Experimental Economics Research Paper

This sample Experimental Economics Research Paper is published for educational and informational purposes only. Like other free research paper examples, it is not a custom research paper. If you need help writing your assignment, please use our custom writing services and buy a paper on any of the economics research paper topics.

Being able to test theories and understand the underlying mechanism behind observed phenomena is crucial for scientific progress in any discipline. Experimentation is an important method of measurement in the natural sciences as well as in social sciences such as psychology, but the use of experiments for gathering economic data is a much more recent endeavor. Economics has long been regarded as a nonexperimental science, which has to rely on observations of economic behavior that occur naturally. Experiments, however, have found their way into the economist’s toolkit in the past few decades and are now being employed commonly in main-stream economics research in many diverse subfields such as game theory, industrial organization, labor and development economics, and, more recently, macroeconomics.

The very first experiment in economics is known to have been conducted by Bernoulli on the St. Petersburg’s paradox in 1738 (see Kagel & Roth, 1995, for more on the history of experimental economics). However, more formal experimentation started in the 1930s with individual choice experiments and flourished especially with the advent of game theory with the work of von Neumann and Morgenstern (1944) on the theory of decision making and games. Also around that time, the first “market experiments” were run by Chamberlin (1948) to test competitive equilibrium. Gradually, experiments started being used more and more widely in many areas of economics, and the number of experimental research papers published in economics journals has been growing rapidly and is now on par with more “classical” fields such as economic theory. The development of experimental economics as a field has also been parallel to advances in the field of “behavioral economics.” Behavioral economics aims at integrating insights obtained from psychology into economic models, frequently uses experiments as a method for collecting data and finding out patterns of behavior that are inconsistent with standard theory, and builds new models that can explain the observed behavior in experiments. The most evident recognition of the importance of experimental and behavioral economics was the 2002 Nobel Prize in economics, which was awarded to Daniel Kahneman and Vernon Smith for their contributions to behavioral and experimental economics, respectively.

But why and when do economists need experiments? One of the main goals of empirical analysis in economics is to understand how different models of economic decision making fare in understanding observed economic behavior and outcomes and therefore test the predictive success of economic theories. However, it is not always possible to conduct proper tests of theories or to measure the effects of different economic policies using naturally occurring data because of at least three reasons. First, naturally occurring data may simply not exist. For example, in testing models of strategic interaction, oftentimes it is important to know what people think or believe about other people’s actions, and although actions are observable, beliefs are not. Similarly, reservation wages and workers’ outside options are also not observed in naturally occurring data but are important to understand agents’ behavior in labor markets. Laboratory experiments, on the other hand, allow us to collect data on these unobservables.

Second, experiments allow randomization into treatments of interest, reducing selection bias and giving the researcher the proper counterfactual for causal inference. Consider the following example. Suppose that we are interested in the effects of tournament-type compensation schemes (where one’s wage depends on his or her performance relative to others) versus piece-rate compensation schemes (one’s wage depends only on his or her own performance) on worker productivity. If we collect real worker productivity data from firms that use each of these two types of incentive schemes and make a direct comparison, we cannot be sure whether any productivity differential we observe comes from the true effect of incentives or from firms or workers’ unobservable differences that are correlated with productivity. For example, more able and more ambitious workers may think that they have better prospects in a firm that uses tournament incentive schemes, and this motivation differential will bias the estimates of the “treatment effect” of incentives on productivity. That is, the samples under the two incentive schemes are “selected” and not directly comparable, and this incomparability will blur any type of inference we can make. Although econometri-cians have been devising methods that could alleviate some of these problems (such as instrumental variables and matching techniques), having direct control over the data-generating process makes inference much simpler. In the context of the incentives example, by assigning workers randomly into two treatments, one where they work under a tournament incentive scheme and one under the piece-rate scheme, it would be possible to measure the “true” effect of incentive schemes on productivity. This is possible because random assignment acts as a control for individual characteristics, considering the fact that if you have a sufficient number of subjects, the two groups will look sufficiently alike along dimensions such as ambition, ability, and so on, which we would like to control. By assigning subjects randomly into “treatments” that differ along the dimension of interest, experiments take care of the selection problem through randomization and can accurately isolate the effect of the focus variable.

