Kikuyu counterfactuals

GUEST POST: Ethnic favouritism and its discontents

A few weeks ago, Johan Fourie discussed his recent work with Laura Maravall and Jörg Baten, which showed that African leaders disproportionately hail from the more educated regions of their respective countries. Most quantitative studies of ethnic favouritism by African presidents overlook this endogeneity problem, and consequently the effects they find may be overstated. In a related paper on ethnic favouritism in education, Elliott Green and I have argued that this is far from the only problem dogging the measurement of this phenomenon.

The quest to measure ethnic favouritism in Africa has spawned a small academic cottage industry. It seems fair to say that the interest in this topic is not driven by normative concerns about fairness per se, but by the presumption that ethnic favouritism or discrimination influences political behaviour, and fuels grievances and populism that result in poor policy outcomes and social strife. But for that to be the case, surely measures of ethnic favouritism need to be intuitive and visible to the layman, so that citizens hold reasonably accurate perceptions of the privileges bestowed or withheld from them. By unpicking a set of studies on ethnic favouritism in education, we show how complex and unintuitive these measures quickly become, and how unstable the results are once subjected to robustness tests.

Our paper follows the crowd and focuses on Kenya, which has proven to be a popular laboratory for such studies because of the undeniable influence of ethnicity on Kenyan politics and the sharp transitions in ethnic leadership that make favouritism comparatively easy to spot. Power passed from a Kikuyu president (Jomo Kenyatta) to a Kalenjin president (Daniel Arap Moi) in 1978, and back to a succession of two Kikuyu presidents in 2002 (Mwai Kibaki followed by Uhuru Kenyatta). Conveniently for studies on education, the two groups that have dominated the presidency stood at opposite ends of the educational attainment spectrum, with the Kikuyu having the largest educational advantage at independence (as predicted by Maravall, Baten and Fourie’s model), while the Kalenjin had below average attainment.

A series of papers have found evidence of ethnic favouritism in education in Kenya (Franck & Rainer, 2012; Kramon & Posner, 2016; and Li, 2018). Using retrospective data on schooling achievement, they seek to demonstrate that people who shared the ethnicity of the president during their school-aged years have higher attainment than they would have had, if the ethnicity of the President had been different. But how do they conceptualise this counterfactual world of re-shuffled Presidential ethnic identities? And is such a counterfactual all that politically relevant?

In one sense, level differences alone are a measure of favouritism, at least if most of the costs of schooling are financed by the central government. If the Kikuyu in Kenya, for instance, were to retain their educational lead throughout the postcolonial period (which indeed they did), successive governments would have spent more of the recurrent education budget (per capita) in Kikuyu regions, than in any other part of the country. It may well be such absolute inequalities that fuel inter-ethnic grievances, irrespective of how they emerged or how difficult they prove to dismantle. But clearly none of the papers on educational ethnic favouritism has this counterfactual in mind. None assume (understandably), that, say, teachers and resources would have been transferred out of Kikuyu regions after independence to quickly eliminate the regional enrolment gap, had Kenya’s first President not been Kikuyu.

Alternatively, one might choose to assume that a government unencumbered by ethnic bias will try and equalise the amount of additional, new resources it delivers to each group or region (e.g. numbers of schools constructed), so that the net effect is that each group gains, for instance, one additional year of attainment on average over a given period. This is how Kramon and Posner (2016) and Li (2018) measure ethnic favouritism in their primary specifications. This approach has some intuitive appeal. Residents might, hypothetically, be able to compare and gauge that their community was given roughly the same level of investment (per capita) as a neighbouring one. The problem with this measure is that years of schooling have an upper bound. Thus this measure loses relevance if any group starts approaching universal primary attainment (as the Kikuyu did in the 1970s). In Kramon and Posner’s model, this convergence effect owing to the Kikuyu slowdown in attainment growth coincides with the shift from a Kikuyu to a non-Kikuyu president, and is mistaken for favouritism.

Alternatively, if we place more weight on demand as the driver of attainment growth, we might expect the absolute gains in attainment to remain the largest in the communities that already have a head-start. This may in fact be a more realistic assumption in Kenya in the early independence period, where the cost of primary school construction was typically financed by local communities rather than the central government. In this scenario, one might deem the outcome fair if the growth rate in the years of schooling was equal across groups, so that attainment growth for all groups was, for instance, 10% in a given period. However, such a measure suffers from the same upper bound problem as above, and depending on the constellation of groups, it could have the unintuitive result that ‘favouritism’ is uncorrelated with the largest absolute educational gains.

An alternative, and more defensible, approach, therefore, to assume a quadratic growth function, where for each ethnic group, the rate of attainment growth slows as values approach the upper limit. This is the underlying logic in the models used by Franck and Rainer (2012) (who test this for several countries), as well as Kramon and Posner, in one of their alternative specifications. However, this makes the notion of favouritism far less intuitive, as favouritism is defined in relationship to a complex counterfactual, where levels and rates of attainment growth will vary across ethnic groups and time. The exact shape of this group-specific curve could be quite sensitive to the start and end point, and, depending on a group’s place on this curve, can yield counterintuitive results.

Furthermore, when we replicate such models for Kenya with the inclusion of additional data and some small improvements to said specifications, the results prove unstable, mostly statistically insignificant, and the effect sizes are very small. We look at secondary school attainment too and find similarly inconclusive results.

In a separate paper with Andrew J. Harris, I have also looked at Kenya’s provision of university education. As the private returns to university education in Africa are large and the degree of state-subsidisation has historically been high, this would presumably be an area particularly ripe for favouritism. We categorise graduates from the University of Nairobi, Kenya’s most prestigious public university, by ethnic group, based on their names. While our intent was not to explicitly test for favouritism, the group shares alone are illustrative. They show little change in the Kikuyu share of graduates during either Kenyatta’s or Moi’s presidencies. In fact, the Kikuyu share of graduates started to fall only after Kikuyu President Kibaki came to power, presumably for reasons that have little to do with the President’s ethnicity.

All in all, we find surprisingly little evidence of favouritism in educational attainment in Kenya, although our results do not conclusively disprove ethnic favouritism either. But if the effects are so small and complicated to identify that they cannot be visualised in a few simple charts, then it seems doubtful that public perceptions of the President’s meddling in the provision of education services will be remotely accurate or helpful in delivering political support.

Photo by Ken kahiri on Unsplash.