Potentials of neuroeconomic analysis
The Ultimatum Game
A critical discussion about the potentials
References
The emerging field of neuroeconomic research can be classified as a subfield of Behavioural Economics(Camerer, 2007, p.C26). Behavioural Economics is based on the premise that more accurate assumptions underlying economic models lead to better predictions. The way that neuroeconomics may lead to more accurate assumptions is by adding a mechanistic theory of choice to the existing mathematical and behavioural theories of choice(Camerer,2008,p.369 –370). These existing theories do not contain what the mechanistic theory might provide, i.e. how the brain solves economic decision tasks or, in other words, how the decision process influences the decision. In contrast, the existing theories treat the brain as a black box(Camerer et al.,2005,p.9). The mathematical theory of choice is about axiomatic foundations. The behavioural theory of choice is the revealed preference approach: One observes what people choose in one situation and deduces their preferences and what they will choose in future decisions under the assumption that their behaviour is consistent(Binmore, 2007, p.111)
Neuroeconomic analysis might present evidence that preferences are not the starting point of the decision process(as the revealed preference framework assumes) but that preferences and therefore decisions depend on internal states that neuroeconomics might reveal(Camerer,2006,p. 31 – 32). Results of neuroeconomic experiments that deal with the trust game, conducted by Zak et al.(2004), can be interpreted as an example of the state – dependency of preferences: if individuals receive oxytocin, they change their behaviour in the way that they give more money to the second player than otherwise. The suggestion that this data(like the activaion of certain brain areas) is important for making decisions leads to the fact that neuroeconomics seeks to use this “non-choice data” to explain behaviour. This is in contrast to existing approaches where only choice evidence is used (Gul/Pesendorfer, 2005, p.6).
Using neuroeconomic research, mainly three goals might be reached. First, it might become possible to test economic models and develop new ones with brain data(Camerer,2008,p. 372). This testing can surely be done with new developped models that are based upon brain data. Moreover, it could be possible to test whether models that make assumptions about the decision process are compatible with new findings. One example for such a model is the Dual – Process – Model formulated by Kahnemann(2002) which assumes that behaviour is caused by an interaction of reasoning and intuition. While some authors argue that models should only be judged by choice evidence, as will be discussed in the last section of this paper, Camerer(2008, p.372) reasons that it could not be wrong to check if models also make sense comparing them with brain data. It might also be possible to discriminate between models using brain data, if there is more than one model that tries to explain behavioural results(Rustichini, 2005, p.203). If standard economic models make assumptions about brain data is crucial for the potential of testing them with brain data. This question will be further discussed in the last section of this paper.
Second, neuroeconomics might be able to inform the researcher why observed anomalies exist. A researcher has only the possibilty to observe the behaviour of participants and interpret it or to rely on their reports(Camerer et al.,2004,p.573). Neuroeconomics might provide more objective details about behaviour in experiments. Third, behavioural models are build on the mentioned observation and analysis of behaviour. This knowledge is then used to explain behaviour with these models. This circular reasoning is often criticized and might be avoided with the more objective data that neuroeconomics provides(Kenning/Plassmann, 2005, p.343)
The remainder of this paper is structured as follows: In the next section, a neuroeconomic research about the Ultimatum Game is presented as an example of how the mentioned potentials might be accomplished. In the last section, these potentials will be critically evaluated.
Sanfey et al.(2003) test the neural activity of human beings while playing the Ultimatum Game. The Ultimatum Game is a sequential – move game with two players. The first mover(“proposer”) proposes a split of an amount of money(in this task 10 $) between the two players. The second mover(“responder”) decides whether to accept or reject the offer. If he accepts, both players get the proposed split. If he rejects, both get nothing. Because standard economic theory assumes that players are rational maximizers who only care about their own payoff, backward induction leads to the solution that the proposer should offer the smallest possible amount(in this task 1 $) because the responder will accept every offer that raises his own payoff. Experimental evidence in contrast shows that responders often reject offers that they consider to be too low and proposers make higher offers than in the rollback equilibrium(Oosterbeek et al., 2004, p.171)
Sanfey et al. use fMRI(functional Magnetic Resonance Imaging) which is an instrument to visualize brain activity. While this is a good technique to measure where brain activity occurs, it has a poor temporal resolution(Camerer et al.,2005, p.12). This fact might be important because of the experimental design used in the presented paper. The aim of this research was to identify the role of affective and cognitive processes in decision – making. The use of the brain scanner can be justified by the fact that affective and cognitive processes can be distinguished by where they occur in the brain(Camerer et al., 2005, p.17). In contrast, although it is not clear if standard economic models make these assumptions, utility maximization in standard economic models can be seen as a process in which the decision – maker is deliberately comparing costs and benefits which means that these models are supposed to assume that decision – making is done by cognitive processes alone(Camerer et al., 2005, p. 10).
