random probability samples of 1,000 individuals and 10,000 vaguely-described nonprobability samples 4,000 individuals were drawn from the population. Our goal is to use the SUDAAN procedure WTADJX to estimate population means for 14 externally-chosen survey variables of interest from each of the 10,000 blended (5,000-individual) samples using information from a set of not-of-interest (NOI) CCTC survey variables (chosen by us) as well as variables for which the population means are known. Our method adds probability-sample estimates for the NOL variables to the calibration targets of the nonprobability sample. We will assess how well our method estimates the population means of the variables of interest and their standard errors.