Likewise, in a controlled experiment, which variables are exogenous and which are endogenous is clearly known, and this allows the experimenter to make causal inferences about the association between the variables, whereas with natural data, there are usually many factors changing at the same time, making it hard to disentangle the effect of a certain factor on the variable of interest. Another rationale for using experiments, perhaps of more theoretical interest, is that it may be difficult to test theories of one-shot interactions with naturally occurring data since factors such as reputation are oftentimes present because interactions naturally take place mostly in repeated settings. All these advantages suggest that experimentation can be a very valuable tool for gathering data on economic decision making.

While economic experiments have generally used the laboratory as their setting and university students as subjects, “field experiments” have been receiving much attention recently. By testing behavior in a setting that is more “natural” on several dimensions, field experiments provide insight on the “external validity” or generalizability of the results from laboratory experiments and complement lab experiments in improving our understanding of economic phenomena. According to Harrison and List (2004), experiments may differ in the nature of (a) subject pool, (b) information and experience that the subjects bring to the task,

(c) the commodity being transacted, (d) task and institutional rules, and (e) the environment that subjects operate in. These five factors are used to determine the field context of an experiment and result in the following taxonomy: (a) conventional lab experiments, which use a standard subject pool of university students, abstract framing (e.g., choices are labeled A, B, C, etc., rather than with words that suggest context), and an imposed set of rules; (b) artifactual field experiments, which are conventional lab experiments but with a nonstandard subject pool (e.g., actual workers from a firm rather than university students); (c) framed field experiments, which are artifactual field experiments but with field context in the task, commodity, information, or environment (e.g., rather than trading a “virtual” commodity on the computer, subjects trade a real commodity in a real field context); and (d) natural field experiments, which are framed field experiments in which subjects are not aware that they are in an experiment.

In what follows, we will first provide a discussion of some of the main methodological issues involved in experimental research and then provide a selective account of the main areas of application where experiments have been employed in economics. Because of space issues, we are bound to leave out many important domains and applications, so our treatment here should be taken as an illustrative approach that provides examples of how experiments can contribute to our understanding of economic behavior.

The Methodology of Experiments

The goal of the experimenter is to create a setting where behavior can be measured accurately in a controlled way. This is usually achieved by keeping constant as many variables as possible and varying others independently as “treatment variables.” To achieve as much control as possible, the experimental method in economics is based on “inducing preferences” by using appropriate monetary incentives that are tied to the consequences of decisions made during the experiment. A typical economics laboratory experiment involves recruiting subjects, who are usually university students, and paying them a fixed show-up fee for their participation, plus the amount they earn during the experiment. The amount they earn during the experiment, in turn, is tied to the payoff consequences of the decisions they make. This feature of economics experiments, in fact, is a main characteristic that distinguishes economics experiments from psychology experiments, where decisions are often hypothetical and not incentivized.

For properly inducing preferences, the experimenter should have control over the preferences of the subjects and should know what the subject is trying to attain. For example, in a game theory experiment, we would like the numbers in the payoff matrix of the game to represent the players’ utilities, and we therefore pay subjects according to the payoffs in the matrix. Naturally, however, there may be unobservable components of utility affecting subjects’ behavior. For instance, if one runs a 100-period experiment with the same type of choice being repeated every period, subjects might get bored and start acting randomly. Alternatively, some subjects might have preferences over other subjects’ monetary earnings, and if they know the payoff distribution among the participants, this might affect their behavior in ways that are unrelated to the hypothesis of interest. To make the effects of these unobserved factors minimal, one should make monetary rewards as strong as possible. Having salient monetary rewards will also ensure that subjects are motivated to think carefully about the decision, especially when the decision task is cognitively demanding. Attaching strong monetary consequences to decisions reduces the occurrence of random decisions and minimizes the errors and outliers that would otherwise be observed more often in the data.

