De-confounding causal inference using latent multiple-mediator pathways

Causal effect estimation from observational data is one of the essential problems in causal inference. However, most methods rely on strong assumption that all confounders are observed, which impractical and untestable real world. We develop a mediation analysis framework inferring latent confounder for debiasing both direct indirect effects. Specifically, we introduce generalized structural equation modeling incorporates structured factors to improve goodness-of-fit model observed data, deconfound mediators outcome simultaneously. One major advantage proposed it utilizes pathway structure cause via multiple debias without requiring external information confounders. In addition, flexible terms integrating powerful nonparametric prediction algorithms while retaining interpretable theory, establish identification effects based deconfounding method. Numerical experiments simulation settings normative aging study indicate approach reduces bias