G-Estimation Introduction

G-estimation is a method that:

• Estimates parameters of structural nested models
• Uses the relationship between treatment and outcome to find causal effects
• Works by finding the effect that "removes" the association between A and residuals

Key idea: If we knew the true treatment effect, we could "de-confound" the outcome:

$Y - \psi A = Y^0$

This $Y^0$ should be independent of A given confounders.