I use this space to publish thoughts on a range of issues that are too complex for twitter and too short or trivial for academic publication.
Some post-publication reflections on the “educational gender Kuznets curve”
June 2023 (latest minor updates: 6 September 2023)
Educational gender Kuznets….what?
In August 2020, Joerg Baten, Felix Meier zu Selhausen, Elisabeth Kempter and I published an AEHN working paper in which we trace gender gaps in educational attainment (years of education) in 21 African countries over most of the twentieth century and coin a new term: the “educational gender Kuznets curve” (EGKC). We define the EGKC as an initial growing gender gap in years of schooling as male education expands, followed by a subsequently decline in the gender gap. We observe this pattern in many (but not all) African countries, as well as the largest developing world regions across the twentieth century. The curve is evident when we backward-project census data (obtained from IPUMS), and when we use the Barro and Lee dataset on educational attainment. Since this initial publication, the article has been published in the Population and Development Review in September 2021, and has been summarized and discussed on various blogs and Twitter threads. On Google Scholar, the article has so far received 39 citations (per 31 May 2023). This sounds like a decent reception, but how well has the EGKC – which I consider to be our key finding and contribution – been understood and picked up by colleagues?
Not so well, I’m afraid. To see what elements of our study have found their way in the literature, I checked every single reference so far to both the working paper and journal article. I was able to trace 38 citing publications, and found 49 instances in which “Baten et al. 2020” or “Baten et al. 2021” were used to back up specific statements. As a disclaimer, let me emphasize that I am not criticizing any of the citing studies here. My intention is not to blame any specific author(s) for misciting our work, and I am of course grateful with their engagement with our work in the first place! My goal is merely to trace and reflect on the reception of the article so far. Another disclaimer is that this exercise only looks at academic papers: some blogs and Twitter threads have given more attention to the ECGK (many kudos to you, Dr Alice Evans!). In the table below I summarize the different uses of our study in academic literature:
Table 1. Analysis of the reasons "Baten et al." is cited
Source: my own analysis based on reading and classifying citations found through Google Scholar
Notably, the most important reason to cite “Baten et al.” so far has been to substantiate the claim that Africa is performing badly in terms of gender inequality today. Sometimes, citing studies link this to specific “persistent historical” or “African cultural” factors to explain such bad performance. That we are mainly cited for this reason is odd – I would in fact say incorrect. Firstly, our data, deriving from national censuses digitized and made available through IPUMS International, takes us up only to cohorts born in the 1980s (and thus educated, largely, before the year 2000). This means that we have little to say or contribute in terms of knowledge about trends in African educational gender inequality in recent years (we do flag current gender disparities in the introduction, but solely based on secondary literature).
Secondly, one of our key conclusions is that educational gender inequality is not static, but that it evolves with the overall expansion of education: at first, boys accumulate a large share of all educational expansion which leads to a rising absolute gender gap in attainment. At some point, then, girls begin to account for a growing share of accumulated education, which ultimately leads to a reduction of the absolute gap as well, which typically peaks around 4 years of male education on average. One of our findings which I find most interesting and important, is that Africa actually has a remarkably low educational gender gap across all stages of the ECGK, as shown in Figure 1a below. This is robust to using the attainment ratio (M/F) instead of the attainment gap (M-F), as shown in Figure 1b below. This means that cultural factors or historical events particular to Africa are unlikely candidate explanations for gender-skewed educational outcomes. Instead, it is the comparatively low expansion of male education in most African countries that in and of itself can fully account for why educational gender inequality remains high in Africa today, in a comparative perspective. So far, none of the academic references to our work have picked up on this insight.
Figure 1a. Gender gaps and male educational expansion in developing world regions, birth cohorts 1890-1980
Source: Baten et al. 2021, p. 831
Figure 1b. Gender ratios and male educational expansion in developing world regions, birth cohorts 1890-1980
Source: Baten et al. 2021, Appendix (Supplemental Materials)
Note: years indicate the center-year of the birth cohorts, so 1890 are people born between 1885 and 1894.
