People playing games: the human face of experimental economics.

Eckel, Catherine C.

1. Introduction

Research in economics is focused primarily on the behavior of groups of individuals aggregated into markets and economies. Economists have paid less attention to understanding individual decision making, which (by rights) is the realm of psychology. Much of economics appears blind to the individual, and this results in the neglect of issues that can be useful in understanding market phenomena such as the gender or race wage gap. Combining a psychological interest in the characteristics and propensities of individuals with an economist's perspective can give insights into important economic outcomes. Experiments, both in the lab and the field, examine the behavior of people in situations that economists are interested in. My purpose is to show how experiments, people playing games in the lab, can provide a window on individual behavior, a measure of the nature and variability of the preferences that underlie economic models, and a better understanding of puzzling economic phenomena.

Altruism and trust are two aspects of human behavior that might seem orthogonal to understanding economies. Yet they are very important factors for the success of an economy. Altruism ensures that the people in the economy who do not (or cannot) win under the market system are taken care of. Altruists also increase the production of public goods. Trust is said to grease the wheels of commerce by enhancing trade among strangers and by making complete contracts unnecessary. Because of the cost of writing and enforcing fully specified contracts, most economic exchange must operate without such contracts, and can do so successfully because of high levels of both trust and trustworthiness in the economy. The research I present today focuses on these two important areas of behavior, and the use of experiments to study them.

2. Economic Man in Residence

Economic man is a simple agent who single-mindedly maximizes his utility, and who lives in our models. He is useful as a building block in economic models, game theoretic among them. He arose, not as a way to describe actual human motivation, but rather as a simplification (as most models arise) intended to capture important aspects of behavior. Much has been made in recent years of the systematic failure of experimental subjects to behave like economic man. This challenge to the vision of man as a rational being has led to the development of many models that extend the range of predicted behaviors to include the kinds of cooperation, trust, and reciprocity that we observe in the lab. Some, indeed, have pronounced economic man dead and, dead or alive, apart from a few especially earnest graduate students in economics, surely no one really believes in him. Nevertheless, he is useful. Assumptions of rational behavior do have handy formal properties, allowing the (relatively) easy aggregation of "maximizing monads" (a term I believe is due to Deidre McCloskey) into markets and economies, and extensions of predictions about individual rational behavior into greater understanding of complex system-level phenomena.

Reliance on economic man in our models does not require that citizens be selfish, but in practice most models assume that agents care only about their own well-being. In the same vein, such reliance does not require that agents behave identically, but in practice, individuals are modeled as being identical. Just as no one really believes people are completely selfish, no one really believes that all people are the same. People are different from each other, and they treat each other differently. The differences are not random: Heterogeneity is systematic. For example, in experimental games, about a quarter of experimental subjects really do behave like economic men (and women), no matter what situation we put them into. Perhaps 20% are altruists, and behave as if the welfare of others is very important to them. The rest could be termed "conditional cooperators," cooperating or behaving altruistically when the costs are low or the benefits high, trusting when the potential gains to trust are high, reciprocating when it is the right thing to do. In a way, these conditional cooperators seem familiar to economists-they often exhibit a kind of economical cooperation. But subjects may be contingent in other ways, as well. In two settings with identical payoff structures, a subject may behave very differently depending on the context. (1) And sometimes the contingencies are things we do not approve of and teach our children not to pay attention to, superficial things such as the mere appearance of a counterpart.

Who are the economic men and women, and who the altruists? Who is trusted and who not? An exploration of the systematic differences in behavior across different categories of persons (gender, racial, etc.) and of the systematic difference in which different categories are treated by others can improve our understanding of market outcomes.

3. Measuring Altruism and Trust

To investigate heterogeneity in behavior requires reliable measurement. Self-reported survey measures of altruism, trust, or trustworthiness have several shortcomings. First, there is no incentive to report correctly, and economists are naturally skeptical of such "cheap-talk" claims. Second, there may be an incentive to misreport. To take a simple example, suppose I ask if you are altruistic or trustworthy. You may exaggerate your virtues to impress me or someone else who might be looking on, or to validate your own self-image. You may not even be consciously aware of your exaggeration, believing you would behave in a certain way. There is plenty of evidence that people over-report socially desirable behavior, such as voting, volunteering, or even exercising.

Laboratory experiments were developed for testing theory and for teaching, but that is not all they are good for. (See Holt (2003) for interesting uses of experiments.) In particular, experimental games can also be designed explicitly to measure preferences. Measuring preferences is important because so many economic models require parameterization of a utility function in order to have empirical content. Knowledge of the arguments that affect utility and their sensitivity not only to price but also to other elements of the environment can help make model predictions more precise.

Experiments are incentivized choices: Something (usually money) is at stake, making misrepresentation costly. Decisions made in the lab are real, not hypothetical. Two important elements of the experimentalists' creed are: Thou shalt pay, that is subjects are really paid according to the decisions that they make in the experiment; and thou shalt not deceive, that is everything we tell subjects in an economics experiment is true. An experiment might give a subject an opportunity to exhibit altruism, trust, or trustworthiness, at some cost. To say that you are altruistic in the lab means giving up some of your money. Doing so provides a behavioral measure of a preference for altruism.

