F.A: It's not uncommon that researchers in experimental economics have backgrounds in other fields, obviously you have a background in game theory, and moreover far from all have pursued a consistent career in the field. Why did you start in experimental research?
W.G: The decisive reason was that I wanted to learn game theory. I was interested in experimental economics, but I thought "maybe, this is nothing for me." Later, the same thing happened with evolution. Then we had a series of local conferences where game theorists and social psychologists met. So, there was a long tradition of experimental research going back to the late 50's. When I came back from one of those conferences in 1977 I had 1000 marks in my bag for running experiments. Actually, in the winter semester of 1977-78, which was my first semester as a professor in Cologne, we ran ultimatum experiments, and also Vickrey-auction-type experiments; it took us four years to publish the results, but ever since I have been involved in experiments. I don't have a life-long plan of running so and so many experiments, of doing so much theory, and so on. When I was invited to this conference two years ago I knew I had to offer something, and this demand effect is pretty much how it works.
F.A: Today quite a few economic graduate students start directly with experimental economics. What are your thoughts about this development?
W.G: A good background in theory is not bad, of course. It goes both ways, actually. Many experiments are theoretically inspired, and vice versa. In order to do experiments they must know statistics, and they must be trained to solve software problems; they really need to be quite sophisticated. In fact, one thing which is nice is that experiments can be taught differently. I teach experimental economics completely differently compared to other things, I teach it in an "apprentice" fashion. I use learning-by-doing in small groups from the start. This is a very nice aspect of experimental economics, actually.
H.H: Experimentally oriented economists have often imported methods and knowledge from psychology. However few economists would argue that experimental economics is a sub-discipline of psychology. To what extent is there a marked division line between experimental economics and psychology?
W.G: The disciplines become increasingly close. To some extent, they are preoccupied with the same problems. Also, many of the paradigms in psychology are of a game theoretic structure. Sometimes we have a lot to learn. For instance, recently I ran experiments in Bielefeld dealing with the game theoretic concept of mixed strategies. A psychologist (Yaakov Kareev) pointed out that it is not easy for people to randomise. In fact people tend to "over-randomise" - change too often. When people try to randomise there are too few long sequences, like head-head-head; his idea is that we have a short-term memory of about seven draws and people try to balance heads and tails within such short subsequences. In fact, I don't see the dividing line; I was once the president of an economic psychology association. More generally, I think game theory is social science; what we call experimental economics, could as well be called experimental social science. I consider myself a social scientist.
H.H: You indicate that there is a substantial intersection between experimental economics and psychology, but is there something that the intersection does not contain?
W.G: For instance, I personally would not like to engage in a questionnaire study. If I collect data, I will always provide monetary incentives; I am satisfied with my evidence only if there are significant monetary incentives backing them. Another major difference is that we have slightly other ethics, for instance there is a great reluctance to be untruthful to subjects in experimental economics, whereas in psychology they would simply say "you have to debrief them," that means you have to tell them afterwards. But there are many psychologists who follow the same practice as I do. I don't see a clear dividing line. I am more for crossing dividing lines than drawing them.
F.A: There have been instances recently of theory being developed from experiments. This seems to be a new phenomenon in the sense that previous theoretical breakthroughs have been initiated by other sources. What is your opinion about the virtue of this development and do you think it will change the economics discipline fundamentally?
W.G: You have to be very clear on what kind of new theories have been developed. There are so- called new equilibrium ideas. I refer to this as a neoclassical repair shop. Starting point is that orthodox theory - meaning that you apply the common laws of rationality to material incentives and derive a solution - is strongly rejected in many instances, for instance in case of the ultimatum game. And then, what those people will do is to somewhat repair the game-theoretic presentation; they have richer utility functions, but they would never question rationality. The same can be said about psychological game theory: there is no questioning of rationality, what people do is that they somewhat change the game-theoretic presentation so that the game-theoretically rational behaviour moves somewhat closer to the observed behaviour. And that I call neoclassical repairs. Something can be learnt from this. What can be learnt is that maybe one natural type of argument like for instance inequality aversion, can account for many experimental effects. That you can learn from it, but it's never the true explanation.
