3
$\begingroup$

I've encountered lots of causal inference terms and jargons (under the Neyman-Rubin potential outcome framework), and I had a question regarding how Pearl's DAG restrictions relate to ignorability and unconfoundedness.

Is the "back-door" the same as ignorability?

And is the "front-door" the same as unconfundedness?

My motivation: Wikipedia page on ignorability seems to suggest that Pearl's back-door is a type of ignorability.

Pearl devised a simple graphical criterion, called back-door, that entails ignorability and identifies sets of covariates that achieve this condition

Related Questions:

Causal Inference: Moderation and Mediation

Causal Inference: Ignorability and Collider

Causal Inference: Selection Bias and Endogeneity

$\endgroup$
0

1 Answer 1

4
$\begingroup$

Ignorability, unconfoundedness, and satisfaction of the backdoor criterion mean the same thing. They all refer to there being a set of variables (possibly empty) that when conditioned on (i.e., stratified upon), the causal effect of the treatment on the outcome is nonparametrically identified as the association between them. You can think of it as all confounders being measured (i.e., no "omitted" variables), but it is a little more complicated than that, because it also involves not conditioning on the wrong variables (e.g., colliders). Satisfaction of the backdoor criterion is sufficient for nonparametric identification of the treatment effect.

The front door criterion is another condition for which satisfaction is sufficient for nonparametric identification of the treatment effect. It is almost never used because the condition is almost impossible to satisfy in real data. I encourage you not to focus on the front door criterion as you educate yourself about causal inference. It is a niche topic that serves solely to illustrate the power of causal graphs to illuminate theoretical ways to compute causal effects, but not one that is used in practice. It has little to do with ignorability or unconfoundedness.

$\endgroup$

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.