On the basis of these general ideas, Nobel Prize winner Vernon Smith (1982) put forward the following precepts for achieving proper control:

Nonsatiation: This means that individuals should prefer more of the reward medium used in the experiment (usually money) to less of it.
Saliency: This assumption means that the reward medium is suitably associated with the choices in the experiment— for example, if one wants action A to be a better action for the individual than action B, then action A should be associated with a higher monetary payoff than B.
Dominance: This means that the reward structure should dominate other factors associated with participation in the experiment (e.g., boredom). High monetary stakes can help achieve this.
Privacy: Individuals’ utility functions may have unobservable components that depend on the utility or payoffs of others. For example, a subject’s decisions may depend on how much others are earning, if he or she cares about “fairness” of payoffs across subjects. By withholding information about other subjects’ payoffs and giving subjects information about their own payoffs only, this potential issue can be mitigated, and better control of preferences can be achieved.

While inducing preferences is very important for experimental control, there are some cases in which experimenters do not want to induce preferences but are interested in obtaining information on the natural (“homegrown”) preferences of subjects. For example, we may be interested in knowing how much an individual values a certain object (e.g., the item for sale in an auction). Likewise, we may be interested in knowing the beliefs of the individual. As mentioned before, the latter can be especially important in testing game-theoretic models, where subjects’ beliefs about others’ possible actions are important in shaping optimal strategies. Experimental economists have devised incentive-compatible mechanisms to elicit subjects’ true valuations for an object (e.g., the maximum amount of money they would be willing to pay to buy the object) and use various techniques to elicit subjects’ beliefs truthfully. While discussion of these methods in detail is beyond the scope of this research paper, it is important to note that these methods have allowed economists to obtain crucial information that is not available in the field.

One last principle that is relevant for running a “good” experiment is “design parallelism” (Smith, 1982). This principle refers to the need for laboratory experiments to reflect naturally occurring environments to the extent possible. This is very much related to the concept of “external validity” of the experiment, in other words, how much the experimental decision resembles decisions that individuals face in the “real world,” which will affect the extent to which the experimental results can be extrapolated to natural economic environments. Although the external validity of an experiment is important, it should be noted that adding more and more complexity to an experiment to increase external validity could result in loss of control and compromise the “internal validity” of the experiment, making the data useless. One should therefore be careful in choosing an experimental design that can be as realistic as possible while still maintaining proper control and avoiding confounds.

Applications of Experimental Methods in Economics

Individual Decision-Making Experiments

The experiments conducted in this area analyze non-strategic decision making by a single individual, in a context with no interaction between the subjects in the experiment. The goal is to understand the decision-making process of the individual and the motivation behind the observed behavior. The topic is strongly related to the psychology of judgment and illustrates the interdisciplinary aspect of experimental methods quite well. In fact, as we will explain below, this strand of the experimental economics literature has collected quite a large body of observations that are inconsistent with standard economic theory and has been an important part of the research in behavioral economics.

Experiments of individual decision making investigate choice in different contexts: over time, under static uncertainty, under dynamic uncertainty, and so on. A very important set of experiments here concerns decision making under uncertainty and provides tests of the relevant standard economic theory, which posits that individuals are expected utility maximizers. As mentioned before, this means that individuals maximize the expected value of utility, defined as the sum of utility from different possible outcomes, each multiplied by the respective probability that that outcome occurs. Anomalies, or observations violating expected utility theory in these experiments, have been frequent. For example, a well-known departure from the predictions of expected utility is the Allais paradox. Consider the following example, with two decision tasks (Kahneman & Tversky, 1979):

Decision 1:

Option A: $3,000 for sure
Option B: $4,000 with 80% chance, $0 with 20% chance

Decision 2:

Option C: $3,000 with 25% chance
Option D: $4,000 with 20% chance

A vast majority of subjects select Option A in Decision 1 but Option D in Decision 2, which is inconsistent with expected utility theory since the lotteries in Decision 2 are equivalent to Decision 1 (when one divides all the winning probabilities of Decision 1 by 4, Decision 2 is obtained, and this across-the-board reduction in winning probabilities should not affect the choice according to expected utility theory).

