For DR. Adeline Zoe
Nurse1
RefereeReport_7feb2020.docxComments
This paper addresses an important topic: How do health shocks early in life affect longer-term educational attainment, and does this differ by gender? Using a health shock in Niger in 1986, the authors assess how the shock affects longer-term educational attainment, finding a significant gender gap.
There is much to like about this paper: The research question, the data, and the proposed estimation strategy. At the same time, there are several key assumptions (which are crucial for the estimation strategy) which are not necessarily supported by the data, and which seem to contradict other research in this area. This is primarily related to assumptions about the homogeneity of ethnic groups across regions and the lack of domestic migration. It would be very helpful for the authors to more substantially support these claims using micro data, or, if they are not supported, conduct robustness checks.
Introduction
· The introduction focuses on climate-induced epidemics, related to the meningitis outbreak in Niger in 1986. The link between climate change and this epidemic is unclear. Are the authors suggesting that climate change in 1986 was causing changes to the Harmattan patterns and hence the outbreak? If so, then it might be important to clarify this. If not, then perhaps try not to explicitly make this link.
Conceptual framework
· The authors speak of two channels through which meningitis could have differential effects on education attainment. For the biological channel, do the papers cited provide evidence that meningitis affects girls differently from boys from a biological perspective, although in a different context? Or are these for other diseases?
Harmattan and Diseases
· It seems as if a key element of the identification strategy is the assumption of low levels of interdistrict migration in Niger (p. 15), and state that household primarily travel from the desert regions. While the authors cite data from Afifi (2011) and the DHS data to support this, this does not necessarily correspond to a number of surveys over the course of the past 10 years on domestic and regional migration (Aker 2011, Aker 2018, World Bank 2017, etc). It would be helpful if the authors could do much more to substantiate this claim, especially using micro data, as annual population statistics can be spotty and do not necessary show within-regional variation. This is a key point, as it is key to their estimation strategy by having assignment at the district level.
Data and descriptive statistics
· The authors argue that the “Using data from Niger also allows us to exploit homogeneity in religious, ethnic, and income characteristics across individuals in the country to more cleanly capture the effect of meningitis epidemic exposure” by citing that “Niger is 98% Muslim, over 50% Hausa and most of the population is poor and employed in the agricultural sector. Source: US Department of State, CIA.” While this last sentence is technically true, it masks significant heterogeneity across Niger across the eight primary ethnic groups by region, not only from North to South but also from East to West. The border that emerged in 1906 created a Niger that included eight primary ethnic groups (Hausa, Songhai/Zarma, Toureg, Fulani, Kanuri, Arab, Toubou and Gourmantche) that were, for the most part, situated in geographically distinct regions. Niger’s ethnic groups are separated by geographic area, with the Zarma and Songhai ethnic groups concentrated to the West, the Hausa ethnic groups in between Konni and Zinder and the Kanuri ethnic groups farther East, and the Fulani ethnic groups interspersed. These regions inhabited by these ethnic groups have been relatively stable over time; ethnographic maps from the late 1950s show a similar pattern of settlement as those from 2008 (Asiwaju 1985, Miles 2005, Aker et al 2014). It would be helpful for the authors to dig more deeply into differences in regional and district composition to better substantiate this claim.
· Since the authors are looking at the impact of meningitis on education (classifying it as low and high exposure), it would be helpful to show how different the high and low areas were prior to the meningitis outbreak (similar to a balance table). If DHS data aren’t available from the 1980s, then using time-invariant characteristics would be helpful.
· For the correlation between Harmattan winds and meninigitis cases: While Figure 5 is compelling, it only compares two seasons. Did previous work on this by Perez Garcia Pando et al (2014) look at these patterns across several years? In reviewing their paper, it seems as if they have data from 1986 to 2008, but it isn’t clear that they looked at the specific correlations between these factors (wind, dust, etc) and meningitis timing (although I could have read this too quickly). It would be helpful to know if Perez Garcia Pando et al documented this pattern more widely across a number of years, and hence your “check” in Figure 5 is primarily used to check the validity of these patterns for the specific years in question.
Identification Strategy
· In the first sentence of the identification strategy, the authors state that “the intuition for our identification strategy is that in an environment that is homogeneous with limited interdistrict migration, households in districts affected by the 1986 meningitis induced by Harmattan, experience the epidemic as either an income or direct health shock, while households in the other districts remained unaffected.” As mentioned above, the authors would need to do more to substantial this, given that some other evidence from Niger suggests that there is heterogeneity by ethnic group across districts, as well as non-trivial migration. For the latter point, if the authors could, at a minimum, show migration data from 1986, this would be helpful. If that isn’t possible, then they could show a balance test, showing that exposure to meningitis is not correlated with ethnicity or migration (migration could have been an out-response to meningitis, but pre-1986 levels would be helpful).
· While the authors’ key question is looking at the differential impact of meningitis on educational attainment, a natural question is whether this affected overall educational attainment. Did the authors look at this in their previous 2017 paper? If so, it would be helpful to mention this again in the estimation strategy.
· How was MENIN intensity defined and developed? What is the R0 for meningitis? Does the number of people infected per 100,000 matter so much as some threshold level where it become an epidemic?
Results
· Figure 4: Are the authors using a K-S test to test these?
· Table 2: While the results are strong and negatively statistically significant for girls, they seem small in magnitude – ie, out of approximately 1.5 (or 2) years of education, this reduction is not 1% of the control mean (if I am interpreting this correctly). Is this important to point out and discuss?
· The manuscript says “Central African Francs” (CFA) – this should be the West African CFA Franc, is this correct?
· Perhaps I am missing something, but Table 7 is difficult to read and interpret. Figure 6 suggests that meningitis exposure affects age of first marriage, but then the table shows no s.s. results (although the t-statistic seems to be above 2 in some cases) and there would not be a s.s. difference between boys and girls if a F-test was done. And why would Table 8 not assess this for males and females as well, rather than just the female sample?
