Networks of neuroeconomics

What is neuroeconomics? I tried to answer this nagging question in a quantitative and visual way:

What do these big blobs mean?

The big, green one stands for all the journals published in neuroscience. The strings connecting this blob to some other blobs in the graph represent citations. The size of the blob represents the proportion of papers in neuroeconomics published in this discipline, according to a dataset spanning 1999-2009.

Not surprisingly, neurosciences is the biggest one. But, where is economics for example, and which size is it? Ok, I hope this was teasing enough (to neuroeconomists at least…): this is explained in much more details on a presentation I gave a few days ago, the slides of which you can find there. Note: if you are a neuroeconomist, a scientometrician, or a social network analyst, I am particularly interested in your feedback! Post a comment on this blog entry!

NB: I also posted this blog entry earlier today on http://historyofeconomics.wordpress.com/, with marginal modifications.

I am currently reading a fascinating book, “The Cognitive Revolution in Psychology” by Bernard J. Baars (1986).

With a long introduction, it provides informative material for an outsider like me on how the cognitive turn played out in psychology, and presents a clear historical background getting back to Wundt and the early experimentalists, and the origins of the behaviorist revolution. Then it is followed by a series of interviews of participants in the cognitive revolution: from the opponents (Skinner and others) to the enthusiasts, and the followers.

On the substance, I was stroke by how much behaviorism, which is the methodological orthodoxy that was overthrown by cognitive psychology, shares features with today’s textbook economics. Both share the status of a well-guarded orthodoxy: in their interviews, psychologists remember that behaviorism in psychology was exclusive, displaying a “nothing but” attitude: variables should be related to nothing but observable behavior, which disqualified the discussion of concepts like “memory” or “representations” ! Those words were taboo in psychology at least until the mid-1950s.  Looking back, psychologists consider that the methodological rigorousness of behaviorism, which insisted that each concept be operationally defined and testable, had the effect to strip psychology from its substance: the study of cognition, consciousness,  emotions and rational behavior were discouraged, virtually banned indeed, because these concepts did not readily translate into tightly defined behavioral variables that could be observed in an experimental setting.

I could not help but be reminded of a similar taboo in today’s economics, where the formation of preferences, or how the process of choice unfolds, is declared “out of bound” right from the introductory chapters in microeconomic textbooks: only an individual’s observable behavior, as it is instantiated in the outcome of the choice it performs, is to be taken into account.

Reading this book, neuroeconomists will also be strongly reminded of Princeton economists Gul and Pesendofer’ essay published in 2008, in which they defend a “nothing but” approach to the revealed preference approach – dismissing any kind of evidence from “inside the head”, and advocating bluntly a “mindless economics”. Behaviorists (or operationalists…) of the purest ink!

The cognitive revolution in psychology crystallized around the mid-1950s, early 1960s. Forty years later, nothing of that sort happened in economics, it seems to me. With behavioral economics and neuroeconomics, maybe that economics will jump directly to the next train: the neurocognitive turn. Or will it miss that one also?

Post-script: on an approaching topic, Wade Hands has a paper forthcoming in the CJE, which is a nice read.

This coming Saturday I will start a three-week visit to a university with teams performing research in neuroeconomics. The purpose of this visit is to perform an “ethnography” of neuroeconomics, focused on how interdisciplinarity works in practice. Among other purposes, this study will provide me with qualitative insights which will complement one of the other projects I am currently running on the bibliometric study of neuroeconomics.

A bibliometric study can be many things, in this case I am interested in how publications in neuroeconomics reflect its interdisciplinary nature. Online databases such as ISI Thomson help a great deal in performing such a study, and the remaining difficulties are probably of the conceptual sort (see the previous post!).

The results of those bibliometric studies are most commonly represented in graphs of social networks: they help read a tonne of information in just one picture.  I am currently training myself to use them, getting gray hair at pre-processing bibliometric files which then can be fed into those softwares…

This is how a social network can look like in practice (click on the pic to expand):

A co-citation network - green and yellow nodes indicate those papers citing heavily the original set

Opinions about neuroeconomics vary enormously – to begin with, there is little agreement about what even *counts* as neuroeconomics.

