A0830
Title: Causal inference with non-probability samples and misclassified confounder
Authors: Hua Shen - University of Calgary (Canada) [presenting]
Abstract: Causal questions often arise in social and biological sciences, but challenges occur when data come from non-probability samples and key variables are misclassified. Methods for estimating average treatment effects are presented using a non-probability sample with responses and a probability sample with auxiliary information. A binary confounder is misclassified in one or both samples, leading to biased estimates if unaddressed. To correct this, a latent-variable approach is developed via an expectation-maximization algorithm. The method is combined with a double-robust estimator, requiring correct specification of either the outcome model or the sample selection model. Simulation studies and an application to smoking data from the center for disease control and prevention demonstrate the approach's effectiveness.
