I’m currently preparing some slides for a talk on the uses of economic theory for pure empirical economists, seemingly a contradiction if there ever was one. In this paper – notes given to a conference on economic methodology – Varian discusses the broader question of whether theory is good for anyone. More importantly, he quotes Wordsworth in the introduction, and I’m easily suckered into a paper that begins some good verse.
Varian argues the critical characteristic of economics is that it is a policy science, more like engineering than physics, more like medicine than biology, in that our work has a broad goal of assisting policy decisions which will improve human welfare. He notes that policy sciences, for some reason, are less often studied by philosophers of science than the hard sciences. Though I generally disagree with Friedman’s famous claim that economic theory ought be judged on its ability to predict things we care about predicting, to the extent that economics is being written for the benefit of policymakers such a claim is more plausible.
The obvious use for theory is that data cannot only speak to the past, not the future. This is Hume’s induction problem. Theory provides a justification for extending results about the past into the future, because theory in the social sciences generally is a proposition about some consistent, general action or motivation for individuals and firms.
Welfare calculation is also inherently theoretical. To inform policy, we need to know not only what happened, but whether what happened was good for people. But what could that possibly mean? Consider the introduction of some good like Honey Nut Cheerios by a monopolist; this is Hausman’s famous example. Theory tells us precisely how we can calculate from data the welfare gain in willingness to pay using only demand data and the assumptions that firms maximize profit and individuals maximize utility.
In my presentation, I’m going even further and arguing that modern theory – games and mechanisms – are critical even for purely positive empirical economists. The reason is threefold. First, theory allows the justification of econometric assumptions, some of which are otherwise untestable, such as ignorability in propensity score matching. An example in health would be testing for a linear effect of health report cards on health outcomes, finding no effect, and calling it a day. The theory of reputational games suggests that reputation often shows arbitrarily small effects on choice unless the reputation mechanism is used by a sufficiently large proportion of the population. That is, the effect can be discontinuous, so a zero coefficient in a linear regression does not imply that report cards will be ineffectual as they become more common.
Second, many empirical papers propose counterfactual policies. Theory can tell us when these counterfactual policies are contradictory to the assumptions made in the original empirical estimation. For example, consider a paper that shows large gains from moving to employer based insurance to a health exchange because the exchange offers more choice over plans and can therefore better match heterogeneous preferences for premium/co-pay bundles. If the hedonic estimation of the value of some plan characteristics comes from a world where firms offer more than one plan to their workers, then we want to ensure the important heterogeneity is consistent with offering multiple plans. Unfortunately, rather heroic assumptions about the labor market are necessary to give multiple plans per employer in an equilibrium which does not devolve out of steady state due to adverse selection. The easiest assumption allowing multiple plans it that employees make mistakes when choosing plans. But if that is the case, the welfare implication assuming perfect choice on a non-work-based health exchange is invalid.
Third, theory allows us to interpret the results found in the data. The implications of action in strategic problems is often non-obvious: Varian points out that in tennis, improving one’s backhand can lead to the backhand being used less often in an equilibrium game of tennis. In the context of health, there is a paper which proposes avoiding adverse selection within a firm by taxing the healthy and subsidizing the sick; that is, rather than pay $4000 of each employee’s health insurance bill, we pay $2000 to Healthy Type and $6000 to Sick Type. We don’t see examples of this in the real world. Why not? A simple explanation is that labor markets are competitive, and because firms must offer a worker the welfare he gets in his outside option, the healthy and sick employees will sort into two firms and avoid the adverse selection altogether.
I would add the following:
the role of theory is that it helps one to organize thoughts about any particular (empirical) issue. even in cases where the original theory was not intended for.
I’m not very convinced about the role of using theory to predict something, rather an ever advancing body of theory helps us generate more generalizable ways of understanding phenomena, so that we will know what we are looking at, even if we do not know how the whole thing will play out.
and phenomena are always empirical in nature! I think the connection between theory and empirics is vastly closer than what most people imagine.
ed leamer in his review of tom friedman’s “the world is flat” tells it best: “all (mathematical) models in economics are metaphors…and the hardest thing about a new language (i.e. the mathematics) is understanding the metaphors”. sometimes internal consistency (this relates to point number 2 you write here) isn’t necessarily an asset to be protected at all costs…