In my historical study of neuroeconomics,  I am confronted to this difficulty right from the first step. Before even analyzing it, what is neuroeconomics, the field that I am studying? There is yet no journal of neuroeconomics which would map and delineate the topic, and there is of course no single parent field from which this sub-field can be traced from.

Lost, but with a map

Mapping neuroeconomics

The Society for Neuroeconomics is a useful rallying point where neuroeconomists can be found, but there is an obvious North American bias. More, an unknown proportion of scientists attending this conference would be reluctant to be labeled neuroeconomists, if I refer to the interviews I conducted there. So?

There is always the possibility to ask “the experts”, as it is customary to do in scientometrics. That is, I would not placate any arbitrary definition of neuroeconomics on the field, but would ask some renowned neuroeconomists what they would consider classic papers in neuroeconomics, or who do they consider to be the leading figures in neuroeconomics, and then start from there.

The problem with this approach is that leading neuroeconomists are truly extremely busy people, so the sample of experts that I could tap from would be very low, and hence surely not representative of all the currents represented in neuroeconomics.

There are many other ways to define a field, all with their particular drawbacks. One is to refer to the indexing of papers performed by Thomson Reuters‘ ISI Web of Knowledge, a database which records and indexes virtually all peer-reviewed journals and their papers on the planet since 1988. If a paper is indexed with the keyword “neuroeconomics”, then it can count as neuroeconomics. Authors who published a certain number of articles indexed with the keyword “neuroeconomics” can be considered to be neuroeconomists. However, this approach is equivalent to delegating the task of defining neuroeconomics to the employees in charge of indexing the papers at Thomson Reuters: given the immensity of their task, probably not the best experts in neuroeconomics.

I am in the process of finding my own (and hopefully, consensual) solution to this arduous problem of mapping a field which has still not a stabilized identity. But from experience, it is an issue where everybody can quickly come up with their preferred procedure. So, what do you think? What would be your procedure to arrive at a definition of *who is a neuroeconomist*, and *what is a paper in neuroeconomics*?

Computational neuroeconomics?

Computational neuroeconomics?

Neuroeconomists develop their own jargon, as it is to be expected from a consolidating community of scientists with distinct interests. But denominations, categorical classifications, and basic concepts in neuroeco are very much still in the early stages of their definitions – they have not been “blackboxed” yet.

“Computational neuroeconomics” is one of such terms. I was a bit tired of nodding to my interlocutors when computational neuroeco popped up in interviews, without being sure to understand how different it was from “not computational” neuroeco, or from computational modeling in cognitive neuroscience.

A first possibility was that it could resemble this class of models where connectivity of different brain regions is represented by an analogy with the architecture of a computer.  This is the kind of model used by Peter Dayan and Szabolcs Kali in a paper in 2004, who discussed memory storage and retrieval.

It could have also been the models inspired not by computer hardware, but by softwares: algorithm processes which demonstrate that starting with very simple building blocks and logical rules, an organism could  achieve complex cognitive tasks like letter recognition and other sensory to motor tasks.

But computational neuroeconomics seems in fact to represent an alternative, third possibility.  An entire session was devoted to it in the third day of the annual meeting of the Society for Neuroeconomics. It featured papers which were basically game theory applied to social cognitive problems. The language of game theory provides concepts  to think many useful parameters of behavior: strategies, payoffs, probabilities, types. As I understand it, the task of computational neuroeco is to operationalize those concepts. In the speeches of the session, it was interesting to see how the presenters navigated between mathematical sophistication, and constant references to pragmatic issues in social behavior: what theory of mind emerges from repeated interactions between players, or how risk minimization is accomplished through learning.

Is it a new approach in neuroeconomics? Not really. When one thinks about it, it is “simply” further work in a direction impulsed by Paul Glimcher and his collaborators since the very beginnings of neuroeconomics, when they introduced expected utility and then Bayesian Nash equilibrium in their studies of neurons in the LIP area for monkeys.  In this light, computational neuroeco appears to be at the very core of the new discipline.

Darts and pool

This is the second annual meeting of the SfN that I attend, and this time I am there to do interviews and publicize an online survey on interdisciplinarity which I designed for neuroeconomists (are you a neuroeconomist? Drop me an email at clevallois@rsm.nl, and I will send you the link to it).

