Local Average Treatment Effect (LATE) is a causal effect identified for a *local* subgroup of the population, rather than the population-wide average treatment effect (ATE), which averages the treatment effect over the entire population. What "local" means depends on the design pattern: - In the [[Design Pattern I - Instrumental Variable (IV)|IV]] design, LATE is the effect on *compliers* — subjects whose treatment status is moved by the instrument. Not everyone assigned to treatment is expected to comply with their assignment, and LATE applies only to those who do. - In the [[Design Pattern II - Regression Discontinuity (RD)|RD]] design, the analogous local estimand is the treatment effect at the *cutoff value* of the running variable — the effect for units whose covariate is at or near the threshold. The IV LATE is local in *population* (which units the effect applies to); the RD treatment effect at the cutoff is local in *covariate space* (where on the running variable the effect is identified). Both are local effects, but they apply to different subsets of the population and the inference does not generalize beyond that local region without further assumptions. > [!info]- Last updated: May 13, 2026