Why Nations Fail and the long-termist view of global poverty

Within the effective altruism community, people often talk about “long-termist” vs “short-termist” worldviews. The official distinction between the two is that short-termists prioritize problems by how they affect people alive today, while long-termists prioritize problems by how they could affect humanity’s entire future trajectory. In practice, people usually treat this as synonymous with prioritizing either existential risk reduction (if long-termist), or scaling up proven global health interventions (if short-termist).

It’s a bit surprising that each worldview should have exactly one favorite cause area, though. Couldn’t you have short-termist work on existential risk, or long-termist work on global poverty? In reality, these supposedly discrete worldviews seem more like correlated clusters of various different beliefs:

It’s understandable why some of these are correlated, but there must be a lot of people who fall through the cracks between the clusters. What if you share short-termists’ skepticism of weird claims and hypothetical risks, but you’re willing to focus on first-principles reasoning and work on a long time scale?

You’d still want to focus on something that’s a problem today, so you’d probably want to work on global poverty. But you’d dismiss GiveWell’s top charities as treating a symptom and not a cause. Why do these countries need charity in the first place? South Korea used to be just as poor as anywhere in Africa, but today it’s incredibly prosperous, while sub-Saharan Africa has made way less progress. If we could move the lowest-growth countries from their current trajectory onto South Korea’s, we’d have done much more than any single malaria-eradication campaign could.

If that were your worldview, you’d really enjoy Why Nations Fail, one of the best attempts I’ve seen at getting a first-principles understanding of what affects countries’ long-term economic growth.

First of all, what is economic growth? It’s when people produce more (or more valuable) stuff with the same effort.¹ The first and least controversial point in Why Nations Fail is that for a nation to keep on doing more with less, its individual citizens need to be incentivized to become more productive. In particular, the state should not set up systems where, whenever someone gets more productive, other people come and take away the extra stuff they produced. Those systems are what the authors call extractive economic institutions, and they include things like slavery, serfdom, indentured servitude, roving bandits, guilds, collectivized agriculture, nationalization of private assets, officials requiring bribes, kangaroo courts, banana republics, and other [animal or vegetable] [civic institution].

You might think you could grow your economy under extractive institutions by forcing people to become more productive even if they won’t get to keep the surplus. This does often work in the short term (and the short term can last a surprisingly long time)—for instance, Soviet Russia grew at about 5% annually from 1930-1970,² despite having extremely extractive institutions. But that only worked because the growth started from a low base. Soviet Russia’s economic institutions, while horrible, were arguably still less extractive than the serfdom that preceded them. Plus, the growth came mainly from “easy wins” like people switching from horrifically inefficient farming to horrifically inefficient industry.

In fact, growth under extractive institutions can only come from easy wins—easy enough that they’re obvious to whoever has the job of forcing other people to be more productive. (Nobody is going to generate their own productivity-boosting ideas, because they have no incentive to.) Any half-decent Soviet central planner could have told you that industry would be more productive than farming, so they were able to force peasants to make the switch. But nobody was good enough at central planning to fix all the ways their industry was inefficient—let alone come up with new inventions or technologies as fast as the US was. So once all the low-hanging fruit was plucked, circa 1980, the Soviet Union ran into a productivity wall and collapsed.

What’s more, even if an innovation is obvious to the extractive elite, they might actively try to crush it because it threatens their hold on power. For instance, in 19th-century Europe, it was obvious that factories and railroads could be a huge productivity improvement, but the extractive rulers of peripheral Europe wanted none of them. Here’s Austria:

When a plan to build a northern railway was put before [Emperor] Francis I, he replied, “No, no, I will have nothing to do with it, lest the revolution might come into the country.”

And here’s Russian Finance Minister (1823-44) Count Igor Kankrin:

[R]ailways do not always result from natural necessity, but are more an object of artificial need or luxury. They encourage unnecessary travel from place to place, which is entirely typical of our time.

It might seem cartoonish for a ruler to ban technology out of fear of TEH REVOLUTIONZ—until you remember how many extractive states today, like Ethiopia and China, have resisted the Internet for similar reasons.

The next part of Why Nations Fail asks: if economic institutions explain long-term growth, what explains long-term economic institutions? The answer, maybe obviously, is politics.

Extractive economic institutions tend to persist more when social power is concentrated in a small elite (“extractive political institutions”) that can use this power to take the economic surplus for themselves—or just destroy it if it looks scary and potentially destabilizing. They persist less when social power is distributed widely in the population (“inclusive political institutions”) because in that case the people being extracted will often have enough political power to fight back against the extraction.