The program is remarkably different from the last year. Much less rat studies, and a lot of papers and posters on social interactions in humans. I am not sure whether it reflects an inflexion in the selection process by the organizing committee, or a new direction in neuroeconomics. A participant at the diner hinted that it merely reflects the changes in priorities laid out by funding agencies.

A few labs are overrepresented – Duke and CalTech; and the usual big names are all around. I could interview Peter Bossaerts, Colin Camerer and Paul Gimcher, and I should continue this series tomorrow.

For this evening, I hesitate between catching up with some sleep and fight the jetlag, or join the conference-sponsored dart-and-pool evening at the pub around the corner. Hm.

That is big, really big.

The free-rider problem is simple. It describes those situations when a group of individuals would benefit from a common action, but each individual separately would prefer not to make any effort to make this action happen.

Like: as a group, we would like to have an environmental policy to stop global warming, but when asked how much tax I personally would be ready to pay to implement this policy, I refuse to declare that I’d be willing to pay much. Even if the amount that I would pay would be more than compensated by the benefits of an environmental policy! Simply because hey, if the environmental policy is decided by others and payed by others, once implemented it will also benefit me, so why would I bother paying for it? It is much easier to let others pay for it, and then, I’ll benefit from it anyway!


Another free rider

It is the eternal problem of the free rider: “I would like to have the benefits of the collective action, but I would prefer if the costs were payed by my neighbor.” The problem is, of course, that if everybody thinks like that, then everybody states that they would not pay much, and the budget for the collective action is never gathered. That’s too bad, because the collective good would have enhanced everybody’s welfare!

It has huge implications for tax policy, or any issue where a collective action would be required. And the difficulty faced has always been that when you ask people “how much would you be ready to pay for this collective good”, they tend to understate what they are really willing to pay – always hoping that the collective good will be build anyhow – but payed by their neighbors.

What if we could “read” in people’s mind what they would really like to pay for a collective good? This would allow to know how much people would each be ready to pay for the collective action. The collective action would then be undertaken, only if its benefits would be superior to the sum of the payments that each individual declared to be ready to make.

This is the experiment conducted by a team of CalTech neuroeconomists, just published in Science. They gathered groups of subjects, and scanned their brain while they stated the amount they were willing to pay for a given collective investment.

To be precise, this what not a simple lie-detector mechanism: the accuracy of the scan to detect the “honesty” of your choice was just 56%. But it is enough to act as a threat to participants: they will be punished with heavy taxes if they are found to be willing to pay very little for an investment which will repay them much. It acts as an incentive for participants to reveal their true preference each time they are asked about a potential collective investment!

Ian Krajbich, lead author of the study

Ian Krajbich, lead author of the study

The results of this “Neurally Informed Mechanism” as they call it are astounding: the total welfare achieved by this experiment is 93% of the ideal case, which means that free-riding has almost completely disappeared! This is a remarkable result, given that traditional experimental settings do not score better than 23%.

So, the free-rider problem, or the problem of collective action finally solved? Neuroeconomics made an interesting step in this direction. Huge!

The study raises some questions though. For example, the procedure followed shows that the participants were convinced with lengthy, technical arguments that the “not free-riding” strategy was the most advantageous one.  Those arguments were true, and the experimenters deduce that if the participants did not free-ride, it is because they understood it was in their best interest. But one can also object that they did not free-ride simply because they had been brain-washed about not free-riding: they simply trusted the experimenter and played in the fashion that was strongly suggested. If that is the case, then the 93% result is not so amazing.

This doubt is even greater when one wonders about the actual role played by the fMRI scan: crucial or not? At 56% of free-riding detection, just above fifty-fifty, one doubts whether the participants refused to free-ride because the threat of the scan detection made it the best strategy to play, or simply because they were impressed by this big machine and the intense strategical training they received from the researchers (see the supplement online material of the article, esp. the section “strategy” on pp. 42-45, which shows how much the participants were lectured about NOT free-riding. )

More about it? Antonio Rangel, an economist in the CalTech team reports on the Science article in an interviewed by James Hugues, from the Institute for Ethics and Emerging Technologies. Click here for the interview.