Reply to: Are we over-investing in baselines?
Thank you Alaka, for your thoughtful piece! I agree that we overestimate the frequency and extent to which baseline surveys play a role in policy dialogue.
And yes, a lot of baseline data goes unused, and we often lack data on indicators not yet fully known at the time of a baseline study, eg. maladaptation.
We often find ourselves rebuilding baseline data-sets retro-actively, but we more and more work with scenario-based counter-factuals in rapid impact evaluations.
Reply to: Weekly links July 27: Advances in RD, better measurement, lowering prices for poop removal, and more...
I found the Planet Money article on the sanitation cartel to be well done. They not only described the situation very well, but also documented a solution.
There's no shortage of low level corruption in the water and sanitation sector. Drinking water markets in developing countries are hardly perfect, and it sometimes affects the success of water projects. I've written about such situations on my blog
Reply to: Review of Randomistas: How Radical Researchers Changed Our World
No doubt randomized trials as a kind of "crowdsourcing" can be revealing, but they don't help much with helping you think of what to test, such as the outcome you are looking for. We all tend to be blind to our own core purposes, the same way a camera and its lens never turn up in the photos they help take. I think we could also do hypothetical tests of economic theory.
For example we might do a randomized trial of civilizations that grow their economies till their environment collapses, compared with ones that at their healthy limits scatter their profits as the systems of nature do in ecologies.
Reply to: Assessing the Severity of Experimenter Demand/Social Desirability Effects
Thanks for this interesting post and the reflections. It seems like this paper, by my colleague in Stockholm Jon de Quidt and coauthors Johannes Haushofer and Chris Roth, would also be super relevant to those interested in how to bound these effects? http://www.nber.org/papers/w23470
Reply to: Finally, a way to do easy randomization inference in Stata!
Thanks for putting this together as it is really helpful!
I have a similar question. I'm working with school data that is structured at the student x year level for ten years. I have a treatment that was introduced to a number of schools at different time points (ex: some got it in 2010, some got it in 2015), with the treatment variable as dichotomous (1 during treatment years, 0 during non-treatment years and schools that never received the treatment). I'm currently trying to run an ritest with clustering at the school level. The code looks something like:
ritest teatment _b [treatment], cluster (school): reg outcome treatment year $controls, robust cluster (school)
I keep getting the error that my treatment isn't constant within the clusters (which it shouldn't be), but am not sure how to resolve this issue. Any ideas about what could be going wrong?