In the context of a randomized experiment with non-compliance, I identify treatment effects without exclusion restrictions. Instead of relying on specific experimental designs, I exploit a baseline survey which is commonly available in randomized control trials. I show the identification of the average treatment effect on the treated (ATT) as well as the local average treatment effect (LATE) assuming that a baseline variable maintains similar rank orders as the control outcome. I then apply this strategy to a microcredit experiment with one-sided non-compliance to identify the ATT. In microcredit studies, a direct effect of the treatment assignment has been a threat to identification of the ATT based on an IV strategy. I find the IV estimate of log revenue for the ATT is 2.3 times larger than my preferred estimate of log revenue. R package ptse is available for this analysis.
↧