When I was in grad school, I saw an amazing job talk on the impacts of school construction in Indonesia. That paper, by Esther Duflo, showed long term education and earnings gains, and was one of the key factors in launching the causality revolution in micro development.
Fast forward to 2021, and new paper by Richard Akresh, Daniel Halim and Marieke Kleemans looks at the really long term impacts of that school construction.
To recap: the intervention was a massive school construction push by the government of Indonesia between 1973 and 1979. In all, about 61,800 primary schools were constructed. There was geographical variation in where schools were built and so Akresh and co. follow Duflo in using a difference and difference approach – comparing cohorts exposed to those who were not.
Akresh and co. are using a national survey (291,414 households!) from 2016, so this will not only let them look at the original beneficiaries, but the kids of those beneficiaries. And the structure of the data also lets them look at gender differences.
What do they find?
School construction means more school! For men, this is 0.27 more years, and 0.23 for women (this is pretty close to Duflo’s original estimate). Men were 2.6 percentage points more likely to complete primary school and went on to be more likely to complete lower secondary (2.3 ppt) and upper secondary (2.6 ppt). So, building primary schools gets more men through higher education too. Women are 4.1 percentage points more likely to complete primary, but with no impacts for secondary (and this is statistically different from the male estimates). There were no impacts on tertiary for either men or women.
In terms of work, the program led to a 0.6 percentage point increase in the probability that men work, and 1.1 percentage point increase in the probability that they worked in the formal sector. There are no significant impacts on the probability that women work.
The households of female beneficiaries (controlling for spouse’s exposure) show significantly higher expenditure (2.5 percent), driven by food and non-food expenditure. They also pay 6.4 percent more tax. There are no significant impacts on men’s earnings or tax payments.
On the marriage market, both men and women do better. They have significantly more educated and literate spouses. And women who were exposed to the program have significantly less children: they are 3.3 percentage points less likely to have kids aged 0-14.
What happens to the kids? A father being exposed to the school construction has no impact on the education of his children. But a mother being exposed is great for her kids -- translating into an additional .17 years of education, a 1.2 percentage point increase (significant at 10 percent) in completing lower secondary, 1.4 percentage point increase in completing upper secondary, and 0.9 percentage point in completing tertiary. So, mom may not have been more likely to go to college, but her kids are. (and in case you are wondering, effects are not different across boys and girls)
Now, to get groovy, Akresh and co. do a fairly serious cost-benefit analysis. The beneficiaries have higher expenditure, so that gives us the benefit. But the grooviness enters because Akresh and co. also looked at tax payments – so they can figure out when schools will directly pay for themselves. Starting with taxes, and a bunch of reasonable assumptions (including actually paying the teachers), Akresh and co. calculate net benefits (in 2016 USD) at $1.6bn, with an internal rate of return of 6.2 percent. Turning to the increase in household expenditure, the IRR jumps to 14.5 percent and net benefits are $37bn. And this doesn’t include the benefits for the children of beneficiaries. In fact, if you really want to geek out and play with the assumptions (or add kids to the mix), Daniel Halim has a website where you can tweak the assumptions all you want.
This is a neat paper. Some seriously long-term impacts. Fiscal sustainability. Intriguing gender differences. And a cool use of existing data that is a sequel to a cool use of existing data.
Join the Conversation