ECON 151 Lecture Notes - Lecture 27: Internal Validity, Random Assignment, Uptodate
Document Summary
Control variable has to be prior to treatment (unaffected by treatment) X are control variables unrelated to t; Typically any variable collected before random assignment are ok (baseline variables) X is not necessary as t is random. X improves the precision of the estimation. Still relevant as it measures the general effect of the policy/program/schooling. Independence of t and or the treatment is really random. Check to see if you did your work correctly. To estimate the late we just need to estimate tu: Di = 0 + 1ti + 2 x + . Again 2 x is not necessary but useful for precision. Late = itt/ 1 = reduced form/first stage = iv estimation. Itt is go-to when you have perfect compliance. D = participation (not the same as assignment) where d is the participation in the treatment. D is endogeneous and therefore the structural model is biased. Yet, t is a good instrument for d: