1. THE BOTTOM LINE

Let me start with the bottom line: Neuroeconomics will remain a hot topic in economics during the coming decade, probably one of the hottest. This is not because of any truth that is waiting to be discovered or some urgent real-world problem that needs to be solved. Rather, the evolution of economics (and probably other disciplines as well) is subject to forces similar to those that dictate the emergence of any other fashion trend.

The rise of neuroeconomics is coming at a point in time when economic theory is not producing any exciting insights. The game-theoretical revolution in economics is complete – it has probably even gone too far by pushing aside other economic models. It is a time in which economists have already recognized the limits of rationality and are increasingly interested in models of Bounded Rationality. However, these models are due almost entirely to introspection and modelling convenience and as such are perceived as arbitrary. Neuroeconomics is offering us the illusion that results achieved through introspection can be supported with hard evidence. For many economists, who are obsessed with the desire to become scientists, this is an opportunity to become associated with real science.

I am not a big fan of neuroeconomics. I heard about it for the first time in the autumn of 2000 at a small conference organized at Princeton University. Even then I was assigned the role of critic. However, my attitude to neuroeconomics is more complicated than that and in some ways I actually find neuroeconomics to be appealing. I believe in judging academic work by a subjective criterion: whether or not it is interesting and neuroeconomics certainly meets that criterion. So why do I find it difficult to accept neuroeconomics? I can think of two reasons. The first is my position on the mind-body problem. I fear the approach in economics in which decision makers become machines with no souls. The second reason (discussed in Rubinstein Reference Rubinstein, Blundell, Newey and Persson2006, and Harrison Reference Harrison2008) is neuroeconomics’ style and rhetoric. Conclusions are hastily drawn on the basis of scanty data. Lack of knowledge and uncertainty are swept under the rug. Colourful diagrams, which mean nothing to economists, are presented as clear evidence. To me, they look like a marketing gimmick like those used to sell a new product in the supermarket. The statistics used are not well understood by the vast majority of the users. Brain researchers are rushing to use economic terms without fully understanding their subtleties. The field suffers from a lack of self-criticism and a reluctance to discuss “details”. In informal chats following a seminar, I hear considerable scepticism – which I didn't hear in the seminar itself. I have attended dozens of seminars and read probably a hundred papers on neuroeconomics and I almost always have the feeling of being (unintentionally) manipulated. My sense is that tough competition has led to low standards. A researcher knows that if he doesn't publish a new “discovery”, then someone else will and the stakes in a new field are very high.

However, let's put aside the problems of style, rhetoric and pretension for a moment and assume the following scenario: We are able to map all brains onto a canonical brain. The functions of the different areas of the brain are crystal clear to us. The machines used in experiments are cheap enough that thousands of subjects can be experimented on. And finally the data are clear and double-checked. The question would still remain: what is the potential role of brain studies in Economics?

When I read Neuroeconomics papers, I don't get the impression that those who make claims for Neuroeconomics know how it will change Economics (McCabe Reference McCabe2008 states this quite explicitly). Part of the problem is that many brain researchers misunderstand the meaning of economics. Imagine that brain studies were to discover that there is a centre of the brain that determines whether we prefer Bach to Stravinsky or Stravinsky to Bach and that currently 1% of the general population prefer Stravinsky. This would be a really momentous discovery. It would also be a really important piece of information for a manufacturer of Stravinsky discs, but not all neuroeconomists understand that it would have no significance for economics.

It seems that even neuroeconomics researchers with a good understanding of economics are in the dark about how it will re-shape economics. A year ago, I challenged several ardent supporters of neuroeconomics to show me even one neuroeconomics paper that is likely to change economics. I have yet to receive a satisfactory response. Nonetheless, I'm not going to argue that neuroeconomics will have no effect on economics but just that the neuroeconomics literature does not help us to imagine what that effect will be.

2. WHAT THEN CAN NEUROECONOMICS DO?

The standard way in which we model a decision maker in economics is by a choice function c that attaches to every choice problem A, in the relevant domain, a unique element c(A) ∈ A. The statement c(A) = a means that the decision maker chooses a when the set of available alternatives is A.

Some of us study a somewhat richer structure in which the choice function receives as input not only the set of alternatives A, but also additional information f, which is irrelevant to the mental preferences of the decision maker. In Salant and Rubinstein (Reference Salant and Rubinstein2008), we called such information a frame. A frame can be, for example, the ordering by which the elements in A are presented or a default alternative. The statement c(A,f) = a means that when the choice problem A is presented to the decision maker framed by f, the element chosen by the decision maker is a.

