In the honor of Halloween (today), let’s talk about the nightmare of insect swarms, composed of millions of voracious insects, devouring everything they encounter.
Short-lived shocks can have long-run consequences when they affect the level or timing of critical investments in human development . But there is still much to learn about the how and how much of these longer-run effects. Fortunately the growth of both longitudinal and geo-referenced data is abetting this search. I recently attended a workshop at the Paris School of Economics  on this general topic where many of the papers focused on natural disasters such as earthquakes, floods, and tsunamis. And there was also a paper by Philippe De Vreyer, Natahlie Guilbert, and Sandrine Mesple-Somps on the socio-economic consequences of locusts  – the first to my knowledge on this scary topic.
A swarm of desert locusts can contain millions of individual insects covering several square kilometers at one time and, as the authors note, even though they can affect wide areas of Africa, the Middle-East and West Asia, very little is known about the impact of a locust swarm. A swarm is an interesting phenomenon because it can be absolutely devastating to the farms and villages affected, eating every bit of plant crop and food, but it also leaves neighboring farms and villages untouched. The capricious destruction path is largely determined by the prevailing wind patterns of the day. It is this highly localized effect that allows the authors to identify the impacts of such an infestation.
One past complication to inferring the effects of a swarm is that the acrididae family of short-horned grasshoppers only emerges as a swarm after periods of high rainfall – a locust plague is known as “the curse of good rains” – and so its impact will be muted in any aggregate data by high overall levels of crop yields. While it has been estimated that a locust swarm only marginally affects national agricultural output , localities can be completely devastated.
The last major locust invasion in Mali occurred in 2003-2004, and before that was in 1987-1989. This second most recent outbreak is the topic of the study, and the authors leverage regional variation in this major locust outbreak to investigate its lasting impacts on the educational attainment of kids and young adults who were either primary school-age or younger at the time of the infestations. They accomplish this by matching historical geo-referenced data from FAO on locust attacks with individual education information, grouped into birth year-village cohorts, from the 1998 Malian census.
The authors essentially adopt a difference-in-difference analysis by contrasting the educational measures of cohorts before and after the swarm outbreak in affected and unaffected areas (while controlling for rainfall and other time-varying local characteristics). The coincidence of pre-crisis trends in education measures across infested and un-infested localities lends additional confidence to the identifying assumptions of the study.
It turns out that children who were on the cusp of entering primary school (around age 7) when their communities were exposed to locust invasions eventually attain less schooling by an order of one entire grade for girls and .44 grades for boys. Even if the child was in-utero or a toddler at the time of the invasion, their likelihood of ever enrolling in school 10 years later is reduced by 5 to 8 percentage points. These are large impacts for such a sudden and short-lived shock.
Now there are several challenges in this analysis, not the least of which are challenges of accurate measurement. These measurement challenges include the inability to measure migration and exact age very well through the census data, as well as the fact that the measure of locust infestation is largely self-reported by village officials to the locust observation center. Given these challenges it is all the more remarkable that any impact was found.
To address these challenges, the authors conduct robustness tests:
- Since locusts are an agricultural shock, the authors run placebo tests by investigating whether urban cohorts were affected by infestations. No impacts were found among urban children.
- The census asks about the region of birth for each individual so the authors can investigate the sensitivity of results to migration. And indeed there is some evidence of selective migration, with children, especially girls, of migrant households achieving higher levels of education. By reassigning individuals to their home regions, and then further randomly assigning to villages within regions (since home village is unknown), the analysis can generate confidence bounds for estimates that account for migration. Almost all of the main dif-n-dif estimates fall within these bounds, suggesting that selective migration plays either a small role or no role at all in the these impact estimates.
- The analysis can also control for certain supply side characteristics such as the number of village schools operating in each year and finds the same results.
Of course there are questions left unanswered by this work as there is an unfortunate lack of information on channels – are the observed outcomes due primarily to a nutritional shock that affects the cognitive capacity of young children, or a more general income shock that compels poor households either to withdraw children from school or delay their enrollment? Or are both channels in operation?
Regardless of the exact channel perhaps it’s time to give more attention to safety nets in good harvest years, targeted to locust victims.