Horseshoes, Hand Grenades, and Treatment Effects? Reassessing Bias in Nonexperimental Estimators

Working Paper 16
Publisher: Oakland, CA: Mathematica Policy Research
Mar 30, 2013
Kenneth Fortson, Philip Gleason, Emma Kopa, and Natalya Verbitsky-Savitz
Nonexperimental methods, such as regression modeling or statistical matching, produce unbiased estimates if the underlying assumptions hold, but these assumptions are usually not testable. Most studies testing nonexperimental designs find that they fail to produce unbiased estimates, but these studies have examined weaker evaluation designs. This working paper addresses these limitations and finds the use of baseline data that are strongly predictive of the key outcome measures considerably reduces bias, but might not completely eliminate it.