For instance, in the early 1900s in the United States, various industries (steel, oil, banking, chemicals, farming tools, etc.) ended up under the control of extractive monopolists who charged too much, paid too little, pocketed the difference, and spent the difference on gold-plated toilets. But even the world’s fanciest toilets couldn’t defeat the mass populist movement that rose up to oppose the monopolies (because the US was sufficiently pluralistic that mass populist movements were very powerful), and so they were eventually forcibly broken up and regulated.

In states with less pluralistic politics, the elites can get away with this kind of extraction. For instance, when the British colonized Sierra Leone, they set up monopoly cocoa and coffee purchasing boards that bought low from the farmers, sold high on international markets, and pocketed the margin. After independence, instead of dismantling the boards, Sierra Leone’s presidents turned them into a slush fund, driving the margin up to 90%.

Sierra Leone illustrates another sad fact about extractive economic institutions, which is that they persist surprisingly much across nominally different political regimes. What happened in Sierra Leone happened nearly everywhere in sub-Saharan Africa: the colonists were ousted by a revolutionary leader, who quickly realized that the colonists’ giant money-printing machine was still in place, and decided that maybe the entire colonial apparatus didn’t need to be smashed. And of course, in order to keep their hold on the economic surplus, the new “democratic” government quickly resorted to repression and violence. (Why Nations Fail is very clear that what matters is de facto, not de jure, democracy, and that confusing these two will lead to very wrong conclusions.)

Over long timescales, then, it seems like a mix of inclusive and extractive institutions is unstable. If institutions are inclusive enough, then any group that deviates and tries to institute extractive norms will get shut down by the rest of society. If institutions aren’t inclusive enough, one of those groups will eventually succeed, and then be incentivized to consolidate their power even further until the society is totally extractive.

The most glaring exception to this rule—which many reviewers have pointed out—is China since Deng Xiaoping, which has relatively inclusive economic institutions, but extremely authoritarian politics, for the last 40 years. Acemoglu and Robinson acknowledge this, but claim, basically, that China is not in a stable equilibrium:

All the same, [Chinese] growth will run out of steam unless extractive political institutions make way for inclusive institutions. As long as political institutions remain extractive, growth will be inherently limited, as it has been in all other similar cases.

This is a kind of weak explanation as written because they don’t even mention the mechanism by which it will be limited, but with some charity I think you can construct a decent argument. Concretely, it seems like the Why Nations Fail model predicts that China’s current, inclusive economic institutions are not sustainable without inclusive political institutions to support them. In the current regime, the Chinese political elite’s local incentives are to extract as much of the economy’s surplus as possible, because they can; it’s not sustainable to just hope they keep being altruistic enough to not do that. At some point, unless China’s political institutions give real power to a broad majority of people, the elites will give in to temptation and start taking all the economic surplus for themselves.

(Note by the way that China’s recent economic slowdown does not confirm this model’s predictions, unless you think the slowdown was caused by the political elite becoming more economically extractive, which doesn’t seem like the case.)

If you’re worried about having powerful authoritarian nations hanging around, this is a comforting model to have. On the other hand, China has been “out of equilibrium” for almost half a century, and Why Nations Fail doesn’t offer any framework for thinking about how long these disequilibria can last.

The strongest claim in Why Nations Fail, which is mostly left annoyingly implicit, is that institutions are the dominant factor in shaping economic growth, over long enough time horizons. Many of their reviewers disputed this.

To be fair, a lot of those reviewers seemed to have rounded off this claim to “institutions are always the dominant factor,” and then complained about the book not explaining China (despite the half-chapter explaining that China is out of equilibrium), or not sufficiently explaining Bolivia vs Vietnam over the last 30 years (if China can be out of equilibrium for 40 years, so can other countries), or similar. But even over long periods of time, it’s reasonable to ask whether institutions were really the most important possible thing.

The most obvious alternative candidate is geography (including natural resources). Why Nations Fail makes some attempt to argue that geography isn’t sufficiently explanatory, citing (a) the divergence between North and South Korea, and (b) that after European colonization of the Americas, the correlation between latitude and productivity went from negative (tropics more productive) to positive (temperate regions more productive):

As we saw in the last chapter, at the time of the conquest of the Americas by Columbus, the [tropical] areas… held the great Aztec and Inca civilizations…. In sharp contrast… the [modern] United States, Canada, Argentina, and Chile, were mostly inhabited by Stone Age civilizations lacking these technologies. The tropics in the Americas were thus much richer than the temperate zones, suggesting that the “obvious fact” of tropical poverty is neither obvious nor a fact. Instead, the greater riches in the United States and Canada represent a stark reversal of fortune relative to what was there when the Europeans arrived.

