How do we adjust for the confounder of a confounder in order to compute unbiased estimates of the treatment effect of $A$ on $D$? See the causal graph (DAG) below:

confounder of a confounder

What do we call the confounder $C$ (of a confounder $B$) - does we give such variables/confounders any special name?


1 Answer 1


Thank you for including the DAG!

The answer here is pretty straight-forward: you simply condition on both $C$ and $B.$ Neither $C$ nor $B$ is part of a collider, so you're not opening up new paths by doing so, and this conditioning closes all backdoor paths from $A$ to $D,$ enabling you to get the unbiased effect of $A$ on $D.$

I'm not aware of any special name for the $C$ relative to $B$ except "parent". I'm also not sure I would call $C$ a "confounder of a confounder". A confounder is a variable that sets up a backdoor path from the cause in which you're interested to the effect in which you're interested. But you're not interested in $B$ as a cause of anything, so it strikes me as moot in this situation. I would just call $C$ the parent of $B$ and leave it at that.

  • 2
    $\begingroup$ I agree. "C" is simply another confounder, because of its residual effect even after conditioning on "B". You're right, "confounder of a confounder" is misleading because a confounder is defined by causing two things: if C caused only B and A OR if C caused only B and D, then C would not be a confounder, and adjusting for only B would satisfy the backdoor criterion. $\endgroup$
    – AdamO
    Feb 21, 2023 at 20:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.