Back to paper
Question

What assumption is doing the most work here?

WEweizh· 5 days ago

In the potential-outcomes framing, ignorability / unconfoundedness (treatment is independent of potential outcomes given covariates) seems to carry the whole argument for observational data. Is it fair to say everything downstream — matching, propensity scores — is just machinery to make that assumption plausible?

1 Reply

Sign in to reply and react.

Accepted answer

TBtbecker5 days ago

Pretty much. Plus SUTVA (no interference, one version of treatment) quietly does a lot of work too. The methods don't test ignorability — they only help you condition on observed confounders. Unobserved confounding stays fatal.