Many people are tweeting this paper, because it is forthcoming in QJE. The big splashy headline results are being tweeted. But there are a few things from the guts of the paper that many people may not realise.

(1) Educational expansion in 1980-2019 ‘explains’ more growth if growth rates were lower. So education ‘explains’ the most in Latin America, Sub-Saharan Africa & rich countries, the relatively low-growth regions. Ed. explains the least in fast-growing countries like China & India

(2) Likewise with income groups *within* countries. Education ‘explains’ so much of the income growth of very low income groups because their incomes grew the most slowly…

Although, within the poorest groups, the upper end of the global poor benefited more frrom educational expansion than the most extremely-poor.

(3) Because of the preceding, by the paper’s OWN results, educational expansion has NOT contributed to income convergence *between* countries at all!!! ( although educational expansion contributed to the reduction of global interpersonal inequality )

( I note for the record that I am clarifying what the paper is saying, not asserting what I believe to be true. I happen to think the contribution of educational expansion to economic growth in fast-growing regions is underestimated via human K externalities etc. )

Final note: The Gethin QJE paper also quietly weighs in on "What was responsible for the global poverty reduction since 1980, the market or the state?" debate. Direct & indirect (via educational spending) transfers account for a big share of the reduction in global poverty

So now, in the online global economics culture wars between blue globe neoliberals and the anti-capitalist left, one of them can riposte, "see, it was NOT neoliberalism, it was govt intervention" whilst the other "increased transfers depended on growth enabled by free markets!"

@Afinetheorem But if the effect of education also shows up in faster & more adoption of better equipment, or via Kremer-style O-ring effects (or better governance), then the effect gets 'lost' in K and A.

@Afinetheorem the paper itself says its estimates are conservative and lower bound, for the simple reason that private returns are easier to measure than the social returns.