A completely different story would be something where you'd say: "OK, in light of the experimental data, I now really try to develop a theory about how people come up with their decisions." Let me give you one example: I was always very much inspired by the study of Roth and Malouf. They had two people sharing a unit pie - but the pie was probability. Either one or the other wins a monetary prize, and what the two can bargain about is with what probability each wins. And what was varied in the experiment was how much the parties knew about the monetary prize which they could win. Once nothing was commonly known, each knew their own monetary prize, but not that of the opponent, and vice versa. The usual tendency was to split the probability equally, so each won the prize with probability one half. When those monetary prizes became common knowledge and were different, however, they tended to choose probabilities giving the same monetary expectation to each. So what I read from this, and what I see now everywhere, is that we have a hierarchy of equity standards; in the case at hand monetary expectation would rank higher than winning probability. There is hope for developing a more sophisticated type of equity theory here.
H.H: Some theories have at certain times found themselves in some kind of embarrassment of riches by in some sense delivering too much. This is said about the IO literature of the 80's. Do you have any concerns that this is a problem in experimental economics?
W.G: Yes, I guess we might get the same. In American courts, they consult experts to provide evidence. And for some time they used experimental evidence. For instance in these new frequency auctions, they also used experimental evidence sometimes to argue that this institutional rule is better than the other one. I guess we know how to frame an experiment to get the right data. Maybe somebody wants a certain type of results, and pays me generously for this, and maybe I would know some ways to make it more likely to get this type of result, in a clean type of experiment. I just want to say this because what I think is that we will get evidence for all kinds of effects. There is a risk that the same thing will happen as in IO, that a second type of folk theorem applies; you could say "OK, give me any type of behaviour, I can define a game predicting this behaviour and even run an experiment that just explains this kind of absurdity". There may be something like this.
H.H: Even small differences in experimental design tend to produce differences in results and it is not uncommon that experimentalists use drastic phrases like "the devil is in the details": With this in mind to what extent and in what way do you think it is possible to make generalizations based on experimental results?
W.G: Let me give one idea: Nobody would translate the reactivity numbers you observe in an experiment into real life. But qualitatively you can infer, because people are not brainwashed when they participate in experiments. When participants go into the lab, and they stay there for two or three hours, they can leave it with fifty to one hundred marks. And if you go to shop in a supermarket this is similarly relevant. One artificiality is of course that certain experiments cannot really be used for normal people because they have a quite demanding analytic structure. They could be, for instance, of the game-theory type or something like this. Think about a second, third or fourth-price auction for example, which are very difficult to explain to people; they might understand the English auction, but sealed bids might be very difficult for them. And so sometimes you need students to understand the rules. We sometimes have gone to high schools for participants. Psychologists quite often go to consumers or households. They sometimes go to households and ask questions over three months. But in general, no one is so naive to translate the reactivity numbers, the parameters that you observe directly, but qualitatively you learn a lot.
H.H: If you were to mention three robust non-trivial findings in experimental research, what findings would you suggest? Would your suggestions be the same to "Joe on the street" as to economists?
W.G: That's really difficult. I see a lot of fairness and equity aspects in certain types of situations, but there are other places where you don't observe fairness, in market places for instance. I think you have to defend theories by restricting their area of application. Equity theory is something about say, joint ventures, and there it's rather robust. But as I said, fairness is not the only motivation we have. It has to fight with competing motivations, and in some situations it's a dominating force and in others it may be dominated by other motivations. That would be one thing. The other thing is that decisions and evaluations have to be derived. They are not readily available. Even some psychological theories, like prospect theory, do not answer how I have to make a decision. But if it's really a difficult decision - like "should I marry girl A or girl B?" - what's really important? What should matter? In economic theory decision making doesn't make any sense, it's just exercise. So that leads to the need to develop the concept of bounded rationality. What else? Of course one important finding is that losses matter more than gains. Going down is something what we don't like very much. We are able to do it, but it is painful. These are three. Also "Joe on the street" is an economist, all people are.
F.A: From time to time it's argued that certain theories contain vague concepts that are very tricky to control in empirical tests. Sometimes such concepts are less respectfully denoted as theories garbage cans, since by referring to them as anomalies they can be made consistent with theory. According to your views, are there any garbage cans in economic theory?