To be a good measure it should have three characteristics. It should be replicable, giving the same result when repeated. It should be internally valid, measuring what it is supposed to measure. And it should be externally valid, correlated with behavior outside the lab. All these things are possible in lab experiments, where it is possible to control the experimental environment and specify carefully the variations or treatments we want to implement. Experiments are replicable and internally valid (if they are designed well). External validity can be tested by collecting information from subjects and by following them outside the lab.

4. Heterogeneity across People: Altruism

To illustrate the ways in which experiments can be used as measures, consider the question: Are women more altruistic than men? To measure altruism we use the dictator game. This is not much of a game, but rather an allocation task. One person, the dictator (though we don't use that loaded word in our carefully neutral instructions), is given an amount of money by the experimenter. He then is given the opportunity to donate some of this money to an anonymous recipient, who was recruited separately to the experiment and is in a different room, never observed by the dictator. In this game, economic man would clearly keep all the money for himself, since there is no reason to do otherwise.

Originally invented as a way to understand anomalous (i.e., contrary to game theory) results in bargaining studies (Forsythe et al. 1994), its use to measure altruism is more recent (Eckel and Grossman 1996a, b). In our early experiments we used a "double blind" protocol developed by Hoffman et al. (1994), which guarantees anonymity between subjects and between the experimenter and the subject, and so removes one of the reasons a person might give away money--to impress the experimenter or other subjects. This protocol was developed to improve the internal validity of the experiment, that is, to make sure it measured altruism.

The distribution of choices in a dictator game experiment with men and women as dictators is shown in Figure 1 (Eckel and Grossman 1998). Men on average donate $0.82 from a $10 endowment; women donate $1.60, a statistically significant difference. More than half of the subjects in this particular protocol give nothing, and more men than women fall into this category. On the other hand, more women than men give $5, half of their endowment, to an anonymous counterpart.

[FIGURE 1 OMITTED]

So by this measure, yes, women are more altruistic than men.

A variation on the dictator game matches subjects with a charity recipient instead of an anonymous person. Charities are more deserving of support than random, anonymous individuals, so we might expect higher levels of giving. (This is a rational response to the change in the characteristic of the recipient: The benefits are greater.) Table 1 shows a selection of results from an experiment where the subject chooses from a list of charities that her contribution will benefit (Eckel and Grossman 2003). With a $6 endowment, men give on average $2.63, or 44% of their endowment; women, 52%. At the higher $10 endowment we see a similar pattern, with men giving 45% and women giving 53%. In this study, we test a variety of endowments and several different levels of subsidies for charitable giving (such as matching amounts), and in all of them women give more than men. This provides further evidence that women are more altruistic than men. (2) This experiment is not restricted to gender differences: It could be (and has been) conducted to compare any groups of people, and also can be adapted to look at how people treat others.

5. Heterogeneity across Partners: Altruism

When the dictator game is conducted with subjects other than university students, there is a strong tendency for subjects to divide the endowment equally (e.g., Burks, Carpenter, and Verhoogen 2007; Whitt and Wilson 2007). If a large fraction of the subjects do the same thing, this is a problem. A good measure needs not only to capture altruism if it is there, but also to vary with intensity of preference. Piling up on a 50/50 split is not desirable. To get around this problem, we've been exploring a variation on the game that we call the comparative dictator game (CDG). This game introduces additional variation by giving subjects a frame of reference. In the CDG, subjects make several dictator game decisions, each with a different counterpart. One of the decisions is then chosen at random and paid. (See Eckel, Johnson, and Thomas 2006 for more detail.)

While we developed the game to enhance heterogeneity in giving to an anonymous partner, it also illustrates one of the ways in which people vary their decisions based on the characteristics of their counterparts. In this case the counterparts vary in their social distance, though they could be set up to vary in many different ways. We match them with a stranger, a friend, and a family member. They play all three games. After completing all three, one is selected randomly, and the subject is paid. The counterpart is also paid: We put the money in an envelope and either deliver or mail it to them. In this study we see that subjects vary quite a bit, not only in their overall level of giving, but also in how strongly they respond to the change in social distance. In the lab, from a $20 initial endowment subjects give on average $3.05 (15.2% of the endowment) to a stranger, $5.53 (27.7%) to a friend, and $7.13 (35.6%) to a family member. Behavior not only deviates from the selfish prediction of the naive rational actor model, but also varies by the characteristics of the counterpart. People are altruistic, and for most subjects, altruism is contingent, in the sense that it varies with what they observe.

6. Heterogeneity across People and across Partners: Trust

Like altruism, trust typically is measured using answers to survey questions. The trust game provides behavioral measures of both trust and trustworthiness. The most common measure is the question from the General Social Survey, "Generally speaking, would you say that most people can be trusted, or you can't be too careful in dealing with people?" In contrast, trust can be measured experimentally by putting subjects in a situation where there are gains (and risks) to trust. This measure is a simple, transparent game that examines trusting and trustworthy behavior directly. It is incentivized, and therefore misrepresentation is costly. In order to appear trusting a subject must put her own earnings into the hands of someone else, and in order to appear trustworthy, a subject must reciprocate trust by giving up some earnings.

