I like this style of post and would love to see more of it on the forum. An interesting reanalysis.
I agree that this study is not remotely closing the book on whether educational interventions should be prioritized. Not least because even a significant effect on wages wouldn’t rule out the possibility of negative spillovers—control students lose out because of labor market competition—which would of course matter a lot for policy implications.
I know it’s very ancillary to the post, but I’m not a fan of the last paragraph. The findings of natural experiments are less robust because the world is a random place. Tons of important policies can’t be studied through RCTs, and it feels wrong to have an outlook that throws out the signal from natural experiments just because there’s more noise relative to RCTs. RCTs have their own problems, too—for example, an education scholarship RCT could give us clean estimates of the returns to education but it wouldn’t be able to tell us the effects of a large-scale educational policy like Inpres, which will have problems of scaling, a different population, etc.
Hi Karthik. Without belaboring shades of emphasis, I basically agree with you. But you know, I’ve just spent thousands of words criticizing someone’s work and I want to end positively, within reason.
I like this style of post and would love to see more of it on the forum. An interesting reanalysis.
I agree that this study is not remotely closing the book on whether educational interventions should be prioritized. Not least because even a significant effect on wages wouldn’t rule out the possibility of negative spillovers—control students lose out because of labor market competition—which would of course matter a lot for policy implications.
I know it’s very ancillary to the post, but I’m not a fan of the last paragraph. The findings of natural experiments are less robust because the world is a random place. Tons of important policies can’t be studied through RCTs, and it feels wrong to have an outlook that throws out the signal from natural experiments just because there’s more noise relative to RCTs. RCTs have their own problems, too—for example, an education scholarship RCT could give us clean estimates of the returns to education but it wouldn’t be able to tell us the effects of a large-scale educational policy like Inpres, which will have problems of scaling, a different population, etc.
Hi Karthik. Without belaboring shades of emphasis, I basically agree with you. But you know, I’ve just spent thousands of words criticizing someone’s work and I want to end positively, within reason.