W.G: There is something like risk attitude for instance. But in economic theory it's just the shape of the utility curve for money, and this is not rich enough to express the concerns which make us sometimes shy away from risk and sometimes search for risk. In actual life people sometimes buy insurance for very stupid minor damages, and on the other hand, people spend a lot on gambling and so on. So, I don't know. Sometimes you learn - I wouldn't call it garbage cans. I also would never refer to anomalies; what the notion of an anomaly says is that we have the norm of rational behaviour and you somewhat deviate from this norm. But actual behaviour is not just one step away from this kind of norm - it's a universe of steps away. But, of course, we often learn something. Consider the endowment effect; there could be a more basic thing behind it. Maybe the endowment effect is somewhat related to the idea that losses matter more. I wouldn't claim that those effects are simple anomalies but I learn a lot from the observation, and by framing them as anomalies they are easily understood; it serves a pedagogical purpose. Thaler, for instance, became quite popular because of his series of anomalies, but in general I think I really want to explain things rather than referring to them as anomalies.
H.H: I think many economists and game theorists associate your name with ultimatum games and you have done quite a few ultimatum game experiments since the beginning of the 80's. In 1995 you reflected on this in an article in JEBO. Now, five years later, what main conclusions can you draw from this research?
W.G: Yes, it was the first experiment I ran and there was a time when I was afraid that people would say "he always runs this," but I have started doing it very actively again. And one reason for doing it is that I know much data around this paradigm with certain institutional details varying - for instance, we have considered selling the degree of veto power. So, I'm sometimes interested in other aspects, for instance in why people veto. And I do it quite often by doing something close to the ultimatum game because I have data to compare it with. So partly this is just a working-horse which lends itself to studying new types of questions. My motivation for the basic ultimatum game was that I wanted to study the simplest bargaining situation; actually I wanted to develop a kind of bargaining theory. That was my first idea when I got these 1000 marks.
You have to see that like in game theory, experimental economics also started with characteristic-function type of work. There were many characteristic-function type experiments, and I was a bit scared by these because you just have the characteristic function. Then you have to make sure how this is really run as an experiment since for a cooperative game, it is not clear how they are going to really do it. You need instructions telling them how to announce a coalition and how to announce the payoff distribution, and so on. So you have all the procedure details and this became quite complex since subjects confronted each other face to face with possibilities of free communication. I was a bit scared by those things. So that resulted in studying the simplest kind of situation, and to have a more strategic kind of situation with very exact rules. And I hoped that first you do this and then you slowly proceed with more complicated games; actually, there is a more complicated game in that 1982 paper but nobody reads the second part. And as you see, you got stuck with the original situation because it's game theoretically simple, but apparently it's psychologically very rich. That we have to learn. Other people learn other things from it, they would say for instance, "OK, it was a standard example to show that game theory doesn't predict well." But that was not my main intention, because that would have been overkilling an already dead man. I knew that game theory is not good in explaining behaviour. So that was not the main motivation. But, of course, that may have been the primary reason for why it became so famous. I heard that Robert Aumann often talked about this kind of game, so others also propagated the idea. It came to me as a surprise, but people just caught on and talked about these results, sometimes asking "are those students in Cologne stupid?".
H.H: You mention (in the paper you will present at this symposium) that one reaction to experimental results that seems hard to explain by the rational choice approach is that experimental hypotheses about subjects' preferences do not truly reflect the real ones. In your paper you suggest another explanation; by substituting forward looking by backward looking or adaptive behavior. Could you develop your thoughts on this theme.