In the trust game, first studied by Berg, Dickhaut, and McCabe (1995), two players are endowed with an amount of money, say $100. (This is to get the attention of an audience of professional, income-earning adults. Student subjects respond to lower stakes, say $10.) The first mover must decide how much (if any) to send to a second mover. Any amount sent is tripled by the experimenter. The second mover must decide how much, if any, to send back to the first mover. (The returned amount is not tripled.) Suppose the first mover sends $60. This would be tripled to become $180 received by the second mover, who must then decide how much to send back. The first move is "trust," the second is "trustworthiness," and the multiplier produces the potential gains from trust. Subjects engage in the game for real money, and they keep what they earn. In the first study, and most of those that followed, the decision makers were anonymous to each other. This was accomplished by implementing a protocol very similar to the one in Hoffman et al. (1994) developed for the dictator game. In this game, an economic man in the second position would keep everything; knowing this, the first mover would send nothing. However, when real subjects play this game, first movers send on average about half of the endowment, and second movers, on average, just reciprocate. Trust (barely) pays.

As with the dictator game, this game was invented to test hypotheses about game theory, and only later came to be used to measure and compare levels of trust and trustworthiness across types of people or across societies. In these games frequently there are systematic differences in trust across types--that is, people behave differently--as well as across partners--that is, people discriminate. For example, in a study of Harvard undergraduates, Glaeser et al. (2000) find that members of different races paired together are less likely to exhibit trustworthiness; less is returned by a second mover of a different race. Fershtman and Gneezy (2001) find that Israeli men are less likely to trust Eastern than Ashkenazic Jewish counterparts; and Ashraf, Bohnet, and Piankov (2006) and Burns (2005) explore discrimination among black, mixed-race, and white South Africans, and find plenty.

Our trust game protocol differs in important ways from earlier studies, and the changes allow us to examine individual interactions in real time without compromising anonymity (Eckel and Wilson (2006) provide details on the protocol). (3) In a series of experimental studies, we match subjects at two different sites over the Internet to play the game, and subjects see each others' photographs (framed like a passport photo) while making their decisions. In the first phase of the protocol, the first mover sees his counterpart's photo, then decides how much of a $10 endowment to send. He then guesses how much the second mover will return to him: If correct, he earns an additional $1. The second mover first guesses what the first mover will send, again earning $1 if correct, and then she receives the tripled amount sent by her assigned counterpart. Her next task is the important one: to decide how much to return. In the second phase, a separate group of subjects is recruited whose purpose is only to rate the photos for a set of characteristics. They are paid by the photo. We match things up so that people at one site rate the photos that folks from that site were matched with. I'm going to describe results from two of our studies.

Study 1: Attractiveness

In Study 1, we examine the effect of attractiveness on trust and reciprocity (Eckel and Wilson 2005; Wilson and Eckel 2006). Attractiveness is interesting to us because it is consequential in labor markets and other economically important situations. There is a substantial beauty premium in labor markets: 14.4-16.8% of earnings for men and 9.2-11.7% for women (Hamermesh and Biddle 1994). Attractiveness is associated with greater electability (Ottati and Deiger 2002) and with faster advancement (e.g., for attorneys, Biddle and Hamermesh 1998). In psychology, stranger attribution studies examine the characteristics that people believe others embody. Here, all sorts of positive attributes are ascribed to attractive persons. They are thought to be more intelligent, competent, and skilled, and more is expected of them (e.g., Webster and Driskell 1983). The beautiful also receive better treatment from others, even with respect to judicial decisions: Attractive persons get lower sentences (Stewart 1984)! (These and other studies are discussed in Eckel and Wilson 2005.)

There is some evidence that attractive people are of higher quality. Studies in evolutionary biology indicate that symmetric persons are judged more attractive, and symmetry is an indication (and a credible signal) of the quality of an organism. This symmetry is evident in people's faces, and can be read by subjects who are viewing only a facial photo (Rhodes et al. 1998; Zebrowitz and Rhodes 2004). Psychologists point out that the superior treatment of attractive persons, beginning in early childhood, can lead to the development of higher quality. Whether by nature or nurture, attractive people may indeed be superior. They score higher on intelligence (measured as IQ, with all its attendant shortcomings), health, extraversion, confidence, and social skills. While the stranger attribution studies are not entirely wrong, perceptions exaggerate any inherent differences in intelligence or productivity (Langlois et al. 2000).

In the workplace, trusting and trustworthy employees are thought to be more productive; certainly we can well imagine the flip side--the negative effects of suspicion and deceit. Trust and trustworthiness are likely to be rewarded in the labor market, and if attractiveness signals greater trust/trustworthiness, then this may partially justify the higher earnings. Differential treatment of the attractive may make them more trusting and trustworthy. The world is a safer, kinder place for them, after all. On the other hand, a feeling of entitlement may make them less trusting or less trustworthy. In the lab we can ask, are attractive people more trusting/ trustworthy, or do we just think they are?