The reversal of fortune happened, they argue, because colonists were more likely to install extractive institutions in the densely populated (wealthiest) areas of the Americas, where they could rely on forced indigenous labor; and inclusive institutions in the sparsely populated areas where there was no forced labor supply and settlers needed to be incentivized to be productive. This difference in institutions eventually led to stagnation in the tropics and faster growth in the temperate regions.

This doesn’t seem like a very definitive refutation, though. The Koreas example only shows that institutions dominate when geography and culture are held constant (plus, it’s cherry-picked to be maximally extreme). The example of the Americas is stronger, but I can imagine alternative models. For instance, maybe being tropical was an advantage before the industrial revolution because it meant less lethal weather; but a disadvantage afterwards because it meant that your comparative advantage was farming, rather than industry, and farming-specialized economies grow less quickly because they don’t develop compounding technical knowledge.

The example of the Americas’ reversal of fortune also doesn’t address more complex geographical factors, like that iron, coal or navigable rivers are necessary for industrial growth. In the medium term, geography of this form seems obviously very important; it’s no coincidence that England was rich in all three. In the long term, it seems like they might become less important, because the types of resources that are important will change. (Today if you’re a developing country with no iron or coal deposits, you can trade with other countries to get them; if you don’t have navigable rivers, you can build railroads or highways.) But neither of those is a perfect substitute, so it seems like there’s still a strong case for geography mattering.

On the other hand, if you’re trying to encourage development, not just understand it, it’s not clear how actionable the pro-geography argument is. Even if geography is important, it’s not particularly tractable to change! The only implication I can think of is maybe that liberalizing immigration becomes more important, so that it’s easier for people to move from resource-poor to resource-rich areas.

If true, the claims in Why Nations Fail have huge implications for (global-poverty-focused) effective altruism. Global poverty EA currently focuses on scaling up development aid interventions that have a strong evidence base of randomized controlled trials. But if you buy Why Nations Fail’s argument, you should probably prefer to take countries with extractive institutions and move them towards inclusivity, if that problem is tractable.

To me, it does seem tractable. For instance, in the recent election in Senegal, where I live, the incumbent president had a huge spending advantage, which was widely rumored to be because he diverted public funds to his campaign. (Not coincidentally, many prominent opposition politicians are currently in jail for misuse of public funds—the result of an investigation initiated by the same incumbent president.) The current Senegalese media is reluctant to investigate because the media in the capital are largely owned by people high up in government. Sponsoring the development of independent investigative media in Senegal seems like a (relatively) straightforward step towards improving Senegalese political institutions.

Of course, this take is only based on a couple hours of conversation with my Senegalese coworkers, so it’s definitely not optimal. Maybe it would even make things worse! But hopefully it illustrates the general point, which is that we know what inclusive institutions look like, so trying to move things in that direction is mostly a question of strategy and execution towards a known goal, not of solving some sort of mystery.

Why Nations Fail provides a compelling theory, with clear implications, supported by many examples, a few of which (like the Koreas or Americas examples) even had some kind of “control group.” But as the book goes on, it becomes more and more glaring how much it relies on anecdotes, not studies. Unfortunately, that’s probably because the empirical situation is abysmal.

Acemoglu and Robinson might not agree with that claim, since they were publishing empirical studies of institutions in economics journals for a long time before they wrote Why Nations Fail. But they also chose not to cover most of that empirical research in the book, and I can see why: I couldn’t find a single study that seemed really defensible.

Some of the problems I saw were silly and avoidable, like treating Likert scales as continuous instead of leveled-categorical or obviously violating the assumptions of instrumental variables. But as far as I could tell, the bigger issues were core to any attempt to empirically study the causes of growth:

• The sample size is tiny. There are only 200 countries, many of which are too small or too weird to be informative. (We do have hundreds of years of growth data for some countries, but different years of growth for the same country aren’t independent data points.) The “effective” sample size is even lower because the “error term” is often spatially correlated, which most studies do not properly account for, causing widespread p-value inflation.

• When you’re analyzing countries’ growth, everything is endogenous. Your institutions in year 1 affect your growth in year 2, which affects who gets more power in year 3, which affects how your institutions change in year 4. If you measure a correlation between growth and institutions, it could just as easily be that fast growth causes better institutions, not the other way around.

• Institutions and growth have a really complicated causal structure—both of them have a huge number of causes and effects. In situations like that, and especially with a sample size of 200, correlational studies never work. Even if you try to control for all the confounders, there are just too many subtle things that can go wrong.

