Graphical causal models and collider bias

3 May, 2021 at 16:53 | Posted in Statistics & Econometrics | Leave a comment

causal-inference-in-statisticsWhy would two independent variables suddenly become dependent when we condition on their common effect? To answer this question, we return again to the definition of conditioning as filtering by the value of the conditioning variable. When we condition on Z, we limit our comparisons to cases in which Z takes the same value. But remember that Z depends, for its value, on X and Y. So, when comparing cases where Z takes some value, any change in value of X must be compensated for by a change in the value of Y — otherwise, the value of Z would change as well.

The reasoning behind this attribute of colliders — that conditioning on a collision node produces a dependence between the node’s parents — can be difficult to grasp at first. In the most basic situation where Z = X + Y, and X and Y are independent variables, we have the follow- ing logic: If I tell you that X = 3, you learn nothing about the potential value of Y, because the two numbers are independent. On the other hand, if I start by telling you that Z = 10, then telling you that X = 3 immediately tells you that Y must be 7. Thus, X and Y are dependent, given that Z = 10.

Students usually find this collider attribute rather perplexing. Why? My guess is the reason is most students — wrongly — think there can be no correlation without causation.

Leave a Comment »

RSS feed for comments on this post. TrackBack URI

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

Blog at
Entries and Comments feeds.