What do(n’t) we know about educational expansion versus gender gaps?
There remains much to be said about the relationship between educational expansion and educational gender gaps. Firstly, it is not obvious if we should focus on absolute gender gaps in years of schooling, or on gender ratios, which indicate the share of total education going to boys versus girls. As we briefly do in the article (and especially in the working paper version), a case can be made for both approaches: it is insightful to know that the proportion of all education going to girls was increasing slightly during the colonial period (ratio), but it is equally useful to see that, as a result, boys were able to accumulate many more years on education (gap), increasing their advantage over girls in knowledge and skills attained. Indeed, the broader literature is quite inconclusive on which of these two metrics is more appropriate to measure gender inequality in education. This is one reason why we evaluate relevant correlates of gender inequality for both the gap and ratio – although the ratio results are relegated to the appendix. The most important correlates – especially the fact that gender gaps are smaller in places treated with early and large-enough historical missions to appear in a widely cited 1924 missionary atlas – are statistically significant in both gap and ratio specifications.
Secondly, our concept of the ECGK is a-historical, in the sense that we generalize a patterned relationship to exist across societies and time periods (as in Kuznets' original curve, which is also supposedly universally applicable across times and places). In practice, it’s perfectly plausible that tolerance towards gender inequality in general and education in particular has declined over time globally as a result of progressive influence of feminist ideology, as well as the efforts of the United Nations and subsidiary organizations. Acknowledging this means taking “history” properly into account. If this is the case, we should expect lower gender gaps along African countries’ ECGK’s (i.e. at similar levels of male educational expansion that in past societies was associated with very large gender gaps), since they are relative latecomers to educational expansion. Or to put it differently, Africa’s better performance of closing gender gaps at given levels of male education (Figure 1) may be explained by the fact that these levels were attained in a (more recent) historical context more prone to educational gender equality.
Although this is not entirely straightforward to test statistically, we can run a ‘horserace’ type exercise to see if the ECGK indeed predicts evolving gender gaps better, or if historical time does. And if we account for both, is there any “Africa” effect remaining? Table 2 below shows these results, in different specifications, at the country-level, using 10-year birth cohorts, taken from the latest version (v.3) of Barro & Lee (2013), from which we include people born between 1885 and 1984. There are reasons to be hesitant about the accuracy of the dataset, especially the earlier cohorts, but here I take it at face value. The table shows the results for the gender gap and ratio, but I prefer the ratio for this exercise, because it is more straightforward and avoids having to enter a squared term in the regression. These are tentative statistics, but they all point in one direction: male education strongly correlates with the gender gap in a pattern that follows the ECGK. Meanwhile, “historical time” also significantly correlates with the gender gap in most specifications (but not for the ratio when we include fixed effects). These two forces operate in conjunction, since they tend both to be significant at the same time. If we control for these relationships, a country being in Africa (which taken as the reference category below) does not make much of a statistically significant difference in comparison to being in other regions with similar histories of colonialism and educational expansion (Middle East and North Africa, South Asia and Southeast Asia). In fact in most ratio specifications, MENA and SA countries come out as significantly more unequal than African countries, on average. This corroborates the message of Figures 1a and 1b above, taken from Baten et al. (2021).
Table 2. Correlates of the gender gap and ratio at the national level in 66 developing countries, birth decades 1890-1980
Source: Version 3.0 of the Barro-Lee dataset (Barro & Lee 2013), downloaded from http://www.barrolee.com/. The underlying .do file can be provided on request.
Note: years indicate the center-year of the birth cohorts, so 1890 are people born between 1885 and 1894.
Why should we care about the EGKC?