Attractiveness has effects on behavior that can be detected in the lab. Solnick and Schweitzer (1999) find that more is sent to attractive counterparts in the ultimatum game, for example. (In this game the first mover proposes a division of a fixed endowment, and the second mover must accept or reject it. If accepted, payoffs are as proposed, and if rejected payoffs are zero for both players.) This behavior is an indication either that first movers want to give more money to attractive counterparts (taste-based discrimination) or that they expect attractive persons to reject lower offers (which is not taste based but rather based on expectations or, in a way, productivity). In the prisoner's dilemma game, attractive people are more likely to be selected as a partner and trusted (Mulford et al. 1998).

In one of my favorite attractiveness studies, Andreoni and Petrie (2006) show that, while the presence of attractive players initially increases contributions in the public goods game, when subjects find out how much others have given, they quickly adjust their contributions downward. This suggests that the subjects are biased in their estimates of what attractive people will contribute to the pie, a suggestion we are able to test directly with our protocol. Of course expectations can't be biased in equilibrium, where expectations have to be on average correct. The story we are telling, and the evidence we find for it, may indeed be an out of equilibrium phenomenon, sustainable for any length of time only in one-shot settings where feedback is nonexistent or infrequent. Repeated interaction provides the information needed to reach equilibrium, but repetition creates a whole new set of candidate equilibria. In our work we tend to focus on initial interactions, which allow us to measure stereotypes and biases and their effect on earnings in a simple setting, ignoring reputational considerations or repeated-game strategies.

In the trust game, we are able to test whether attractive people are more or less trusting or trustworthy (heterogeneity across people), as well as whether attractive people are treated differently (heterogeneity across partners). People may discriminate in favor of attractive people either because they like being nice to attractive people (or want to curry favor with them) or because they expect attractive people to have superior performance. In our protocol we can distinguish between those two motivations because we measure expectations as well as observing behavior.

To see what the experimental environment is like, some of the screens are displayed in the figures below. Figure 2 shows the first mover decision screen. (4) While observing his counterpart, the first mover decides how much to send to her. In a subsequent screen, he guesses how much she will send back. Meanwhile, the second mover observes the first mover, as shown in Figure 3, and is guessing how much he will send her. In a subsequent screen she finds out how much he has sent, and decides how much to send back. After the subjects are paid and go home, we keep their photos and have them rated by a different set of subjects. We show the raters screens like the one in Figure 4, and the raters decide which of each word pair best describes the photo. The 15 word pairs used in the ratings include attractive/unattractive, and those ratings are summarized in Figure 5, which shows the distribution of attractiveness ratings. (5) (Women are more attractive than men, but you knew that.) The data from the phase-I experiment (decisions and expectations), and the phase-II ratings are then combined to analyze behavior.

[FIGURES 2-5 OMITTED]

Table 2 summarizes sending and returning behavior. This table contains data only for subjects in the top and bottom quartiles of the attractiveness ratings. First movers exhibit a beauty premium, sending more to attractive than unattractive second movers. Both attractive and unattractive first movers display this behavior. Unattractive first movers send $5.29 from a $10 initial amount to unattractive second movers; they send $0.90 more to attractive counterparts. Attractive first movers send less on average, but also are more generous with attractive counterparts, sending $0.76 more. Based on previous research, this is the pattern we expected. Table 3 gives some insight into why this pattern is observed. Here data are pooled for all first movers, and presented just by the characteristics of the second mover. We see that more was sent to attractive second movers, and more was expected back from them. This indicates that the superior treatment is due in part to the higher expectations of attractive second movers--a productivity-based explanation. The third column shows that attractive and unattractive second movers do not differ in their behavior, so the difference seen in expectations is incorrect. (Notice also that trust pays for unattractive first movers in this environment, since percentage returned exceeds 33%, but not for attractive first movers.)

The bottom part of Table 2 shows the percentage returned by and to the most and least attractive. The data show an interesting pattern. To our surprise, and in contrast to the results of previous research, second movers exhibit a beauty penalty in their behavior. On average, unattractive second movers return 35% to first movers who are relatively unattractive and 29% to first movers who are relatively attractive. The same pattern is seen for attractive second movers, who return 40% to unattractive and 31% to attractive first movers. (6) What is the source of this unusual behavior? We thought perhaps the result might be a mirage. If second movers send back a lower percentage to less generous first movers, then that could cause a pattern of results like this one. Fortunately, we can control for this in our analysis. In the paper we report regression analysis showing that this effect survives controls for the amount sent. We are able to show that the result comes from the effect of dashed expectations, as summarized in Table 4. This table pools the data for all second movers, and shows their expectations, amount received from, and amount returned to unattractive and attractive first movers. The table indicates that attractiveness confounds intuition: Second movers expected more from attractive than unattractive second movers ($5.50 vs. $5.15), but they received less ($4.12 vs. $5.70). Second movers then punish attractive first movers for failing to live up to their expectations. Interestingly, this punishment is inflicted only on attractive first movers. As shown in Figure 6, only the beautiful are punished when expectations are dashed.