In the world of neuroeconomics, a decision maker is presented with a choice function that receives as input a choice problem A and produces as output a pair (a,x) consisting of an alternative a ∈ A and an array of parameters x. The statement c(A) = (a,x) means that the decision maker who confronts the choice problem A will choose a and will produce a vector x of indicators measured and recorded during the period between the moment he was presented with the choice set and the moment he “pushed the button to select a″ (and probably even after that).

Current research in neuroeconomics takes x to be a very long vector of numbers that represent the activities in various areas of the subject's brain during the deliberation period. But x can have other meanings as well. For example, there are those who take x to be a description of the subject's eye movements while staring at a screen where the choice problem is being presented. In order to assess neuroeconomics’ research agenda, we follow Rubinstein (Reference Rubinstein2007) and look at another, even simpler case, in which x is a single number representing the decision maker's response time.

The platform for the experiments was my didactic website http://gametheory.tau.ac.il. Thousands of students in game theory and microeconomics courses in about 30 countries have used the site, where they were confronted with decision problems and game situations. In addition to the choice made, the computer also recorded response time (RT), i.e. the time that elapsed from the moment the server sent the question until the moment the answer was received. This type of research has some obvious merits: it is cheap and facilitates the participation of thousands of subjects from a much broader than usual population. On the other hand, research of this type has a major disadvantage in that response time is a noisy variable (although so are the brain indicators measured through fMRI). Nevertheless, the huge sample size seems to smooth things out.

The basic hypothesis states that response time is an indicator of the way people think about a problem. Assume that decision makers are to choose between A and B and that their response times in choosing A are concentrated around 20 seconds, whereas the length of time it takes them to choose B is about 40 seconds. We can then conclude that the choice of A is more instinctive whereas the choice of B involves more deliberation and is more cognitive.

As an example, we will use results of an experiment based on the Allais paradox. Subjects were asked to make two hypothetical choices as presented by Kahneman and Tversky (Reference Kahneman and Tversky1979). First, they were asked to respond to Question 1:

Question 2 followed:

The following table summarizes the results.

A majority of subjects chose A and an even larger majority chose D. Those who chose A and D or B and C (49% of the subjects who answered the two questions) violated expected utility theory. Note that although in this case the two questions were presented to the subjects one after the other, they were not very different from those reported in Kahneman and Tversky (Reference Kahneman and Tversky1979) in which a group of subjects was randomly divided into two subgroups and each of the subgroups answered only one of the two questions. The Median Response Time (MRT) of the subjects who chose A in Question 1 (49 seconds) was significantly longer than that of the subjects who chose B (34 seconds) and the MRT of the subjects who in Question 2 chose the risky lottery C (31 seconds) was much longer than the MRT of those who chose D (19 seconds).

Further information regarding the subjects' response times is presented in figure 1. The graph plots the cumulative distribution of response time for each of the four choices.

Figure 1.

Distributions of RT for the choices A, B, C, D.

The fact that the response time for Question 2 is dramatically shorter than for Question 1 might be an outcome of the order in which the subjects responded to the two questions. Nevertheless, I believe that it is also due to the fact that Question 2 is simpler. There is evidence for this conjecture in the data on the small group (consisting of 86 subjects) who answered only Question 2: their median response time was 32 s which is far shorter than that of the group who answered only Question 1. This could easily be checked by reversing the order in which the questions are presented. It was not done since the site is used for pedagogical purposes and therefore I felt it would not have been correct to do so.

We observe that the choice of C requires much more time than the choice of D. The interpretation of this result is quite clear: Taking a risk requires more time for consideration. It is less clear how to interpret the differences in response time between those who chose A and those who chose B. I can conceive of two explanations for the choice of A: (i) Lottery A has a higher expectation than lottery B or (ii) Subjects view a lottery as a two-dimensional vector and when comparing between two vectors they try to “eliminate” one of the two dimensions in which the parameters are “approximately the same” and then make the decision according to the relative magnitudes of the values in the other dimension. (When comparing the pairs A = ($4000,0.2) and B = ($3000,0.25), some subjects find the probabilities to be similar and choose A after comparing the dollar amounts.) Both explanations require attention and calculations that require time.

Measuring response time does not allow us to distinguish between the two explanations. The importance of the distinction lies in the fact that if subjects who chose A maximize expected value, then the assumption of the existence of a preference relation over the space of lotteries of the type (x,p) is not rejected. On the other hand, if subjects used the second procedure, it was argued in Rubinstein (Reference Rubinstein1988) that intransitivities are likely to result.

This is a task for which neuroeconomics might be suited. In a study I am conducting with Amos Arieli of the Weizmann Institute, we track the eye movements of subjects who are facing a screen showing the following four numbers:

The detection of vertical eye movements would support the theory that subjects multiply the prize by its probability in each lottery. The detection of horizontal movements is more consistent with the “similarity” procedure.