I believe that the policy implications of the EGKC are substantial. Firstly, it shows that there are no grounds on which African countries can be singled out for particularly "bad" gender dynamics, arising from culture or otherwise, working against the education of girls. Surely, norms, practices, and policies exist that disadvantage girls, but the data show no reason to maintain that these have been worse in sub-Saharan Africa over the 20th century than in other comparable regions. Second, the data suggests that ‘gender-neutral’ investments in education are probably a good strategy to reduce gender gaps in the medium- to long-term. As male education expands, the absolute gender gap tends to rise before it declines, but the relative gap declines more or less steadily with the expansion of male education. This does not mean that policies targeting girls specifically do not work or are undesirable, but the relationship between gender gaps and educational expansion is so strong, that we should at least consider the possibility that overall education expansion, even if it mainly benefits boys at first, has a larger benign impact on girls’ education in the medium- to long-term than gender-specific policies.
Many issues remain to be explored to understand the EGKC better, however. Why some countries and regions have more pronounced curves than others, and why some even have no female disadvantage as education expands (Botswana!) remains a puzzle to me as well. The gender division of labor, exposure to external influences, religion and national policy are all plausible factors – some of which we, in “Baten et al.” have found to be robustly correlated with the magnitude of the gender gap – but our findings in this respect are tentative at best and need further research. The temporal dynamics also need further study. What are the factors that shift educational investment towards girls, eventually leading to a downward sloping EGKC as absolute attainment gender inequality begins to decline? Is this an educational “saturation effect” among boys (or in other words: “boys first, girls next”)? Is it an emancipation effect, for example when more educated fathers begin to realize the benefits of female education? Is there some labor market (i.e. skilled labor demand) effect that happens to coincide with educational expansion? Does it perhaps have more to do with gender-segregated education and the building of school infrastructure for girls? Of course some of these questions have been addressed in the literature before, but not within the broader framework of and following from the insights generated by the EGKC. Thus, I hope our article will be cited more – not just because it is about "educational gender gaps in Africa" but for its actual contribution -- because I believe it helps advance knowledge on an important topic.
References
Barro, R. J., & Lee, J. W. (2013). A new data set of educational attainment in the world, 1950–2010. Journal of development economics, 104, 184-198.
Baten, J., De Haas, M., Kempter, E., & Meier zu Selhausen, F. (2021). Educational gender inequality in Sub‐Saharan Africa: a long‐term perspective. Population and Development Review, 47(3), 813-849.
Baten, J., De Haas, M., Kempter, E., & Meier zu Selhausen, F. (2020). Educational gender inequality in Sub‐Saharan Africa: a long‐term perspective. African Economic History Network Working Paper No. 54
Why we should be cautious to claim that ‘smallholders’ produce only a third of the global food supply
August 2021 (Updated October 2021)
Our World in Data (OWID) is one of my favorite places on the internet. The platform, founded and headed by Max Roser, recently massively expanded its reach as one of the key disseminators of up-to-date and easily accessible information on Covid-19 globally. Before the pandemic, it was already widely known for its visualizations of global poverty trends over the long-run, which have been widely shared and hotly (sometimes viciously) debated on various social media platforms. But OWID has a lot more to offer: easily accessible time-series on topics ranging from climate change to health, and from violence to economic output. Increasingly, the website also offers informative interpretative articles by its editors, based on data and state-of-the art academic research. Thus, OWID does exactly what I, as an economic historian, believe in: disseminating (long-run) data to inform public debates, weighing and comparing evidence and busting pervasive myths.
In this light, I was surprised by a tweet by Hannah Ritchie, Our World in Data’s head of research. In this tweet, she summarizes the conclusion of a recent article posted on OWID on the contribution of farms of different size classes to global food production. Not 80%, as the UN Food and Agriculture Organization and others have claimed, but only “one-third” of global food is produced by “smallholders”, according to Ritchie, who bases herself on a number of studies that underpin the article. This huge difference arises from a confusion of the terms “smallholder” and “family farmer”: the former is – by definition small, while the latter can operate huge, high-tech farms of 100 hectares or more. There is also a normative issue at stake here. The 80% is often used by detractors of large-scale, high-input farming. But, as Ritchie retorts, if only a third of global food is produced by smallholders, we should put some more faith in agricultural modernization and technology, and stop romanticizing the poverty that characterizes small-scale farming. Another nice example of useful, data-founded OWID myth busting, it seems! But is it really?
