The interactions between the physical things that exist in the world – human beings, tools, animals or pieces of land – can be structured in many ways and institutions can be thought of as the rules that structure and condition these interactions (which may depend on other things such as values, beliefs – culture). The focus on economic institutions is motivated by the hypothesis that given a set of such physical objects, different institutions for organizing their interaction will have very different consequences for productivity and economic prosperity.

Take, for instance, the economic institutions of the Trobriand Islanders famously described by Malinowski (Reference Malinowski1922). These islanders grew yams and taro on their coral islands. Land was not owned by anyone, but was allocated by the chief to families. The chief, ‘notables’ and particularly magicians, decided what was to be grown and there was a complex planting system regulated by religious ceremonies. Families did not consume what they grew but were embedded in a complex system of ‘exchange’ and reciprocal gift-giving. Every man had to give almost all of the produce of his plot to his sisters. Barter existed (there was no money) but the terms of trade between all goods were fixed by tradition. The most famous type of exchange that existed was the Kula Ring, a system of trade between the islands. There were two sorts of trade goods: red shell-disc necklaces (veigun or soulava) that are traded to the north (circling the ring clockwise) and white shell armbands (mwali) that are traded in the southern direction (circling counterclockwise).

Most economists would agree that such a system of economic institutions would not maximize the economic potential of the Trobriand Islands. How do we know this? Well, first we know factually that the Trobriand Islanders were very poor in the early 20th century when Malinowski studied them. We also have strong theoretical priors that such a system created very poor incentives. For instance, since people were not residual claimants on their output, they did not have incentives to exert effort. It is clear, however, that Trobriand society was made up of a vast mesh of different institutions. Was productivity low because of the gift-giving or because land was not privately owned? Was it low because magicians were not good at picking the right time to plant or because they forbade people from planting crops that would have been more productive? Was it low because fixed relative prices again removed incentives to adopt better technology or exert effort? Even elements of this description are uncertain. Land might not have been privately owned but, in practice, lineages that may have had de facto user rights so the powers of chiefs to re-allocate the land were purely notional.

In his paper, Voigt (Reference Voigt2012) poses the question of how we are to measure institutions in a way that will allow us to estimate their causal effects. I like very much his approach to defining institutions and the agenda he lays out is important. I particularly like the use of the word ‘pragmatic’.

There are several sets of issues here, some important and some not important. Amongst the unimportant are that the empirical literature fails to do justice to the distinction between institutions and policies and that both are obviously outcomes. I do not see what the important distinction between policies and institutions is. They are both collective choices which create incentives and constraints. Yes, of course, both are outcomes (endogenous), which is why so much work has been done to trying to control for this econometrically.

The important points are that many existing measures of institutions are subjective, they are complex aggregates of many specific institutions, and there needs to be a clear distinction between the de jure rule and the de facto enforcement. He is critical of attempts to measure the ‘security of property rights’ by asking experts and of measuring ‘democracy’ since this is clearly a bundle of many specific institutions. From these observations, he draws the conclusions that we have to be very specific in specifying and measuring institutions and not look at bundles, we need to be as objective as possible in the way that we do this, not relying on opinion, and we need to measure both de jure and de facto elements.

I am sympathetic to all of these points, but I think there are also problems with them. First, institutional measures which are subjective may be problematical, but here econometrics comes to our aid. Such subjectivity is really just a form of endogeneity, so if you have a valid instrumental variable then you can solve this problem. Hence, this issue should really be bundled into the whole problem of inference. I am not saying objectivity is not important, but it is less of an independent problem than the article suggests. Second, measuring specific institutions is also important but it comes with a large number of problems. Go back to the Trobriand Islands. Here we would need to start measuring the rules that the magicians used for deciding when to start planting, or how the chiefs decided who to give the land to. One obvious problem is that so many of these institutions are specific that one has to ask where the variation comes from to measure the consequences of them. If only the Trobriand Islanders had the Kula Ring, you cannot estimate its effects except maybe if it varied over time. Even if you could, ‘going specific’ creates an important trade-off between internal validity and external validity. We might have estimated the causal effect of an institution that exists nowhere else in the world. Much of the existing work which is criticized in essence takes this mesh of institutions and says ‘the Trobriand Islanders had poorly enforced property rights’. Obviously not picking up the richness of the de facto and de jure institutions and their multifarious consequences (one might even ask: could the Trobrianders have de jure institutions without writing when Voigt says that de jure rules relate to ‘legislation’ – I think they could actually) but nevertheless a possibly useful characterization with much greater external validity than an estimate of the causal effect of the Kula Ring. Moreover, many institutions come as a bundle because they are created by a particular society. For example, when French armies invaded and ‘reformed’ much of Europe after 1792, they tended to impose the same package everywhere, abolishing feudalism and guilds, writing constitutions, later introducing the Napoleonic Code (Acemoglu et al., Reference Acemoglu, Cantoni, Johnson and Robinson2011). In this case, there is again no variation you can use to estimate the causal effect of abolishing guilds. Finally, I agree with the notion that it would be good to measure the rules and their enforcement and how this changes behavior, again though we need the variation. Much of the impact of rules comes ‘off the equilibrium path’ and sanctions which never happen can have big effects on equilibrium behavior. To estimate the effects of such rules, we need natural experiments or shocks which we may or may not have. Perhaps here other methodologies, for example laboratory and field experiments, can be more powerful than observation. However, again this comes with a trade-off: many questions we are interested in cannot be addressed in this way.

So measurement is crucial, but ultimately convincing ourselves that institutions matter will involve a pragmatic mixture of theory, case studies, along with (imperfect) measurement and hypothesis testing.