Causal Inference with Conditional Instruments Using Deep Generative Models

The instrumental variable (IV) approach is a widely used way to estimate the causal effects of treatment on an outcome interest from observational data with latent confounders. A standard IV expected be related and independent all other variables in system. However, it challenging search for directly due strict conditions. conditional (CIV) method has been proposed allow instrument conditioning set variables, allowing wider choice possible IVs enabling broader practical applications approach. Nevertheless, there not data-driven discover CIV its data. To fill this gap, paper, we propose learn representations information confounders average effect estimation. By taking advantage deep generative models, develop novel simultaneously learning representation measured generating given variables. Extensive experiments synthetic real-world datasets show that our outperforms existing methods.