Speaker: Caleb Leedy

Title: Data Integration in Survey Sampling

Abstract:  Data integration, the process of combining information from probability and non-probability samples using common covariates, is an increasingly important area of research in survey sampling. When a sample selection model for the non-probability sample is assumed, we can employ a conditional likelihood approach to estimate the model parameters and subsequently develop an inverse probability weighted estimator. To further enhance estimation, we may incorporate an outcome regression model, leading to a doubly robust estimator of the population parameter of interest. The proposed estimation strategy is implemented using a debiased generalized entropy calibration method. We demonstrate that these model-based estimates exhibit consistency and double robustness, and yield asymptotically valid confidence intervals. The performance of the proposed methods is evaluated through several simulation studies, and we apply our approach to the Culture and Community in a Time of Crisis Survey.