Choices in such lottery choice experiments have also pointed to a “reflection effect”: While individuals make risk-averse choices when the choices involve gains (e.g., a lottery that pays $100 with 50% chance and $0 with 50% chance vs. a sure gain of $40), they act as if they are risk loving when the choices involve losses (e.g., a lottery that involves a $100 loss with 50% chance and no loss with 50% chance vs. a sure loss of $40). That is, losses and gains are treated differently, which is again inconsistent with expected utility theory.

The laboratory evidence that highlights such anomalies has stimulated the development of alternatives to expected utility theory. A well-known alternative, called “prospect theory,” was proposed by Kahneman and Tversky (1979). Prospect theory allows for loss aversion, reference dependence, reflection effect, and probability miscalculations and can explain a significant amount of the anomalous findings in the lottery choice experiments. Through its modeling of reference dependence and loss aversion, prospect theory can also explain a phenomenon called the “endowment effect,” which refers to the observation in experiments that there is a discrepancy between individuals’ valuations of a good, depending on whether they own it or not. That is, the minimum amount of money that individuals are willing to accept in order to part with a good they own (e.g., a coffee mug that has been given to them in the experiment) is higher than the maximum amount they would be willing to pay for the same good when they do not own it, which is inconsistent with standard theory.

Although many experimental results in individual decision-making experiments point to deviations from the predictions of standard economic theory, there is also some evidence that market experience and large monetary stakes can improve the alignment of observed behavior with standard predictions. For example, John List (2006) finds, in a field experiment involving traders in an actual market for sports memorabilia, that inexperienced traders display a strong endowment effect, but experienced traders who have been engaging in market activities for a long time do not. This might suggest that these anomalies or biases might be less prevalent in actual markets, where individuals self-select into economic roles and gain experience through repeated transactions.

One methodological point to be made here is that the type of individual lottery choice tasks mentioned above can also be used to measure subjects’ risk preferences in the laboratory. In many different experiments involving a wide range of economic decisions, it is useful to know the risk preferences of individuals. For example, risk-averse and risk-neutral individuals are predicted to bid differently in auctions, and having information on risk preferences allows economists to obtain better insight into behavior. One such method to measure risk preferences is the “Holt-Laury mechanism” (Holt & Laury, 2002), which involves giving subjects a series of choices between a risky lottery (that has a large spread between the good and bad payoff) and a safe lottery (that has a small spread between the good and the bad payoff), which differ in the likelihood of the good payoff. As the probability of the good payoff increases, the attractiveness of the risky lottery increases. By looking at when the subject switches to the risky lottery as the good payoff probability increases, it is possible to get a measure of the risk aversion of the subject.

Game Theory Experiments

Most of modern microeconomics research involves models of strategic interaction. Since the work of John von Neumann and Oskar Morgenstern (1944) and later of John Nash, game theory has become the fundamental approach in microeconomics, replacing the Walrasian and Marshallian models of consumption, production, and exchange in which individuals take decisions in response to exogenous prices by models of strategic interaction. The theory of games has been used to analyze strategic behavior in coordination games, public good contribution games, auctions, analysis of oligopolies in industrial organization, and so on. In general, game-theoretic models assume that individuals are rational and have stable preferences over outcomes and correct beliefs as to how choices affect the relative probability of possible outcomes. They are assumed to maximize their own expected payoff given their preferences and beliefs and take into consideration material and informational constraints that may exist.

There are two main criticisms to game theory. First, individuals may not be able to be as forward looking as game theory predicts. They may not always behave rationally or may not perceive others to be rational. Second, people do not behave “selfishly” in all situations and may not expect others to behave in that way either. Experimental methods help economists explore how important these issues are. In experiments, information and incentives are held constant, which allows us to accurately test how well game-theoretic principles predict behavior and understand in which situations individuals do behave in line with the theory and in which cases they deviate from it.

Many of the experiments on game theory have tested (a) the rationality assumption or the depth of strategic reasoning and (b) the assumption of “selfishness,” which posits that individuals’ utility is dependent on their own monetary payoffs only. In the following discussion, we will focus on games that have attempted to clearly test these two postulates of standard economic theory. First, we will illustrate the effects of limits on rationality through experiments on limited strategic thinking. Second, we will discuss social preferences experiments that test the assumption of selfishness and exclusively money-maximizing behavior.