• Ordinarily you might try to get around that by using a fancier causal inference technique, like finding instrumental variables—random events that affect a state’s institutions, but aren’t correlated with them. For example, one famous paper tried to do this with European mortality rates in colonized countries (the theory being that colonists were less likely to install extractive institutions in countries where more Europeans had immigrated).

  But a key assumption of instrumental variables is that the instrument only affects the outcome (growth) via its effect on other things you’ve measured (institutions or other covariates). In other words, to get a valid estimate, you’d need to measure every possible way that settler mortality could affect growth rates. Since measuring everything is impossible, and throwing it all into a model with only 200 data points is also impossible, that makes instrumental variables double-impossible. (Which didn’t prevent various economists from trying—apparently badly.)

These problems make me put almost zero weight on the empirical literature for institutions and growth, even before reading about the specific problems with specific papers. But they apply equally well to any attempt to find the causes of economic growth.

So where does that leave us?

Partly, it leaves us with a big methodological gap. The problems above make it relatively harder to get to a true understanding of what matters for development. But it doesn’t mean it’s hopeless to try and make intellectual progress on the question.

Indeed, I can imagine analyses that would make me more confident that the institutional hypothesis is correct—for instance, a systematic review with case studies of many different growth episodes, or many different institutional changes that you’d expect to affect growth. I’m not sure why Acemoglu and Robinson’s empirical work focused instead on what looks like badly-validated instrumental-variables models—maybe because they were easier to do (or easier to publish in top journals)? But given the amount of noise and the limited sample size, case studies seem much more likely to be informative than fancy models with questionable assumptions.

Nor is it unprecedented for lots of experts to agree on something even without airtight studies backing it up. A real-world role model could be something like the laws of supply and demand—which most economists believe in, even though they turn out to be tricky to verify in many real-world cases, like minimum wage laws or housing. Or even like the efficient-market hypothesis, which many effective altruists seem to endorse, far beyond our ability to empirically test it.

But on questions of poverty reduction, the effective altruism community doesn’t seem to be really looking for any methodology other than randomized controlled trials. In fact, this is a major (and not often addressed) criticism of global-poverty-focused effective altruism, including from Acemoglu himself in his Boston Review critique of effective altruism:

[P]recise measurement of the social value of a donated dollar may be impossible. … If, as some economists and political scientists suggest, changes in political and economic institutions are critical for long-run economic growth, then watchdog organizations such as Amnesty may be essential for transforming dysfunctional regimes. Effective altruists don’t (yet?) see the importance of these more political organizations. If this narrow focus continues, it may divert public and media priorities from political factors underpinning economic development.

His concern was shared by Angus Deaton:

More broadly, the evidence for development effectiveness, for “what works,” mostly comes from the recent wave of randomized experiments. … How can those experiments be wrong? Because they consider only the immediate effects of the interventions, not the contexts in which they are set. Nor, most importantly, can they say anything about the wide-ranging unintended consequences.

However counterintuitive it may seem, … [d]evelopment is neither a financial nor a technical problem but a political problem, and the aid industry often makes the politics worse.

Peter Singer’s response (which, as I’ve argued before, misses the point):

[I]f large-scale reform offers some prospect of reducing poverty, then effective altruists will try to assess its chance of doing good, and if the expected value of such action is higher than the expected value of more limited interventions, they will advocate working for the large-scale reforms.

It’s been four years since Peter Singer wrote that, and I don’t see much sign of the prediction coming true. The leading EA global poverty charity evaluator, GiveWell, begged off because institution-focused interventions wouldn’t be transparent enough:

Root-causes-based approaches are, in our view, the kind of speculative and long-term undertakings that are best suited to highly engaged donors (as discussed above).

But the EA community doesn’t seem to have produced many such “highly engaged donors,” with the possible exception of Aceso Under Glass giving to Tostan. (GiveWell itself may be in the process of becoming one. I hope that changes this!) Instead, most global-poverty-focused effective altruists seem happy to follow GiveWell’s recommended charities. Maybe that’s because all the long-termists are working on mitigating existential risks. But for any who are still sympathetic to global poverty as a cause area, Why Nations Fail provides an interesting example of an alternate paradigm.

Thanks to Eve Bigaj, Drew Durbin, Milan Griffes, Alexey Guzey, Holden Karnofsky, Joyce Keeley, Jeff Kaufman, Dan Luu, Aaron T, and Yuri Vishnevsky for their comments on drafts of this post.