[FIGURE 6 OMITTED]

In sum, we find evidence of a beauty premium, in that relatively attractive people are more likely to be trusted. We also see a beauty penalty: Beautiful people are punished for failing to live up to the biased expectations of their counterparts. Attractiveness confounds intuition in the sense that expectations of attractive people are systematically too high. When we published these results we were contacted by many media outlets, who wrote about our results and uncovered examples from the real world. (7)

Study 2: Skin Shade

Let's turn now to study 2, an investigation of race, ethnicity, and skin shade (Eckel and Wilson 2007). The persistent gap in earnings between African Americans and whites in the United States and elsewhere--about 20% after controlling for productivity-related variables (Couch and Daly 2002)--again motivates our work in this area. Many different methods have been used to investigate the gap, including standard econometric studies (Couch and Daly 2002) and audit studies (Bertrand and Mullainathan 2004). The ability to isolate elements of decision making that might affect earnings is the reason for using lab experiments. Our experiments focus on the relationship between skin shade and trust/trustworthiness.

The design of this study is similar in most respects to study 1, with three differences. First, the decision is framed as a loan. While experimentalists prefer neutral language, we have found that a small amount of context can sometimes make the situation easier for the subjects to understand, in effect lowering the cognitive load. Second, instead of allowing the first mover to send any amount, the decision is all-or-nothing: The first mover must decide whether to send the entire $10 amount. We did this to avoid a possible confound in the data. Since second movers tend to return a larger percentage for higher amounts sent, we worried that some ethnic pairings might result in such low amounts sent that we would be unable to disentangle the effect of the amount from the effect of the pairing. Forcing all subjects to send the same amount would avoid this. Third, the amount sent is doubled instead of tripled. (8)

As before, the first mover observes the photo of his counterpart and then decides whether to make the loan (trust). He then guesses how much will be returned. The second mover guesses whether the loan will be sent, then decides how much to return. Photos are evaluated by a separate group of subjects recruited for that purpose. The subjects who participated were recruited from three schools: Virginia Tech, Rice University, and North Carolina A&T (a historically black engineering school). We set the experiment up to gauge discrimination. Subjects were matched randomly, but by recruiting an ethnically diverse set of subjects, we could examine differences in behavior by ethnic pairing. (9)

The screens are similar to those in Figures 2 and 3. Table 5 summarizes the decision to make the loan, separated out by the characteristic of the second mover. In the table, we see significantly lower levels of trust for nonwhite subjects. Whites are trusted with a loan 82% of the time, and nonwhites 63%. In analyzing the data, we initially pursued a modeling strategy that focused on ethnic pairings. However, we discovered that replacing ethnic categories with skin shade ratings effectively captured the variation in decisions. Interacting skin shade with ethnic category did not provide any additional explanatory power (see the regressions in Eckel and Wilson 2007). (Perhaps with more data, we would be able to distinguish ethnic-specific skin shade effects.) Figure 7 shows that skin shade is correlated with ethnic category, but there is considerable variation within categories. Table 6 shows loans, expectations, and returns by the skin shade of the second mover. Paralleling Table 5, we see that the lightest quartile is trusted 84% of the time, and the darkest 53.3% of the time. If the differences are based on differences in expected return, then that would support productivity-based discrimination. However, the average expected return does not differ significantly between the two groups, suggesting that discrimination in trust is taste based. Differences in average return show lower return by darker skinned second movers. However, this difference is small relative to the differences in trust.

[FIGURE 7 OMITTED]

Table 7 shows second mover expectations and returns. Lighter skinned second movers expect less trust from darker skinned first movers, but they are wrong. (Darker skinned second movers are correct in their relative assessments, but at an absolute level they are trusted less than they expect to be.) As with the attractiveness study, second movers who were surprised reacted strongly. In this case, darker skinned first movers are rewarded for making the loan when it wasn't expected.

In sum, we find a skin shade penalty, and a skin shade premium. First movers trust lighter skinned second movers more than those with darker skin shades. Darker skinned second movers never get the opportunity to show they are trustworthy. Lighter skinned second movers expect less from darker skinned first movers, but they are wrong. Darker skinned first movers trust more than is expected of them. Positively surprised second movers reward them, creating a skin shade premium.

Photos, Expectations, and Trust

These two studies taken together allow us to draw several tentative conclusions. The decision to trust is based on expectations, and expectations can be wrong. Our design allows us to distinguish between discrimination based on expectations (productivity) and discrimination based on taste. Attractive people are trusted more, in part because more is (incorrectly) expected of them. Darker skinned second movers are trusted less, despite only a small difference in expected return, indicating that the lack of trust stems from something other than expected return. Expectations also play an important role for second movers, who expect less trust from unattractive and darker skinned counterparts. Both of these expectations are based on stereotypes and are biased, resulting in a beauty penalty as attractive people are punished for dashing expectations, and a skin tone premium, as darker skinned people are rewarded for trusting more than expected.

We see that subjects condition their behavior on what they see--and what they believe about what they see--in the photos of their counterparts. Should they? In one study we compare the trust game with information only (subjects are told gender, ethnicity, and a few other things about their counterparts) and with photos. Correlation between expected return and actual amount returned is 0.10, and not significantly different from zero for the information condition. But in the photo condition, the correlation coefficient is 0.22, and significantly different from zero at p = 0.05. Seeing the photo improves subjects' ability to forecast their counterparts' behavior.

