Definition
Plain language
A hidden third factor that makes two things look connected when one isn't actually causing the other.
As stated in the literature
A variable correlated with both the treatment and the outcome; unless controlled for by design, matching, or regression, it biases causal estimates.
Also called: confounding, confound, confounds, confounded
Why it matters: Without accounting for confounders, you can draw the wrong conclusion about what actually causes an outcome and act on a false lesson.
For example, if students who use a study app get better grades, the real cause might be that motivated students both download the app and study harder anyway.
Heard on the show
“Second concern: specification gaming versus preservation might be confounded.”Episode 001 — When AI Models Quietly Protect Each Other From Shutdown