Matt has a post up on the clear highlight of the first day of Young Lives 2013 - the final plenary by Lant "Dude is so famous he doesn’t even bother wearing a name tag" Pritchett.
Pritchett’s point was fairly simple: in many settings school can be a pretty awful place to be, especially if the curriculum is moving faster than you can keep up with it. Eventually, all but a select few are left behind, leading to a “flattening out” of the learning curve. At this point, you can’t really learn anything when you are this far behind, so why stick around? At one point – and without warning – Pritchett presented an entire slide in Spanish, to give the audience a sense of how this must feel.The bottom line is really quite depressing - there are thousands and thousands of kids out there sitting in classrooms learning absolutely nothing.
The other highlight for me was Karthik Muralidharan's plenary - apparently one of the first papers to measure and illustrate the learning progress (or lack of) of individual children as they progress through school years - on a comparable ordinal scale. The approach is smart, borrowing from the "Item Response Theory" used in GRE tests, and allows you to estimate for example whether grade 5 students can answer grade 1 questions without having to ask them. The key policy take-away was that clearly we need more of this kind of testing being done with the same kids on an annual basis. At present, we have a few snapshot surveys of learning outcomes in random years in random countries, and almost nothing in most countries that can reliably tell you something meaningful about the progress that children are making. Part of this will hopefully be solved in a few years as countries sign up to a new post-2015 development goal on learning outcomes and then realise that they have committed to figuring out a way of actually measuring them. This stuff is important. As Karthik noted, all of the RCT randomista experimental literature looks at how much an intervention impoves treatment schools compared with control schools, but misses the larger point that nobody has a clue how much progress the control schools are making over time if any (as you might expect given general economic growth).
Naureen Karachiwalla and Abhijeet Singh both presented really interesting papers, documenting in detail the role of caste in determining learning outcomes in Pakistan, and differences between public and private schools in India, respectively (bottom line: they perform similarly, but private school teachers cost around a fifth of public school teachers so private schools are a lot cheaper to run).
The nonparametric bayesian econometrics (I think that's what it is...) was maybe a bit much for me first thing this morning, but the point to note for survey designers emphasised by Costas Meghir was that the cutting edge Heckman "latent factor" model tools for estimating human capital, cognitive and noncognitive skills, or whatever you want to call it, are data hungry. You need a few (at least 3) different measures of each concept that you are trying to proxy for.
That's all for now, time to sleep.