And what does this have to do with secondary education in India? A lot, as an interesting new paper  by Karthik Muralidharan and Nishith Prakash shows. Muralidharan and Prakash take a look at the Chief Minister's Bicycle program in the Indian state of Bihar. This program provided funds to the families of all girls who enrolled in grade 9 (secondary school) to purchase a bike (and then monitored the actual purchase).
So this is point one of what makes this an interesting paper: this was a program that was implemented at scale in India's third most populous state (around 100 million people if you are keeping track at home). Of course, since everyone is eligible, this scotches a nice simple randomized experiment.
But Muralidharan and Prakash are not deterred. They start with the District Level Health Survey of 2007-8 (18 months after the start of the program). In case you were wondering what people mean by "big data" this is one example - it covers 720,000 households across 601 districts in India (of which they are using the Bihar and Jharkhand data). It gives them data on education choices, as well as things like distance to school, for everyone in the household -- and it's an already existing data source. (Interesting side note: they discovered this data while doing a referee report for another (health) paper...another good reason to do referee reports).
In the vein of some other interesting papers that use non-experimental data, they break out cohorts -- the treatment cohort is girls 14-15, while those 16-17 at the time of the survey were not eligible and hence can serve as controls. They then compare this difference to boys. However, as one might expect, a significant difference in the pre-program trend in male and female enrollment shows up when they test for it (girls’ enrollment was growing faster).
Off we go to Jharkhand, the neighboring state which, conveniently, was actually part of the same state as current day Bihar until 2001. Now the test for parallel trends looks good -- there really is no significant difference in the relative growth rates of male and female enrollment across the two states prior to the program. So Muralidharan and Prakash give us a triple difference-in-difference estimate. With minimal controls, this estimate shows a 10 percentage point impact of the bikes, but a more conservative estimate with a larger set of controls shows a 5.2 percentage point increase in enrollment for the girls who were eligible to get the bikes. What's the relative significance of this? Muralidharan and Prakash compute the base age appropriate enrollment rate of girls of this at 17.2 percent -- so this translates into a 30 percent bump in enrollment. And since boys clock in with a 30 percent age appropriate enrollment, this also represents a 40 percent drop in the boy-girl enrollment gap.
This seems like a pretty significant impact, but one might wonder about something different going in Bihar during this time (e.g. security got better relative to Jharkhand). So Muralidharan and Prakash give us a quadruple difference-in-difference. Now they segregate the sample above and below the median distance to school (3km). And for folks less than 3km away the program had no impact. For those more than 3km, the impact is estimated at 9 percentage points, suggesting the program really only worked for those far away from school. To examine this distance effect in more detail Muralidharan and Prakash show some non-parametric effects across distance. These reinforce the quadruple diff-in-diff by showing a significant positive impact between 5 and 13 km. This, plus a placebo test on eighth graders, help strengthen Muralidharan and Prakash's case overall, as well as give us some insight into what the mechanism is. (For those of you wanting to see the authors explain the results themselves, head on over to youtube and check out a video on the results through the triple diff here and the updated version here ).
So girls appear to be going to school more as a result of the bicycle program. Do they learn more? First of all, Muralidharan and Prakash cite Kremer, et. al.'s work on teacher absenteeism which shows Bihar at around 38 percent (only beaten by Jharkhand). So that should give us pause. And then Muralidharan and Prakash use administrative data on the secondary school exit examination. And they find a significant increase in the number of girls sitting for the exam, but no increase in the number passing it. So: no learning outcomes, likely due to selection and/or service delivery reasons.
Muralidharan and Prakash then stack these impacts more directly against a CCT in Pakistan (the nearest available program impacts). They show that not only are the absolute program effects larger, but that this program is more cost-effective. They advance a number of thought provoking arguments as to why this might be the case. First of all, since the household wasn't going to buy the girl a bike anyhow, this transfer went completely to the girl and stuck with her -- unlike straight cash which was likely to be shared in the household. Second, giving an in-kind, lumpy asset all at once may have avoided intrahousehold and credit constraints issues that would have limited households spending any equivalent (spread-out) cash transfer on the girl's transport. Finally, the signal of this kind of program may have had externalities in terms of getting girls to bike together (safety) and shifting norms about girls being able to leave the village to go to school. Stuff to think about and to push further as we think about cash versus kind, and the best way to close that still persistent secondary school gap between boys and girls in some countries.