"Good institutions" are tautological or unsupported

The Nobel-winning theory has no value for development policy

2001 saw the publication of The Colonial Origins of Comparative Development by Daron Acemoglu, Simon Johnson and James Robinson. AJR showed that measures of “institutions” (rules and norms governing economic activity) across countries caused those countries to have different long-run development outcomes. In fact, their estimated effects of institutions negate every other factor: after controlling for institutions, there is no income gap between Africa and the rest of the world, or between tropical and temperate countries. If you took their results at face value, AJR had solved development – they had found the root cause of all development differences.

It’s no wonder that the paper went on to be wildly influential. It garnered 20,000 citations, formed the basis of Acemoglu and Robinson’s bestseller Why Nations Fail, and shaped a generation of development policy thinking. In 2024, the three authors won the Nobel Prize “for studies of how institutions are formed and affect prosperity”. It was a total victory.

There’s just one little question.

How exactly do you put institutions in a linear regression?

What’s in an institution?

The 2024 Nobel reignited an old controversy about the empirical credibility of AJR’s studies. To put it lightly, there are problems.

But debating causality has always struck me as pointless. There is a much bigger problem with the Nobel-winning literature: institutions, as measured by empirical studies, are a meaningless concept with no prescriptive value.

Literally speaking, a linear regression tells you the effect of changing one number (X) on another number (Y). If we’re looking at the effect of education on wages, X is years of schooling, while Y is wages; statistics disentangles the effect of one on the other. Here, Y is simply GDP per capita. But what is X?

In AJR’s case, X is a country’s average score on a “protection against expropriation” index, compiled from surveys of investors. So when AJR show that “institutions cause development”, what they have technically shown is that investor perception of expropriation risk proxies for something that causes development. But whether that something corresponds to what we would normally call “institutions” – a complex mashup of courts, parliaments, bureaucracies, etc – is an entirely different question.¹

Of course, it isn’t intrinsically wrong to simplify a high-dimensional concept into a single number. For example, we could measure the sentiment of news articles on a numerical scale, and estimate the effect of news sentiment on business investment. Similarly, you could argue that various institutional quality measures are useful compressions of the concept of “institutions”.

The problem is that compressing institutions into a single quality score makes policy advice meaningless. What are you going to do, tell a Kenyan minister to have institutions that are 7 out of 7 rather than 4 out of 7? If they don’t kick you out of their office, they’ll ask you how the hell they’re supposed to do that. Then you explain how the 7-point scale was built — some combination of rule of law, property rights, contract enforcement, constraints on the executive, corruption, regulatory quality, bureaucratic capacity – and you actually can’t tell them which of those is the important one to move. All of them? A secret other factor that is just correlated with the ones on the list?

We have a bunch of knobs, and we’re pretty sure that turning all of them would increase GDP, but we don’t know what turning any specific knob will do. Or which knob is the most important to turn. Or whether that’s even a meaningful question.

The root cause of this paralysis is that “good institutions cause development” is a tautology. We only think of rule of law, strong property rights, and low corruption as “good institutions” because we noticed that developed countries have those things. It’s not as if we derived the idea that “rule of law matters for development” from first principles, and then checked whether rich countries had it. When you construct good institutions this way, of course you’re going to find that good institutions cause countries to be rich.

When AJR claim that institutions fully explain cross-country differences in GDP, this is the tautology in action. Of course there is no cross-country difference in GDP after you control for a measure chosen because it correlates 0.9 with GDP.

Non-tautological stories are unsupported by the data

I’m simplifying the institutional theory of development more than some would say is fair. There are ways to make a coherent and non-tautological story out of the institutions-development link. In Why Nations Fail, Acemoglu and Robinson expressed the most popular version: countries prosper when their institutions are “inclusive” in the sense that they spread economic opportunity broadly across the population. These include broad-based property rights and well-functioning markets. By contrast, “extractive institutions” that serve only a predatory and rent-seeking elite lead countries to stay poor.

This is not a tautology, because it features a specific story linking institutional quality to development. Unfortunately, this story is not supported by the data. I don’t mean that the particular idea of inclusive vs extractive institutions is unsupported. I mean that any story that distinguishes between types of institutions is unsupported by the data – because the data simply doesn’t separate types of institutions.

Suppose we wanted to answer which of “constraints on the executive” and “property rights” is a more important institution for growth. Ideally, we would compare the incomes of countries with constrained executives but no property rights to the incomes of countries with property rights but unconstrained executives. If the former were richer than the latter (and we were somehow convinced that this difference was causal) then we might conclude that constrained executives are “more important” than property rights. This way, we could build up a coherent picture of which institutions matter and go on to reason why they might matter.

The problem is that “countries with some good institutions and some bad institutions” broadly don’t exist. Good institutions are a package deal.² As long as that holds true, data cannot distinguish between the types of institutions that matter.³

The solution is to be more specific

In response to these problems and others, economists stopped studying “institutions” as a broad category. This is a good thing, because what followed was much more successful at actually identifying institutional problems that affect development.

Take one of my favorite modern papers, Boehm and Oberfield (2020). They study the effect of contract enforcement on Indian manufacturing, exploiting the fact that court congestion reduces the ability of firms to enforce their contracts. They show that in states with more congested courts, businesses use fewer specialized inputs and produce more in-house, even when in-house production is less efficient.

This is high-quality, careful research that tells a clear story about how failures in a particular institution – contract enforcement – hold back development. It makes a basically obvious policy prescription (reduce court backlogs) but it also tells us why the particular institution of contract enforcement matters for development, by allowing firms to build specialized supply chains. In other words, we can make productive statements about institutions – but to do so, we have to stop talking about “institutions” and start talking about a particular institution.

Conclusion

It’s obvious to anyone with eyes that “institutions matter for development”, in the normal sense of “institutions” and the normal sense of “matter”. This is why the statistical critiques of the institutions literature never felt important to me. But studying institutions under an umbrella gives us no insight as to how they matter, and therefore does not generate useful policy advice.

I think about this problem a lot, because I meet many people in the development policy community who grumble that development economics has drifted into irrelevant tinkering, and that we need to return to answering big questions. I sympathize with this view, but it would be a huge mistake to equate “answering big questions” with “answering questions about sweeping concepts”. Just because your question is big (e.g. “how do institutions affect development”) doesn’t mean your answers have any meaning!

As Boehm and Oberfield show, it is possible to do research on institutions that is both granular enough to be policy-relevant and broad enough to be relevant to economic growth. So let’s leave the “institutions” literature in the trashcan where it belongs, and focus on answering specific questions about specific institutions.

1. This is a well-known critique first expressed by Glaeser et al 2004, although it was weirdly absent from the debates I saw after the Nobel announcement. Yes, I have been stewing on this since 2024.

2. This also binds if we want to improve on the cross-sectional evidence with panel data, ala Acemoglu et al 2019. We would need countries to improve some institutions but not others in order for cross-institution differences to be identified from panel data.

3. Langbein & Knack 2010 argue this formally, showing that worldwide governance indicators have such strong correlations with each other that they likely all represent a single factor that can’t be distinguished by different measures.