Fig. 4From: Frameworks for estimating causal effects in observational settings: comparing confounder adjustment and instrumental variablesControlling for U, Z is a proxy of effectively randomized variable ABack to article page