This led us to ask, will subjects pay to see the photos, and how much? In Eckel and Petrie (2006) we report the results of experiments to find out whether and how much subjects will pay to see the photos of their counterparts. Subjects play six trust games for $10, with the amount sent tripled, each with a different counterpart. One game is selected at random for payment. The subjects are given the option to purchase photos of their counterparts at prices ranging from $0.20 to $8. About half of the subjects are willing to pay something--$0.20--to see a counterpart's photo, and 7% will pay more than $2. We find that first movers--especially white first movers--are more likely to buy the photos (about 70% of them do so), and then they discriminate based on the information, trusting whites 50% more often than African Americans. (African Americans also trust whites more, but the difference is not statistically significant.) In addition, white second movers who buy the photos send back less to their black counterparts, in contrast to African American second movers. People will pay to learn about their counterparts, and act according to the information they acquire.

7. Conclusion

Experiments can be used to measure preferences, and to assess the magnitude and consequences of heterogeneity in behavior. People behave differently: There are systematic differences in average behavior across identifiable types or categories of people. For example, women are more altruistic, and attractive people are less trusting. More importantly, people discriminate. Experimental games can be used to measure discrimination without calling attention to it. In contrast to the typical survey approach, subjects make decisions naturally, without knowing what the study is about. This is especially important for behaviors that are socially sanctioned, such as discrimination against African Americans. If a subject is aware of the purpose of the study, it is easy and costless to misrepresent preferences in a survey, but in an experimental game such misrepresentation incurs a financial cost. The experimental approach allows us to observe latent or unconscious discrimination.

What about economic man? Three observations can be made from our studies. First, people differ. The common modeling assumption of a single representative agent is not accurate; we see considerable heterogeneity across individuals along the dimensions we studied: altruism, trust, and trustworthiness. Second, people treat each other differently. Behavior is conditional on the setting. In our studies, nearly all subjects vary their behavior depending on the situation they find themselves in. Third, the factors that people condition on are not just costs and benefits, but also include social elements. Our research highlights the importance of social considerations in economic decision making. Economic agents pay attention to social factors and condition their decisions on everything they know about the decision, including the characteristics of their counterparts. Behavior is conditional, based on expectations that are in turn based on stereotypes and that are biased, but better than chance at predicting the behavior of others. Economic man would not exhibit systematic biases. Perhaps a more accurate model of behavior can be developed by giving economic man a social identity.

Thanks to the Association for the opportunity to stand up in front of everyone and talk about my work. Thanks to the many people who provoked and improved this research through their skepticism and their encouragement: You know who you are! Special thanks for financial support go to the John D. and Catherine T. MacArthur Foundation and directors of the Network on Preferences and Norms, Herb Gintis, and Rob Boyd; and to the National Science Foundation and their wonderful program officers, especially Dan Newlon, Lynn Pollnow, Jon Leland, Laura Razzolini, Bob O'Connor, and Frank Seioli. Thanks most of all to the best coauthors anyone could ask for, including but not limited to: Sheryl Ball, Phil Grossman, Ragan Petrie, and Rick Wilson. Finally, to Genia Toma and Kathy Hayes, my fellow steel magnolias.

References

Andreoni, James, and Ragan Petrie. 2006. Beauty, gender and stereotypes: Evidence from laboratory experiments. Working Paper, University of California, San Diego.

Andreoni, James, and Lise Vesterlund. 2001. Which is the fair sex? Gender differences in altruism. Quarterly Journal of Economics 116:293-312.

Ashraf, Nava, Iris Bohnet, and Nikita Piankov. 2006. Decomposing trust and trustworthiness. Experimental Economics 9(3):193-208.

Berg, Joyce E., John W. Dickhaut, and Kevin McCabe. 1995. Trust, reciprocity, and social history. Games and Economic Behavior 10(1):122-42.

Bertrand, Marianne, and Sendhil Mullainathan. 2004. Are Emily and Greg more employable than Lakisha and Jamal? A field experiment on labor market discrimination. American Economic Review 94(4):991-1013.

Biddle, Jeff, and Daniel Hamermesh. 1998. Beauty, productivity and discrimination: Lawyers' looks and lucre. Journal of Labor Economics 16:172-201.

Burks, Stephen, Jeff Carpenter, and Eric Verhoogen. 2007. Fairness and freight-handlers: A test of fair-wage theory in a trucking firm, with Stephen Burks and Eric Verhoogen. Industrial and Labor Relations Review. In press.

Burns, Justine. 2005. Race and trust in post-apartheid South Africa. Working paper, Center for Social Science Research, University of Cape Town.

Couch, Kenneth, and Mary C. Daly. 2002. Black-white wage inequality in the 1990s: A decade of progress. Economic Inquiry 40(1):31-41.

Eckel, Catherine C., and Philip J. Grossman. 1996a. Altruism in anonymous dictator games. Games and Economic Behavior 16:181-191.

Eckel, Catherine C., and Philip Grossman. 1996b. The relative price of fairness: Gender differences in a punishment game. Journal of Economic Behavior and Organization 30(2): 143-158.

Eckel, Catherine C., and Philip J. Grossman. 1998. Are women less selfish than men? Evidence from dictator games. The Economic Journal 108(448):726-35.

