"Placebo tests" and the meaning of "urban" in agrarian countries

Jed Friedman at Development Impact summarizes a new paper by De Vreyer, Guilbert, and Mesple-Somps that measures the effect of locust swarms on human capital formation in Mali. I very much like the idea behind this paper – early childhood health and nutrition shocks strike me as an increasingly important sub-field of development economics, and the literature on them motivates my own predominant focus on health. The authors find big effects on educational attainment: young boys who experience a locust swarm and the attendant local agricultural catastrophe end up finishing half a year less school than they would have otherwise, and for girls the effect is a full year. Their approach is a difference-in-difference method, which means they compare the changes in educational attainment in locust-affected regions with changes in other areas.

It’s an intuitively plausible approach, but it’s tricky to get right: in a simple “diff-in-diff” you need to assume “parallel trends”, which is that the rate of increase in educational attainment is the same across the locust-hit areas and the unaffected areas. When you can control for other factors, then you need to have parallel trends conditional on those observed, controlled-for factors. The authors have some nice data on school infrastructure that allows them to control for one potential concern: what if more schools were built in unaffected areas, maybe because the government invested more heavily in the places with the best payoff? This turns out not to affect their results.

Another check people usually do when they run a diff-in-diff is a “placebo test”. For example, if you have lots of years of data, then you can look for effects on affected areas the year before the actual event occurs. This is what Jensen and Oster do in their paper on Cable TV and women’s empowerment in India. De Vreyer, Guilbert, and Mesple-Somps do something different: they want to confirm that this is an effect caused by the locusts, and that it is a direct effect and not a general equilibrium or macroeconomic impact. To do this they look at impacts on kids in affected cities. I am hesitant about two aspects of this approach. First, if there are macroeconomic effects I’d expect them to impact the entire country, and not to be particularly concentrated in locust-affected regions. So it’s not clear that we learn anything about macroeconomic effects (or spillovers, as I’d call them) from this.

Second, the more compelling aspect of this is to use the urban areas as a validity check: if this is an effect of the locusts, and not some other correlated phenomenon, then it should impact rural areas alone (barring spillovers). And indeed that’s what they find – there are no effects on children from urban locations. But I’m a bit dubious of this distinction for a typical heavily-agrarian developing country. First, the authors identify 340 “cities” in their data, and the use of scare quotes here is their choice but I agree with it. Mali has an average personal income of $1,127 in purchasing-power-parity dollars, which puts it just ahead of Malawi in terms of economic development. I don’t know a ton about the Malian economy, but Wikipedia also tells me that agriculture is the dominant sector, which stands to reason given its level of development. I would venture that most of those 340 cities are fairly small trading posts (some are evidently neighborhoods of the geographically-larger urban locales, although Bamako, the capital, is excluded from the paper). In that case, I would bet a decent sum of money that these, too, are fairly rural areas in the sense of the importance of agriculture. Just go to a trading center in sub-Saharan Africa and walk around, looking for crops.

Even if these are “true” cities, one thing I’ve noticed during my time in Malawi is that under the surface, almost everyone is still a farmer. In informal conversations, I often try to get at where people’s livelihoods come from, and many Malawians who live and work in cities still maintain farms, either on the outskirts or farther afield. So I’m not sure I’d expect a zero impact from locusts in urban areas, and I’d definitely expect some spillovers. This speaks to one of the big advantages of modern development economics’ focus on fieldwork: there is a lot you can learn just by going to the place you are studying and talking to people. Perhaps this is obvious, but as best I can tell it’s something that economists still rarely do. Could Mali be different, such that the assumption that urban areas aren’t engaged in agriculture is valid? Indeed it could – and one of the best ways to find out is to go there.

Leave a Reply

Your email address will not be published. Required fields are marked *