How can we know how – or even if – institutions matter for growth unless we can accurately measure them? This is the central question posed by Voigt (Reference Voigt2012).

Few institutionalists are happy with the way we measure institutions, but many hold their noses and run the regressions anyway – who wants to look closely at how the sausage is made if it tastes good? Voigt would put a stop to this practice and demand that we examine not only the manufacturing process but also the sausage ingredients, down to the blood, guts, and mouse droppings.

This is important because these ‘institutional’ variables show strong and significant correlations with economic growth in cross-country regressions. Such powerful explanatory power is largely responsible for the growing attention paid to institutions in mainstream economic models and a marked increase in journal articles. At my last count (Shirley, Reference Shirley2008), there were 83 journal articles purporting to show that institutions are positively and significantly correlated with growth in GDP, and the number today is surely much larger. What passes for institutions in these articles varies: security of property rights and contracts, economic freedoms (such as competition and free entry), democracy, trust, religion, social clubs, and other measures of social cooperation, and (negatively correlated) political instability (coups, revolutions, riots and deaths). This list of variables illustrates the problem that Voigt has highlighted. How are we defining and measuring institutions if we include such panoply of phenomena as competition, democracy, social clubs and riots?

Voigt (Reference Voigt2012) joins a chorus of other voices, including my own (Shirley, Reference Shirley2008), in critiquing the most common measures of institutions. As Voigt (Reference Voigt2012) points out, these measures do not refer to specific rules or norms but to broad aggregates such as rule of law. Many are not institutions but the outcomes of institutions (secure property rights) or perhaps even, as Glaeser et al. (Reference Glaeser, Edward, Lopez-de-Silanes and Shleifer2004) assert, of policies.

Admirably, Voigt (Reference Voigt2012) goes beyond most criticisms in offering both a definition of institutions precise enough to permit measurement and a ‘pragmatic approach to measurement’. The definition follows North – institutions are formal or informal constraints that structure human interaction – and Ostrom – institutions are ‘commonly known’ rules that are the result of human action, but not necessarily deliberate human design. Voigt breaks down institutions into five types based on enforcement: self-enforcement, self-commitment, informal societal enforcement (e.g. ostracism), private third-party enforcement and state enforcement. This seems a useful classification, although I question whether all these distinctions are useful. Can we really observe the difference between self-commitment and self-enforcement? [In a less useful exercise, Voigt (Reference Voigt2012) tries to distinguish between economic and political institutions. The differences he cites are fuzzy, and for good reason. North et al. (Reference North, Wallis and Weingast2009) assert that all states are – or once were – limited-access societies with institutions designed to foster and protect alliances between economic, political and military elites. If they are correct, we should expect economic and political institutions to blur into and reinforce each other.]

Based on his definition, Voigt develops a set of theoretical assumptions with implications for measurement. In essence, says Voigt, we must measure the de jure content of an institution, the de facto implementation of the institution, and any informal institutions that might conflict with, reinforce or substitute for the institution. This strikes me as quite sensible, but difficult.

Recognizing that this is a very tall order, Voigt concludes with a ‘pragmatic approach to measuring institutions’. This pragmatic approach is not as simple as Voigt sometimes implies. Voigt notes that a single institution (he cites judicial tenure) is part of a nexus of rules and norms, and we must measure any other institutions that might affect the outcome of interest. For example, if we are interested in the judiciary's independence from political interference, then comparing the observed tenure of judges with the tenure required by law is not enough, as is evident in Feld and Voigt (Reference Feld and Voigt2003). We need additional objective measures to know whether politicians have found other ways to reduce the independence of the judiciary, such as packing the court with allies, paying judges so little that they can be easily bribed, starving the judges of budget to hire assistants or otherwise be well informed, and the like.

Voigt's pragmatic approach would greatly improve the accuracy and specificity of institutional measurement in growth regressions, and I applaud him for his careful and thoughtful approach. But we need to move beyond just doing better cross-country growth regressions. Knack and Keefer (Reference Knack and Keefer1995) achieved a major breakthrough when they showed that broad institutional aggregates matter for growth. Feld and Voigt (Reference Feld and Voigt2003) and others have shown that certain specific and carefully measured institutions such as judicial independence matter for growth. This is more precise, but it is still vague.

Cross-country growth regressions have told us what they can tell us; institutions matter for growth, and some institutions matter more than others. The interesting questions for the future will be different. For example, why are some countries able to create independent judiciaries and others not? Are democracies more likely to create independent judiciaries? How does an independent judiciary affect legal decisions, economic decisions and political decisions? Using Voigt's pragmatic approach to measuring institutional variables may help us address some of these questions, but new breakthroughs will also come from other methodologies. Comparative case studies and field experiments are arguably more likely to yield new insights than yet another cross-country regression, no matter how carefully the independent variables are measured.