I think Dani Rodrik does a good job of explaning the "bifurcation" in this talk
YouTube - Dani Rodrik: Williams College Center for Development Economics 10.14.10
I agree generalization is a problem, though. One of the great things about the existence of the Poverty Action Lab, imho, is there is a nonacademic structure for repeating field experiments in other settings, helping a lot with generalizability, whereas a full professor would not want to do something that's been done before.
Dani Rodrik's weblog: Re-uniting development economics
Here's the link to Rodrik's blog post.
It baffles me why causality should be of primary importance in addressing research questions. True, RCTs and other micro-based identification approaches can answer questions with more internal validity but should that come at the price of less external validity (usually the downside of RCTs along with the GE effects) or the type of questions one should address?
You make a point of JPAL replicating experiments to address the external validity concerns but replications can not address three major problems: one that has to do with the aggregation bias from micro theory; the other that has to do with a concern that a relationship that holds at the individual level may not hold on a more macro level (say, country- level). For example the health-income relationship is a perfect example of a how the relationship between health and income can be addressed with an RCT and even multiple RCTs in multiple contexts but an RCT can never address this relationship on a macro level (does better health lead to higher economic growth, which may be the more relevant and important level to address the policy question at) which is why the micro studies generate significantly different predictions than the macro ones on this relationship. And the third concern is the ever-so-present concern of the GE effects. Again, RCTs alone can not address this GE concern. A scaled-up intervention can generate a completely different relationship between two variables than the one predicted by a great LATE estimate from an RCT.
In many aspects of development content validity (the actual measure of what we intend to measure) can also be a very serious concern.
Aut viam inveniam, aut faciam. Attending: Yale.
There are currently 1 users browsing this thread. (0 members and 1 guests)