In this experiment, 19 participants were in the role of the responder.Each subject played 30 rounds: They were told that they were going to play 10 rounds with a human partner, 10 with a computer partner and 10 rounds which were desigend as a control task so that people just received money by pressing a button. But in fact, participants were deceived and did not play against a human partner. The offers that participants thought were coming from a human partner were the same as the offers coming from the computer and followed a predetermined algorithm: each subject saw 5 fair(5$:5$) and 5 unfair offers(one time 7$:3$ and twice 8$:2$ and 9$:1$). This deception might have influenced the behaviour of the participants and therefore the decision(Zak, 2004, p.1745). The control task is used to perform the so – called “subtraction method”: To ensure the significance of brain activity in the task, the difference between the brain activity in the experimental task and in the control task is determined(Zak,2004, p.1739). One round took 36 seconds; in one round, there were 12 seconds for the subjects to decide whether to accept or to reject.
All fair offers were accepted, while acceptance rates decreased as offers got more unfair. Hence, the behavioural results were similar to former experiments. There were significantly more rejections involving human partners than computers. Comparing fair and unfair offers from human partners, responders showed a greater activation of the bilateral anterior insula(BAI), the dorsolateral prefrontal cortex(DLPFC) and the anterior cingulate cortex(ACC) for unfair offers. There was also found a greater activation of these areas for unfair offers from human parterns than for computer offers. Moreover, for offers from human partners a higher degree of unfairness led to a higher activation. Neuroeconomists can use informations from neuroscience about these activated areas in order to get informed about the brain process that occurs when individuals make decisions. As Sanfey et al. note, the BAI is linked to negative affective states such as pain or distress. This leads to the conclusion that strong negative feelings play a role when receiving an unfair offer. Furthermore, the correlation coefficient between the activation of the BAI and the acceptance rate of a participant can be measured, leading to the statistically significant conclusion that a higher activation of the BAI implies a lower acceptance rate. The DLPFC is seen to be important for cognitive processes such as goal maintenance. This activation can be explained by assuming that the individual wants to accumulate as much money as possible. But receiving unfair offers, the individual needs much more cognitive power to overcome the negative affective states. Looking across participants, it results that offers that have been rejected have a higher BAI activation than a DLPFC activation, while for accepted offers the opposite is true. This finding resembles a main principle about brain processes that Camerer et al.(2005, p.28) refer to as “Competition”: Behaviour is caused by an interplay of different systems which often shift it into conflicting directions. This can be seen here in the interaction between controlled and affective processes. The third region that is found to be activated while receiving unfair offers from human partners is the ACC and can therefore be interpreted to reflect the conflict between the two systems.
Besides the general criticisms of neuroeconomics presented in the next section, there are at least three potential criticisms that can be voiced. First, the short time that participants were able to choose whether to accept or reject the offer might cause inaccuracies due to the poor temporal resolution of the fMRI scanner. The possible serial correlation of brain firing rates within one round mentioned in the next section might also contribute to this. Camerer(2006,p.28) notes that neuroscientific instruments can be seen as complements. To validate the results presented here it might be necessary to derive the same results with other, complementary instruments. Van Wout et al.(2006) conducted a similar experiment, but using skin conductance which can be interpreted as an automatic measure of affective arousal. They found that skin conductance is significantly postively correlated with the rejection rates of unfair human offers. This fact provides a further evidence for the role of affective processes in decision – making(although it is worthy of discussion if skin conductance is a reliable tool). Second, because of the short time to decide, there might be some cognitive overload, biasing the decision towards rejection since the DLPFC has not enough power to overcome the emotional tendencies. With longer time to think about their decisions, subjects might be more aware about the fact that they are able to gain money if they do not reject. As Camerer et al.(2005, p.40) note(in association with a discussion about intertemporal substitution) the prefrontal cortex(to which the DLPFC belongs) can be suppressed by states like stress and sleeplessness. Because the presented experiment showed that the decision whether to reject or accept depends on the relative strength of affect and cognition, future research in this area could head for the question if the decisions might change if this relative strength changes, probably because of cognitive overload. Third, there is no comment about monetary incentives in the paper published by Sanfey et al.