Illustration 1: Why Aren’t We All Chess Players? Experiments on Limited Thinking

We will illustrate tests of rationality through a discussion of the “guessing game.” An important recent set of experiments that has provided a great way to test the depth of players’ reasoning is “guessing games” or “beauty contest games.” (This name is based on John Maynard Keynes’s likening of the stock market to a newspaper beauty contest where readers’ aim is to guess the most beautiful lady, as determined by the population. Keynes noted that in such games, it would be important to guess how others think, how others think others think, and so on.) In the canonical version of the guessing game (Nagel, 1995), a group of players is asked to choose a number from a given range (e.g., [0, 100]). The average of all numbers submitted is taken, and 2/3 (or any fraction smaller than 1) of that average becomes the “target number.” The person whose chosen number comes closest to this target number wins a fixed prize. In other words, the goal is to correctly guess 2/3 of the average guess. The prediction of game theory in this game is that all players will submit a guess of zero. In fact, this solution can be achieved by a process called “iterated elimination of dominated strategies,” which proceeds as follows: If I am rational, I should realize that since the maximum possible number is 100, the target number can never be more than 66. Therefore, any guess above 66 is a “dominated” action—no matter what others might be doing, I should not submit a guess more than 66. But if I know that everyone is rational, I should also realize that no one will submit a guess above 66, which makes any guess above 44 dominated for me because I know the target number cannot be more than 44. Continuing in this fashion, it is possible to reach the unique equilibrium outcome of everyone guessing zero. Experimental results, on the other hand, show that very few people submit guesses of zero and that there are clusters of observations around points such as 33 and 22. Researchers have provided the follow-ing type of model to account for these deviations from the prediction of game theory: Suppose that individuals differ in their “depth of reasoning” and are classified as “Level k thinkers,” where k is the number of rounds of iterated reasoning they can engage in. For example, a Level 0 person just guesses randomly, without any strategic thought. A Level 1 person, on the other hand, thinks that everyone else is Level 0 and best responds to that—meaning, a Level 1 person will assume that on average the guess will be 50 and therefore chooses a number that is 2/3 of that, 33. A Level 2 person thinks that everyone else is Level 1 and therefore the average guess will be 33 and takes 2/3 of that to submit his or her guess. The experimental data show that most people are within three levels of thinking. Although this game may seem of theoretical interest, in fact it highlights mechanisms that are important in many economic settings, such as the stock market, where it is important to guess what others think about the value of a stock and how rational they are in their behavior.

Illustration 2: Do We Care Only About Ourselves? Experiments on Social Preferences

As mentioned before, another important focus of game theory experiments has been testing the assumption of “pure self-interest,” which we define as preferences that depend only on our own monetary earnings, independently of what others earn. A very interesting set of results in this literature comes from experiments that highlight “social preferences.”

The concept of social preferences refers to the concern (positive or negative) for others’ well-being. We distinguish between two types of social preferences. The first type refers to outcome-oriented social preferences. Here, individuals care about the distribution of payoffs, and this encompasses pure altruism, inequality aversion, and possibly a concern for efficiency. The second type of social preferences concerns intention-based reciprocity (i.e., individuals care about the intentions that drive other players’ actions). Here an individual is willing to sacrifice his own material payoffs in order to reciprocate, either rewarding kind (fair) or punishing unkind (unfair) behavior.

Research in experimental economics has helped us understand the nature of attitudes toward social preferences and how these attitudes interact with self-interest. An important workhorse that is used for studying social preferences is simple games in which subjects have to decide on an allocation of money between themselves and an anonymous other subject. Experiments on social preferences generally study such games, which include the so-called ultimatum game, dictator game, trust game, gift exchange game, prisoner’s dilemma, public good game, and modifications to these canonical settings. Subjects make decisions usually for a certain number of periods and are usually rematched with different subjects every period, which enables economists to think about this as a one-shot interaction and abstract from repeated game effects. In the past three decades, a vast number of experiments on social preferences have been conducted either to check the existence of social preferences (in the lab) or to test the robustness of results to different subject pools, stakes, framing, number of players, and other design and procedural variables. More recently, field experiments have been con-ducted to test the external validity of lab experiments and to obtain insight on the strength of social preferences in the field. In the following, we first describe classical experiments on social preferences and some extensions. We then discuss the related field experiments.