Eckel, Catherine C., and Philip J. Grossman. 2003. Rebates versus matching: Does how we subsidize charitable contributions matter? Journal of Public Economics 87(3-4):681-701.

Eckel, Catherine C., Philip J. Grossman, and M. Johnston. 2005. An experimental test of the crowding out hypothesis. Journal of Public Economics 89(8):1543-60.

Eckel, Catherine C., Cathleen A. Johnson, and Duncan Thomas. 2006. Altruism and resource sharing within families and villages in Mexico. Presented at the Economic Science Association Regional Meetings, Tucson, AZ, October.

Eckel, Catherine C., and Ragan Petrie. 2006. Face value. Unpublished paper, Georgia State University.

Eckel, Catherine C., and Rick K. Wilson. 2005. Detecting trustworthiness: Does beauty confound intuition? Unpublished paper, University of Texas at Dallas.

Eckel, Catherine C., and Rick K. Wilson. 2006. Internet cautions. Experimental Economics 9(1):53-66.

Eckel, Catherine C., and Rick K. Wilson. 2007. Initiating trust: The conditional effects of sex and race among strangers. Unpublished paper, Rice University.

Fershtman, Chaim, and Uri A. Gneezy. 2001. Discrimination in a segmented society: An experimental approach. Quarterly Journal of Economics 116(1):351-77.

Forsythe, R., J. L. Horowitz, N. E. Savin, and M. Sefton. 1994. Fairness in simple bargaining experiments. Games and Economic Behavior 6:347-69.

Glaeser, Edward L., David Laibson, Jose A. Scheinkman, and Christine Soutter. 2000. Measuring trust. Quarterly Journal of Economics 115(3):811-46.

Hamermesh, Daniel, and Jeff Biddle. 1994. Beauty and the labor market. American Economic Review 84:1174-94.

Harris, Misty. 2006. Here's a problem we'd all like to have. Leader Post, 25 September, C1-2.

Hoffman, Elizabeth, Kevin McCabe, Keith Shachat, and Vernon L. Smith. 1994. Preference, property rights and anonymity in bargaining games. Games and Economic Behavior 7(3):346-80.

Holt, Charles A. 2003. Economic science: An experimental approach for teaching and research. Southern Economic Journal 69(4):755-71.

Langlois, Judith H., Lisa Klakanis, Adam J. Rubenstein, Andrea Larson, Monica Hallam, and Monica Smoot. 2000. Maxims or myths of beauty? A meta-analysis and theoretical review. Psychological Bulletin 126(3):390-423.

Mulford, Matthew, John Orbell, Catherine Shatto, and Jean Stockard. 1998. Physical attractiveness, opportunity, and success in everyday exchange. American Journal of Sociology 103(6):1565-92.

Ottati, Victor C., and Megan Deiger. 2002. Visual cues and the candidate evaluation process. In The social psychology of politics, edited by V. C. Ottati et al. New York: Kluwer Academic/Plenum Publishers, pp. 75-87.

Rhodes, Gillian, Fiona Proffitt, Jonathon M. Grady, and Alex Sumich. 1998. Facial symmetry and the perception of beauty. Psychonomic Bulletin and Review 5(4):659-69.

Solnick, Sara J., and Maurice E. Schweitzer. 1999. The influence of physical attractiveness and gender on ultimatum game decisions. Organizational Behavior and Human Decision Processes 79(3):199-215.

Stewart, J. E. 1984. Appearance and punishment: The attraction-leniency effect in the courtroom. Journal of Social Psychology 125:373-8.

Webster, Jr. M., and J. E. Driskell, Jr. 1983. Beauty as status. American Journal of Sociology 89:140-65.

Whitt, Sam, and Rick K. Wilson. 2007. The dictator game, fairness and ethnicity in postwar Bosnia. American Journal of Political Science 51(3). In press.

Wilson, Rick K., and Catherine C. Eckel. 2006. Judging a book by its cover: Beauty and expectations in a trust game. Political Research Quarterly 59(2):189-202.

Zebrowitz, Leslie A., and Gillian Rhodes. 2004. Sensitivity to bad genes and the anomalous face overgeneralization effect: Cue validity, cue utilization, and accuracy in judging intelligence and health. Journal of Nonverbal Behavior 28(3):167-85.

(1) For example, Eckel, Grossman, and Johnston (2005) find very large differences in crowding out with small differences in context.

(2) Not every study replicates this result. For example Andreoni and Vesterlund (2001) vary the price of giving and find it is only true for women at high prices. They find that lowering the price of giving makes men but not women more generous.

(3) Anonymity is desirable. If anonymity is breached, we cannot be certain that subjects are playing the game we intend them to play. Instead, the possibility of postgame interaction, positive or negative, may influence decisions.

(4) This photo and others in this paper are of family members of the author.

(5) To get this distribution, we take each rater's ratings and standardize them, centering the distribution at zero and creating z-scores. For each photo, these are then averaged across raters to get a photo score. The distribution across photos is what you are looking at in the figure.

(6) Note these do not average to the percentages in Table 3 because Table 3 includes all first movers, and Table 2 includes only the top and bottom quartile pairings.

