Infants born in areas with epidemic malaria that experience worse malarious conditions during the time in utero than the site-speciï¬c seasonal means face a higher risk of death, especially when malaria shocks hit low-exposure geographical areas, or hit mothers in the ï¬rst trimester of pregnancy.
Infants born in arid areas who experience droughts when in utero face a higher risk of death,especially if born in the so-called hungry season just after the start of the rains.
An excellent working paper from Torsten Persson and collaborators, seen at TSE this week.
A harbinger of risks to come with a more volatile global climate?
For my academic readers: Not only a spectacular dataset, but also a fantastic example of how to go beyond naive reduced form regressions to uncover something about mechanisms. A model to emulate.
One thing that I wish this and other papers might do, however: split the sample before adding structure and exploring heterogeneity.
I work mostly with small samples — I’m lucky to have a field project with more than several hundred subjects. But when you have 5,000 or 100,000 or (in this case) nearly 300,000, you can cut the sample into random halves before building your model, then test it on the other half. Why don’t economists and political scientists do so more often?
Why bother? Well, run enough functional forms and heterogeneity tests and you’re sure to find something statistically significant, whether you like it or not. Pre-specifying all your regressions is never simple, especially since one’s thinking evolves with analysis (not to mention advice from others). So splitting your sample let’s you follow your intuition (not data mine) and then test for spurious results.
In this case, with more DHS survey countries and years coming online every year, it’s an easy replication for future. A nice second year paper for prospective PhD students in the audience…