Instrumental variables
Also known as: Instrumental variable estimation
Use an outside nudge that affects the cause but nothing else to untangle causation.
What it means
Instrumental variables is a strategy for estimating causal effects when treatment is confounded, by exploiting a third variable — the instrument — that influences the treatment but affects the outcome only through that treatment. A valid instrument satisfies relevance (it actually shifts the treatment) and the exclusion restriction (it has no other path to the outcome). Because the instrument induces variation in treatment that is unrelated to confounders, it recovers a causal effect — specifically a local effect among those whose treatment status the instrument moves. The approach is powerful but fragile: weak or invalid instruments can produce badly misleading estimates.
Examples
Using distance to the nearest college as an instrument for years of schooling to estimate schooling's effect on earnings.
Using a man's Vietnam draft lottery number as an instrument for military service: the lottery shifts who serves, but has no other route to his later earnings.
A jobs programme cannot force anyone to attend, so it mails invitations at random and uses receiving one as an instrument for actually turning up.
First described in Wright (1928); developed in econometrics; Angrist & Imbens (1994).