The presented research might be able to reach the mentioned potentials of neuroeconomic analysis in more than one way: First, it is able to neurally explain experimental observations. Second, economic models can be developed containing the derived interactions between affective and cognitive processes or can be tested if already developed. Third, circular reasoning might be avoided by having a more objective method to investigate behaviour. As a result of this experiment, it seems that the activation of certain brain areas is able to predict behaviour. If this is really true, assumptions about the influence of brain activation can raise the explanatory power of economic models. According to the premise of Behaviorual Economics, then this assumption should not be neglected.
There are many authors who doubt that neuroeconomics can fulfill the mentioned potentials. First, when it is about using neuroeconomics to test standard economic models, Gul and Pesendorfer(2005, p.1 – 2 ) claim that the non – choice data coming from neuroeconomics is not helpful for this purpose, because standard economic models make no assumptions about non – choice data. Therefore the accuracy of standard economic models can only be judged by choice behaviour. But testing standard economic models, one needs at least to assume that people choose consistently. For example, if someone wants to test if people violate Expected Utility Theory, he needs to be sure that people go on choosing according to their revealed utility function. This could mean that neuroeconomics can show that the revealed preference approach has to be widened by taking into account the internal states which people are in. This is similar to what Harrison(2008, p.322 – 328) argues. He shows that the revealed preference approach predicts with probability 0 or 1. This also means that a decision – makers mistake could disprove the whole approach. Harrison argues that there should be an approach(which maybe comes from neuroeconomics) that makes it possibe that consistency is not given with a positve probability.
A further critisism coming from Harrison(2008, p. 311 – 316) is about the reliability of neuroeconomic data. He mentions many facts about possible problems of reliability. Two of them should be further explained here. First, as Kenning and Plassmann(2005,p.346) note, brains are like “fingerprints” which means that they are not equal across people. Studying brains involves the need to normalize them. Harrison argues that the reliability of neuroeconomic research is questionable because the method with which this is done matters for statistical inference, so that significant results might become insignificant if someone uses another method. Second, it might be possible that firing rates, which are measured by a brain scanner, are serially correlated. This can be the case within one round of the experiment, but also if participants play more than one round as presented in the research of Sanfey et al. For example, this may be the case because of learning effects. It is problematic that Sanfey et al. as well as other authors using this method do not use statistical tools to examine this and to solve this problem if necessary.
One main problem associated with neuroeconomic research is the fact that neuroeconomics has so far tried to find neural correlates to existing behavioural approaches. The research of Sanfey et al. presented in the previous section is a good example of this critique (Spiegler,2008, p.515). This corresponds to the view of Rubinstein(2008,p.487) that neuroeconomics has not yet presented evidence for new relevant economic concepts. Despite the possible,discussible usefulness of neuroeconomics to inform about anamolies and test models, economists are skeptical if they should invest scarce ressources into this new form of research, because explaining existing anomalies does not help to develop new models. Economic models should make predictions about what people will choose in (new) situations and not explain why they choose one thing over another. So neuroeconomics seems to be not useful in developing new models unless it is possible to derive some new results with the help of neuroeconomics or it can be shown that internal states are really able to predict behaviour. If predictions can be made with the help of internal states like the activation of brain areas, there might be another problem because models might get to complicated if someone tries to map all neural constructs underlying decisions. While this is a question of modeling conventions, from a technical point of view it should be seen positive that the instruments used by neuroeconomics are expected to improve in the near future.
[...]
