Proximal mediation analysis

Summary A common concern when trying to draw causal inferences from observational data is that the measured covariates are insufficiently rich account for all sources of confounding. In practice, many may only be proxies latent confounding mechanism. Recent work has shown in certain settings where standard no-unmeasured-confounding assumption fails, proxy variables can leveraged identify effects. Results currently exist total effect an intervention, but little consideration been given learning about direct or indirect pathways through a mediator variable. this work, we describe three separate proximal identification results natural and effects presence unmeasured We then develop semiparametric framework inference on effects, which leads us locally efficient, multiply robust estimators.