A quick look at the article shows that it uses (and admits to use) a very restrictive definition of what a ‘smallholder’ is: a holding with less than 2 hectares under crops. When I (and some others) responded to Ritchie’s tweet asking if this is a reasonable definition, she gave two responses: 1) FAO uses this definition itself, so we can take it to task for falsely pushing the “80%” estimate, and 2) there is a wide consensus in the literature about the 2 hectare threshold (the OWID article even speaks of a “universal standard”. The OWID article itself suggests a third argument: 3) 84 percent of the worlds’ 570 million farms are under 2 hectare. Let us scrutinize each of these arguments.
I find 1) to be a fair challenge. FAO is an official UN body, and we should expect it to use consistent definitions, and communicate clearly on important topics within its mandate. However, it is also a fact that FAO has many different branches and divisions, and harbors a plethora of perspectives and views (as an aside: it also seems to struggle to maintain some of its basic operations: its GAEZ platform was down for several years, at the detriment of researchers around the world). As a working paper shows, even within FAO there has recently been discussion about the appropriate definition of a ‘smallholding’ based on land size (FAO 2017: 12-15).
However, argument 2) makes me frown. My knowledge of smallholder agriculture is largely confined to 20th century Africa, and I’m certainly not an agricultural scientist. However, I also do not have to be one to know that numerous authoritative scholars use very different definitions of what a ‘smallholder’ is. There is certainly no ‘widespread consensus’ on a ‘universal’ understanding of smallholders <2 hectares. Just a cursory glance at the literature – “relatively large and well-equipped smallholder farmers with 5–10 ha of land” in Jayne, Mather and Mghenyi (2014:1389), or “around 80% of the smallholder farms in SSA are now smaller than 2 ha” in Frelat et al. (2016:461) – serves to prove this point.
Argument 3) also fails to convince me. Yes, 84 percent is a lot, but that such a large share of farms is so small should not come as a surprise. In many rural areas population pressures are mounting, rural households diversify their income earning activities beyond farming, and young generations aspire or (partly) pursue off-farm careers. There is also a composition effect: the number of farmers rapidly declines in countries where farms scale up – from a majority of the workforce to a mere few percent.
So, we should conclude that the justifications for opting for a generic 2 hectare threshold to define and discuss smallholder agriculture are not convincing. Next, we should ask: is there something inherently wrong with this definition, or does it just need better justification? As noted above, it is worth pointing out, as Ritchie does, that a very large (and increasing) share of farmers across the world cultivate small (or even tiny) amounts of land. That these ‘farmers’ are struggling and poor rather than robust providers of the global food supply is also worth emphasizing. But why should we exclude those with slightly larger holdings of, let’s say, 2 to 5 hectares from the definition of what a ‘smallholder’ is? How should we think of the contribution of these farmers to global food production? Indeed, the graph provided in the OWID article reveals that farms of 2 to 5 hectares contribute to global food production beyond expectation: they have very high land productivity – substantially higher than smaller or larger farm size classes. This makes sense. In fact, it is these families, who have either devoted more resources to agriculture (rather than off-farm pursuits) or have proven more successful in its pursuit, that we might expect to produce surpluses beyond mere family survival. Surely, if we include all struggling households and exclude many specialized and successful farmers in our definition, we will – unsurprisingly, and indeed almost by definition – arrive at a pessimistic estimation of smallholders’ contribution to global food production. This doesn’t help the debate.