(7) Consider the beauty queen, Venessa Fisher, crowned Miss Universe Canada 2004: "It's pretty difficult to live up to everybody's standards. People want to criticize everything about you, looking for the littlest things to tear apart," she says. "They're either completely intimidated or have this completely distorted image of who I am." Fisher, now studying broadcast journalism in California and interning at the Dr. Phil show, describes a constant battle to thump the superficial stereotypes people have about beauty and its relationship to personality. "I grew up exactly the same as everyone else ... I'm really normal," she says. "But I guess that doesn't come through until people start to know me." (Ledger Post, Regina, Saskatchewan, 2006).

(8) As it turned out, this combination of choices was not ideal, and we wouldn't repeat it. The combination of loan, one amount sent, and doubling, meant that our second mover data has a large spike at a return of $10, which represents equal splitting of the amount received by the second mover, and just repaying the loan.

(9) In practice, since most of our African American subjects were at NCAT, most pairings that include African Americans are black/white pairings. We have very few black/black pairings, a shortcoming we plan to address in future studies. Therefore, this is largely a study of how white people treat a diverse set of counterparts.

Catherine C. Eckel *

* Catherine C. Eckel is Professor of Economics, School of Economic, Political and Policy Sciences, University of Texas at Dallas. This paper was delivered as her presidential address on November 19, 2006, in Charleston, SC, at the annual meeting of the Southern Economic Association.

Table 1. Dollar Contribution to Charities

Endowment Men Women p-Value (Means Test)

$6.00 $2.63 $3.11 0.06
$10.00 $4.52 $5.34 0.05

Source: Eckel and Grossman (2003).

Table 2. Average Amounts Sent to and Percentage Returned by Players,
Contingent on the Attractiveness Ratings of the Decision Makers and
Counterparts (Upper and Lower Quartiles of Attractiveness Ratings)

 Unattractive Attractive
 Decision Maker Decision Maker
 (First Mover) (First Mover)

Amount sent to second
 mover who is
 Unattractive
 (lower quartile) $5.29 (3.06) $3.79 (2.17)
 Attractive (upper
 quartile) $6.12 (3.19) $4.46 (3.07)
 t-test t = 1.33, d.f. = 100, t = 1.22, d.f. = 112,
 p = 0.09 p = 0.11

 Unattractive Attractive
 Decision Maker Decision Maker
 (Second Mover) (Second Mover)

Percentage returned to
 first mover who is
 Unattractive
 (lower quartile) 35% (21.2) 40% (19.9)
 Attractive (upper
 quartile) 29% (22.2) 31% (19.2)
 t-test t = 0.89, d.f. = 98, t = 2.24, d.f. = 95,
 p = 0.19 p = 0.01

Source: Eckel and Wilson (2005). Standard deviation in parentheses.

Table 3. First Mover Trust, Expected Return, and Return Received
Conditional on the Attractiveness of the Second Mover

 Average Average
Second Mover Average Amount Expected Percentage
Attractiveness Sent Return (a) Returned (b)

Unattractive (lower
 quartile) $4.49 (3.02) $8.06 ($7.22) 34.52 (20.47)
Attractive (upper
 quartile) $5.15 (3.28) $9.5 ($6.59) 35.74 (17.52)
t-test t = 2.08 t = 2.06 t = 0.626
 d.f. = 390 d.f. = 390 d.f. = 382
 p = 0.04 p = 0.04 p = 0.53

(a) Includes all first mover expectations.

(b) Includes only second movers who received something
from the first mover.

Table 4. Beauty Confounds Intuition: Second Mover Expectations,
Amount Received, and Percentage Returned, Conditional on the
Attractiveness of the First Mover

 Second Mover

First Mover Who Was Expected from Got from Returned to

Unattractive 5.15 5.70 35%
Attractive 5.50 4.12 30%

Source: Eckel and Wilson (2005).

Table 5. First Movers: Percentage Trusting Moves
by Counterpart Ethnicity

Caucasian 81.9 (68/83)
Total minority 63.0 (29/46)
African American 60.9 (14/23)
Asian 61.5 (8/13)
Hispanic/other 70.0 (7/10)

Source: Eckel and Wilson (2005).

Table 6. Loans and Returns by Skin Shade of Second Mover

 Average
Second Mover Percentage Expected Average
Skin Shade Loan Made Return (a) Return (b)

Lightest quartile 84.4% (27/32) $7.66 ($3.96) $9.72 (4.34)
Darkest quartile 53.3% (16/30) $8.00 ($5.51) $7.97 ($3.79)

(a) Includes all first mover expectations. Standard deviations
in parentheses.

(b) Includes only second movers who received the loan. Standard
deviations in parentheses.

Table 7. Second Mover Expectations and Returns

 Second Mover Skin Tone

First Lighter (below Median Rating) Darker (above Median Rating)
Mover
Skin % Expect % Trusted % Expect % Trusted
Tone Trust (n/N) (n/N) Trust (n/N) (n/N)

Lighter 77.1% (27/35) 82.9% (29/35) 76.7 (23/30) 66.7% (20/30)
Darker 65.5 (19/29) 86.2% (25/29) 71.4 (25/35) 65.7% (23/35)