Local Average Treatment Effect under Experimental Non-Compliance
RCTClinical Trial
In educational field experiments where students or teachers transfer out of assigned classrooms, unadjusted outcome comparisons measure the policy effect of assignment rather than the direct impact on classrooms that complied.
Picture this
Testing a new training routine where some assigned athletes swap sessions with others requires evaluating the specific subgroup of athletes who actually followed their assigned routine, isolating the true impact on compliant participants rather than diluting the effect across everyone who received an initial ticket.
What the evidence says
Because only 29.3% of the assigned teacher effectiveness translated into actual teacher effectiveness due to student and teacher movement, the LATE IV estimate scaled the reduced-form impact (~0.28) up to 0.955 (p < 0.01, SE = 0.123), proving that the true causal impact for compliant rosters was approximately 1:1 with predicted effectiveness.
- Who was studied
- N = 27,255 randomized students in grades 4–8 across 619 randomization blocks in 6 urban public school districts.
- How
- Instrumental Variables (IV) estimation using assigned teacher effectiveness as an instrument for actual teacher effectiveness to isolate the Local Average Treatment Effect (LATE).
What to do
Interpret instrumental variable estimates in randomized controlled trials with incomplete compliance as Local Average Treatment Effects (LATE) representing the causal impact on compliant study units.
From the source
"The instrumental variable estimator generates an estimate of the Local Average Treatment Effect (LATE). In other words, we are estimating the incremental impact of teacher effectiveness for those classrooms that complied."
Have We Identified Effective Teachers? Validating Measures of Effective Teaching Using Random Assignment