W.G: The main motivation for this kind of robust learning experiments is that I was sometimes attacked in the past, especially because of my work on ultimatum games, along the lines: "OK, he rejects rationality, but does he really offer something else? No it's too difficult." I try to, but the answers are incomplete. Somewhat surprisingly, those people who attacked me because I questioned rationality, now completely deny any kind of rationality. Because in evolution you don't have any cognitive requirements. Some learning theories also have very low cognitive requirements. Sometimes - for instance in stimulus-response learning - you simply would say: "OK, what was good in the past has some reason to be good in the future." That's the only cognitive demand which you have there. So, I said that the truth must be in the middle. We learn from the past, but it's not that we don't deliberate future consequences. First of all I want to show this, and secondly I want to collect stylised facts to get some guidance how I can separate and combine learning from past experience and forward-looking deliberation. And it's a very modest aim: let's get some stylised facts before we rush to develop theories. One thing is, of course, that it depends very much on the type of situation - when social norms are very strong, there's very little learning because people know what to do. If, however, the situation is complex then there is need of adaptation. But there is always, nearly always, consistent evidence of forward-looking deliberation, and we tried to capture this by: "OK, it's a learning experiment where you do not always play the same game." It is the same decision format, but the rules change. And then, if you tell them "now you play another game," then you always see jumps in the decision behaviour. We see that as evidence for forward looking deliberation. The truth is in the middle: behavioural adaptation is neither non-cognitive learning, nor is it complete rationality; i.e. rational anticipation of rule changes; it's in between. And again it's very difficult to organise those things.
F.A: There seems to be some disagreement regarding the extent to which institutional aspects of specific applications should be taken into the laboratory. What are your thoughts on that?
W.G: Why shouldn't we allow everybody to do it according to his taste? People are interested in different things. I'm not interested so much in economic policy, but if somebody is interested in economic policy he might, for instance, take a macro model and build something on top of it which he carries into the lab, this is now possible. You could have a large econometric model and define a game on top of it - for instance labour negotiations. Some aspects are of course difficult to take into the lab. Once we did a marriage-formation experiment. Of course, we clearly say that we don't capture the romantic aspect. Actually, we refer to it as a joint venture, so certain things are clearly more difficult to get into the lab. And we have an overlapping-generations type of experiment with multiple families, so that you can study whether tax systems and social security systems crowd out solidarity within the family or not. Again, these are not real families. But maybe the economic models about those things also don't have the romantic aspects. You just look at the structural aspects, how they feed into behaviour. Certain institutions are clearly difficult to put into an experiment. But if one is ingenious enough, maybe one has new ideas how to do it.
H.H: It has been held that while most research in medicine teach us a lot about white mice, sceptical economists sometimes hold that experimental research teach us a lot about undergraduate students. To what extent do you think that this criticism is fair?
W.G: As I said, you need some training to understand the kind of models we are interested in, and for those things an undergraduate student is quite good. I would love to have more students of mathematics, for instance. They usually capture the situation quite easily because they have an analytic talent. That restricts us somewhat. If the situation is complicated, it might take a long time for normal people to capture the situation. There have been these high-stakes experiments where people would travel to a poor country with a 40 dollar pie. Many results from studies of undergraduate students are rather robust. Especially the ultimatum results are rather robust; other things could be culturally or country dependent. But you have to ask yourselves why should the cognitive and emotional system of undergraduate students be so different?
F.A: Is there any topic in experimental economics or decision making that you consider to be relatively exhausted and for which the marginal benefit of additional studies is small?
W.G: People might say, for instance, that there are enough ultimatum bargaining experiments - just to attack myself, because I don't want to attack others. And there is of course an almost infinite sequence of prisoners' dilemma type of experiments. But again, people might vary the scenario and I do so. Sometimes these are just paradigms which you use to ask something else. There are always additional questions related to this.
F.A: Presumably, you will have a more positive answer on the next question, which is if you think that there are still any large territories unexplored in the experimental field.
W.G: I would like to know those territories, I would enter them as a pioneer. Maybe there are large territories, but if it would look promising, I would immediately jump to that. I guess we are all trying to enter the hall of fame.
H.H: Do you see any particular pitfalls in experimental economics in the future.
W.G: I wonder where all these brilliant young colleagues will find a job. There is so much going on because so many people run experiments. It's a big field now, but there will be some frustration. So I sometimes encourage my younger colleagues to have some additional special field, say law and economics, or import experimental economics to business administration. Experimental economics could have a better future in business administration. They have much more funds, and sometimes they also know much more details. Maybe we could import our ideas in the medical sciences. For instance, I have seen work on an algorithm for the allocation of kidney transplants.
H.H: Then I have the last question here and that is what you think will happen with the market share of experimental research in the future in the economics discipline.
W.G: I don't know, but in the near future it will be stable, maybe even growing, simply because we attract very brilliant young scholars. But otherwise I don't make any predictions. It's also not necessary. People develop how they develop.