Regressing to Colonial Times

Daron Acemoglu doesn’t just want to explain democracy. He also wants to explain economic development.

In a major contribution from 2001, Acemoglu and coauthors Simon Johnson and James Robinson (AceJohnRob) argue that good institutions lead to economic growth.

While the AceRob work on democracy proceeded from game theory peppered with stylized interpretation of a few countries’ history, the AceJohnRob utilizes regression techniques. Indeed,what sets the AceRobJohn work apart from the long-line of “institutions causing growth” literature is an empirical strategy governed by the requirements of state-of-the-art regression work. In contrast, previous institutional research by Douglass North, Jack Knight and others seems motivated largely by positioning the author in relationship to the canon of Western historians and social theorists.
Almost as soon as the new institutional economics emerged on the scene, development scholars were arguing that – rather than institutions causing growth – growth might lead to good institutions. As countries get richer, their values or incomes orient them towards things like inclusive institutions and property rights protections.

The AceJohnRob work attempts to resolve this potential endogeneity / reverse causation problem.

They contrast and compare the impact of institutions on growth for a group of 64 colonized countries under ordinary least squares (OLS) regression and two-stage least squares regression (2SLS) with an instrumental variable. Wikipedia fairly ably explains what this latter method is about. If one is worried that an explanatory variable (say institutions) is partially explained by the posited dependent variable (growth), then one can try to find a third (instrumental) variable that is correlated with the first but not the latter, and add it into the analysis.

AceJohnRob settle on an instrumental variable of European settler mortality in the colonial era, an explanatory variable of average expropriation risk in the modern era, and a dependent variable of per capita growth in the modern era. The idea is this: when Europeans could safely settle in a colonized country, the colonial power allowed them to set up the good ol’ institutions from the home country. Because of a variety of incentive effects, these good institutions tended to persist through time, leading to economic growth. Settler mortality then directly explains good institutions, but not economic growth.

Their results? Settler mortality “explains” even more of economic growth in the 2SLS model as the OLS one, providing support for the notion that institutions explain development, rather than the reverse. They then show that their institutions causing growth theory outperforms other theories that link development to other factors, like identity of colonizer, legal origin, climate, religion, geography, natural resources, soil quality, ethnolinguistic fragmentation, malaria, life expectancy, infant mortality and current fraction of Europeans in the population.

The work seems to have stood up pretty well over time: Jeff Sachs was unable to assert the superiority of his geography as explanation for development thesis. More recent work by David Albouy challenging the settler mortality data seems to have fared little better.

What to make of this tour de force?

Firstly, AceJohnRob dash as many hopes as they do theories. What can countries do today to change the disease environment that confronted white settlers hundreds of years ago? What can they do if they were “cursed” by having lots of people or natural resources to extract, which might have kept vibrant settlements with good institutions from being set up. Catching up to first world standards seems remote in this grim picture.

Secondly, as they note, institutions are treated as a black box. Their preferred explanatory variable is expropriation risk, which is an effect of institutions rather than an institution itself. They assert in a footnote that all other good things (i.e. education, civil liberties) would cluster around expropriation risk. But there is no attempt in this paper to show how this clustering works, or what causes any bit of the cluster to produce growth.

Finally, as Mushtaq Khan recognized in a different context, regression analysis will tend to drown out the experience of the few successful late developing nations. Considerable expropriations through reallocation of land rights seems to have been a common precursor to successful development. Studying the “winners” might yield more interesting insights than studying the unfortunately vast number of development “losers.”