One potentially important task for the neuroeconomics approach is to identify “types” of economic agents, namely to determine characteristics of agents that predict their behaviour in different choice problems. It is difficult to do this simply by observing behaviour. If the choices in Problem 1 and in Problem 2 were independent, then given the distributions of answers to the two problems, we would expect to obtain the following distribution of joint choices:

In the experimental results we observe only a slight (though significant) increase in the proportion of subjects who chose the combination A + C or B + D. Thus, the data on the choice on its own is not sufficient to identify types of agents. The response time data, however, can help us to do so. A point in the graph in Figure 2 stands for a subject. His coordinates (x,y) indicate that x% of the population responded faster than he did to question 1 and y% responded faster than he did to question 2.

Figure 2.

Allais Paradox TR Percentill Scatter Plot.

The correlation between the two variables is quite high (0.58). The two bold lines divide the unit square into two equally sized areas. The vast majority of subjects fall into the centre half.

What is the difference in terms of behaviour between the slow and the rapidly responding types?

In preparing the following table, the population of respondents was divided into four classes according to RT on Question 2 (simlar results are ontained when the RT for Question 1 is used as the base): very fast (the bottom 10%), fast (range of 10–50%), slow (range of 50–90%) and very slow (the top 10%). The more thought the decision maker puts into the problem, the greater the tendency for him to choose option C. While 85% of those in the lower half of the distribution chose the safe alternative D, the subjects in the slowest 10% of the distribution were divided almost equally between the two alternatives.

Another interesting result is obtained by comparing the distribution of choices among the faster half of the population to that among the slower half (as measured by RT on Question 2):

Distribution of choices among the faster half:

Distribution of choices among the slower half:

Longer response time contributes little to consistency. Whereas 49% of the fastest subjects were consistent, no more than 53% of the slowest ones were. However, the composition of the consistent subjects was dramatically different. While almost 80% of the fastest consistent subjects chose B and D, only a minority of the slowest chose this combination. Thus, here is a case in which the “neuroeconomics” data were necessary to provide support for the conjecture that subjects are divided into types (fast and slow respondents) and that there are significant differences in their behaviour.

3. THE BOTTOM LINE (ONCE AGAIN)

The grand vision of neuroeconomics is to use the additional information obtained from brain studies, combined with the choice made by the decision maker, in order to better understand the deliberation process and to use the results to improve economic models. However, it is far from being clear if and how this can be accomplished.

An analogy which I have in mind when thinking about neuroeconomics is of an intelligence agency that tracks enemy activity by continuously monitoring vehicle movements or radio signals. Discovering correlations between enemy tactics and a high level of activity in a particular area or detecting a sequence of high levels of activity at different sites allows the intelligence agency to understand the structure of the enemy's army (such as, where the command centres are located, through what channels the enemy transmits information and so on). This will enhance its understanding of the enemy's decision process. It might also allow the anticipation of the enemy's moves and the prediction of a surprise attack. Does neuroeconomics also work in this manner? Not really. Whereas an intelligence agency is interested in only one “brain”, that of the enemy, neuroeconomics is interested to understand all brains. While good intelligence provides data about the enemy in real time, current techniques (fortunately!) don't allow us to gather real-time data on decision makers (and if they did we would live in an entirely new world which presumably have also new economics).

As I demonstrated in the previous section, neuroeconomics’ approach could probably be useful in two ways: first, once we have enriched economic models of bounded rationality, it would make sense not to simply invent procedures from off the top of our heads but to use models based on our understanding of the mind (see Rubinstein Reference Rubinstein1998). In this respect, I imagine that reliable neuroeconomics data could serve a purpose by providing us with information on how widespread the use is of a particular decision-making procedure. Second, one could imagine that brain studies will allow us to identify types of individuals who share modes of behaviour for a wide range of decision scenarios. If this is indeed the case, we would be motivated to construct models in which the distribution of types is a primitive of the model. Using such models, we would probably be able to derive stronger analytical results. But, the proof is by doing and we are far from showing that any of those could yield new economic ideas.

To conclude, brain studies are of course fascinating and although I have yet to come across a single relevant insight produced by these studies it may be that they will eventually change economics. At the end of the day, neuroeconomics will probably influence economics because economics is a culture and not a science. By “culture” I mean a collection of accepted ideas and conventions that are used in our thinking. Cognitive psychology and behavioral economics have changed economics by introducing concepts and conventions that have become part of the mainstream discourse in economics. I would not be surprised if neuroeconomics infiltrates economics in a similar fashion. If, in addition, neuroeconomics makes less pretentious claims and adopts greater precision, it may become more than just an entertaining sideline.