The Front Door Criterion

The one idea: The backdoor criterion needs you to measure and control for every confounder. Sometimes you can't — the confounder is unobserved, maybe unobservable. If you're lucky enough to fully observe the mechanism connecting treatment to outcome, Pearl's front-door criterion lets you recover the causal effect anyway, by chaining two comparisons through that mechanism instead of controlling for the thing you can't see.

⚠️ The trap: "There's an unmeasured confounder, so this effect just isn't identifiable — give up or go find an instrument." That feels right and it's often wrong. If you can observe every step of the mechanism through which the treatment causes the outcome, and that mechanism isn't itself confounded with the outcome, the front-door criterion identifies the effect without ever touching the unobserved confounder. It's a narrower tool than the backdoor criterion, but when it applies, it applies to exactly the cases the backdoor criterion can't touch.

The backdoor criterion lesson ended on a hard limit: if an unobserved variable U drives both treatment D and outcome Y, no control set fixes it — you need a different research design. The front-door criterion, from Judea Pearl, is one of the cleverer such designs, and it's the textbook illustration of an idea that feels almost too good to be true: you can sometimes get the causal effect of D on Y without ever measuring the confounder, as long as you can see the full pathway the treatment travels through to reach the outcome.

The classic case: does smoking cause cancer?

Pearl's own example, and worth working through exactly because it's the one every causal-inference text uses to make the…