It is also reasonable to question the use of farm sizes altogether to compare farms across the globe. Just for some perspective: farm sizes certainly do not have the same universal properties as the ‘1,25 [or any other amount] dollar a day’ poverty lines at purchasing power, which are really supposed to compare living conditions at barebones level no matter the context (although experts of living standards know that even that standards has many caveats). In comparison, the meaning of acreages are highly context specific, depending on the intensity of cultivation and the agro-ecological context. A grain-farming family in West Africa’s interior savannas, for example, will require substantially more cultivated land to attain the same (caloric) output than its root-crop growing counterpart in the forest zones to its south. Farm sizes in Burkina Faso (3.9 hectares in 1993) and Côte d’Ivoire (3.9 hectares in 2001) might be similar, but that doesn’t mean they yield similar monetary and caloric returns (farm size data from Jayne Chamberlin and Headey 2014:4). In other words: farm size, uncorrected for yields and yield potential, is a poor indicator for global comparison. A very recent paper by Giller et al. (2021) on African farming puts this beyond doubt: “Our results also show that using a standard, area-based farm size definition (e.g. smaller than 2 ha) is not a robust way to define a smallholder farm. In dry areas a substantial proportion of such farms is larger than 2 ha, while they are clearly small farms in economic terms. In high agroecological potential areas with high population densities, a farm of close to 2 ha can already be considered to be a large farm, with the majority of smallholder farms having an area less than 1 ha.”
To conclude, Ritchie poses a strong twofold claim, in the summary, headline and social media advertising of OWID's post on smallholders and the global food supply. First, she claims that smallholders produce 70 or 80% of global food supply are false and unsubstantiated. Second, she posits that the actual number is “one-third”. Part 1 of this claim is myth busting well done. No matter how one defines a ‘smallholder’, one will not reasonably arrive at 70 or 80 percent. The share of global food supply is certainly less. What got me to write this response is Part 2 of the claim, which in fact introduces a new myth into the debate. Certainly, the procedure by which Ritchie arrives at the “one-third” claim is entirely transparent, and indeed founded on the work of others. However, it is also based on a false understanding on a purported but non-existent scholarly consensus on the definition of a smallholder. Thus, if one views its task as fact-based knowledge dissemination (which I’m pretty confident OWID and Ritchie do), claims about smallholders’ contribution to the global food supply should not be made with so much confidence, but attenuated to reflect the fuzziness of definitions and the current limitations of scholarship on the issue which, for example, has only marginally attempted to factor ecological variation into its definition of who is a smallholder and who is not.
References:
Frelat, R., Lopez-Ridaura, S., Giller, K. E., Herrero, M., Douxchamps, S., Djurfeldt, A. A., ... & van Wijk, M. T. (2016). Drivers of household food availability in sub-Saharan Africa based on big data from small farms. Proceedings of the National Academy of Sciences, 113(2), 458-463. 10.1073/pnas.1518384112
Giller, K., Delaune, T., Vasco Silva, J., Van Wijk, M.,Hammon, J., Descheemaeker, K. ... & Anderson, J.A. (2021). Small farms and development in sub-Saharan Africa: farming for food, for income or for lack of better options? Food Security, forthcoming. https://doi.org/10.1007/s12571-021-01209-0
Jayne, T. S., Chamberlin, J., & Headey, D. D. (2014). Land pressures, the evolution of farming systems, and development strategies in Africa: A synthesis. Food Policy, 48, 1-17. https://doi.org/10.1016/j.foodpol.2014.05.014
Jayne, T. S., Mather, D., & Mghenyi, E. (2010). Principal challenges confronting smallholder agriculture in sub-Saharan Africa. World Development, 38(10), 1384-1398. https://doi.org/10.1016/j.worlddev.2010.06.002
Khalil, C.A., Conforti, P., Ergin, I. & Gennari, P. (2017). Defining small-scale food producers to monitor target 2.3. of the 2030 agenda for sustainable development. FAO Statistical Division Working Paper ESS/17-12. http://www.fao.org/3/i6858e/i6858e.pdf
Ritchie, Hannah. (2021). Smallholders produce one-third of the world’s food, less than half of what many headlines claim. Our World in Data. Published online on August 6. https://ourworldindata.org/smallholder-food-production