I think of economics as the one true modernist social science. Our reliance on mathematics and mathematical models introduced a statistical and theoretical rigor that distinguishes economics from the other social sciences. It has also tended separate the discipline from its 18th century roots in moral philosophy and it has reinforced our conceit that economics is more like physics than it is like psychology or sociology.
These are not necessarily bad things. They have provided economics with what I think is a better developed (than other social sciences, not than physics) theoretical framework from which empirical analyses can be conducted. It has resulted in mathematical and statistical rigor that should be the envy of all other social sciences. However, it has also produced a remarkable uniformity of perspective, opinion, and belief among the members of the tribe. In the last, it has come to resemble a religion rather than a science, at least for some.
Had it not been for the recent crises in mortgage and financial markets, most of the high priests and priestesses of our arcane religion would still be worshipping at the alters of rational expectations, efficient markets, and Pareto efficient equilibria; allowing a morbid fear of free-riders and moral hazard to dominate any attempts to address social issues rationally; and chanting the mantra that rational, fully-informed self-interest would save the day (even in situations where information was highly asymmetrical with little hope of remedy).
Economics has been brought down by animal spirits, fat tails, and insularity.
I believe there are several things that will haul us back from the brink of irrelevance and set us on the track to true science. The one I want to write about today is economics becoming a legitimate social science instead of a branch of mathematics. Deirdre McCloskey has written extensively about this. She and Steve Ziliak have been fighting a two-person war to move the discipline to a more enlightened approach to statistical inference for years.
I believe it’s time to refocus attention and discussion on the error term. It is often where much of the action is in our models. It is where unexpectedly catastrophic events dwell resulting in fat tails. It is where our animal spirits manifest and cause us to do the right thing or the wrong thing or the thing everyone else is doing rather than the self-interested, fully-informed rational thing. It is where God and miracles and chance dwell.
The error term is the unexplained residual. Often it contains 70-90% of the model’s variation. It may be reassuring to find that price and income matter and have the expected signs, but if the explanatory power of the model is low, there is almost certainly more we need to know before we can legitimately call ourselves scientists.
Post-modern applied economics, macro and micro, must do a better job of explaining the error term. It is not enough to assume a distribution, a corresponding functional form and then present the most likely model parameters (with 95% confidence intervals). We must investigate the error term by conducting research on what might be left over, what might not be in the model.
This is a well-known problem with economists: omitted-variables bias. It’s time for us to recognize that OVB runs deeper than finding the best instrument (which almost certainly does not exist or if it does, does not vary sufficiently both to reduce bias AND provide precise estimates).
Ironically, psychology, sociology, and anthropology, the social science disciplines we have looked down on these many years, have done much of this for us. Behavioral economists are now leading the charge, but it will require more than improved behavioral models. It will also require a new approach to statistical inference.
What does this mean? Null-hypothesis statistical testing still has value, but its value should be in supplementing a well-researched narrative. P-values and arbitrary cut-points for statistical significance should be recognized for what they are: tools for inference that must be placed in context and tempered by what else is known about the relationships. They are not hard and fast rules that “prove” anything. Observational data are “observed.” This means we have to “observe” more than the variables in the model and the supposed instrument. We must observe context, political, economic, social, and any other even when we can’t include them adequately in the model. We should question the “independent, identically distributed” assumption always. It is clear that independence was highly questionable in finance and mortgage markets. Surely, there are other examples. Finally, there are always omitted variables. In every model. They are in the error term. They should not be ignored even if they are believed to be uncorrelated with regressors. They may well be correlated across individuals.
The error term is and has always been where most of the interesting “action” is. Economic research must return to rhetoric and rhetorical discourse derived from extensive research and supplemented by a sounder approach to statistical inference in order to investigate it. Deirdre McCloskey has said something like this before me.
Is it a girl thing?
