The full subtitle is “A Sober (But Hopeful) Appraisal” and its an article just published in the American Economic Journal:  Microeconomics by Douglas Bernheim.  The link is here (sorry its gated, I can’t find a free version.)  Bernheim is the ideal author for such a critical review because he has one toe in but nine toes out of the emerging field of neuroeconomics.  For the uninitiated, neuroeconomics is a rapidly growing but somewhat controversial subfield which aims to use brain science to enrich and inform traditional economic methodology.

The paper is quite comprehensive and worth a read.  Also, check out the accompanying commentary by Gul-Pesendorfer, Rustichini, and Sobel. I may blog some more on it later, but today I want to say something about using neural data for normative economics.  That’s a jargony way to say that some neuroeconomists see the potential for a way to use brain data to measure happiness (or whatever form of well-being economic policy is supposed to be creating.)  If we can measure happiness, we can design better policies to achieve it.

Bernheim comes close to the critique I will spell out but goes in another direction when he discusses the identification problem of mapping neural observations to subjective well-being.  I think there is a problem that cuts even deeper.

Suppose we can make perfect measurements of neural states and we want to say which states indicate that the subject is happy.  How would we do that?  Since neural states don’t come ready-made with labels, we need some independent measurement of well-being to correlate with.  That is, we have to ask the subject.  Let’s assume we make sufficiently many observations coupled with “are you happy now?” questions to identify exactly the happy states.  What will we have accomplished then?

We will simply have catalogued and translated subjective welfare statements.  And using this catalogue adds nothing new.  Indeed if we later measure the subject’s neural state and after consulting the catalogue determine that he is happy, we will have done nothing more than recall that the last time he was in this state he told us he was happy.  We could have saved the effort and just asked him again.

More generally, any way of relating neural data to well-being presupposes a pre-existing means of measuring well-being.  Constructing a catalogue of correlations between these would only be useful if subsequently it were less costly to use neural measurements than the pre-existing method.  It’s hard to imagine what could be more costly than phsyically reading the state of your brain.