Ultimatum and Dictator Games

The ultimatum bargaining game has been one of the most widely studied games in the past 25 years. In the basic version of this game (Guth, Schmittberger, & Schwarze, 1982), two subjects bargain over the division of a “pie.” The first player (the proposer) has an endowment of money and decides on the amount X to send to an anonymous partner (the responder), who then decides whether to accept it or not. If accepted, both players get their agreed-upon shares. If rejected, both receive nothing. While the rational solution predicts that the proposer should offer the smallest possible share and the responder should accept it, Guth et al. (1982) find that, on average, proposers offer 37% of the pie and that low offers are frequently rejected. Since then, numerous other experiments using the ultimatum game have been conducted, and many possible methodological explanations (stakes, subject pool, nature of the game) for the gap between theory and empirical results have been tested. Results are robust:

(a) The modal and median ultimatum offers are usually 40% to 50%, and mean offers range from 30% to 40%;

(b) very low offers (0%-10%) and “too fair” offers (51%-100%) are rarely observed; and (c) offers below 20% are rejected half the time.

The equilibrium in the ultimatum game is easy to compute and is exempt from bounded rationality or confusion as possible explanations for the results, which makes the ultimatum game one of the most common experimental designs used for inference about individuals’ social preferences. For instance, when a responder rejects a positive offer, this means that his or her utility function has more than only a monetary argument. For example, a rejection of a positive but low offer could reveal a concern for negative reciprocity, as the responder is willing to sacrifice his or her own monetary payoff to punish the proposer’s action. When the proposer makes a higher offer, however, it could mean a preference for fairness, a fear of rejection, or both. Further experiments using the so-called dictator game disentangle the two explanations and show that both have some explanatory validity. Forsythe, Horowitz, Savin, and Sefton (1994) conducted a dictator experiment that mainly removes the responder’s move from the standard ultimatum game—proposers have the power to decide on the allocation of the pie, and the responder cannot reject. In this game, offers are found to be less generous than in the ultimatum game (the mean allocation is about 20% of the pie), but there is still a significant fraction of people who give positive amounts of money to the other party.

Trust and Gift Exchange Games

Trust and gift exchange games are both sequential prisoner’s dilemma games, and they represent situations where contracts are necessarily incomplete, allowing for a controlled study of the nature and effectiveness of trust and reciprocity in economic interactions. The trust game (Berg, Dickhaut, & McCabe, 1995), which is also called the investment game, considers a situation in which one individual (the investor) transfers to another (the trustee) the power to make a decision that affects the utility of both. More specifically, the investor has an initial endowment A and decides the amount X of A to invest. The money invested is multiplied by r and transferred to the second player, who decides the amount Y of rX to return to the investor. The trustee, therefore, plays a dictator game with an endowment decided by an initial investment made by the recipient. The investor receives A – X + Y, and the trustee receives rX – Y.

When players are entirely selfish and only interested in maximizing their own monetary payoffs, the second player would never return any positive amount of rX. Given that the investor knows that the second player will behave selfishly and he or she will receive nothing in return, Player 1 invests nothing, keeping A for himself or herself. Although it yields a socially inefficient outcome (the total payoffs would be maximized if the investor transferred everything), this “subgame perfect” equilibrium of zero investment holds because (a) players cannot sign binding and irrevocable contracts before the beginning of the game, and the trustee cannot be punished for not sending a positive amount back to the investor, and (b) the game is played only once or with different partners, so there is no role for reputation formation and strategic behavior on the part of the trustee.

In Berg et al. (1995), each player was matched only once with another player, and subjects’ anonymity among themselves as well as anonymity from the experimenter was guaranteed. Parameters used in their experiment consisted of $10 for investor’s endowment, and any amount passed to the trustee was tripled by the experimenter. The results deviate substantially from standard predictions, supporting both the hypothesis of pure trust on the investors’ side and the hypothesis of trustworthiness on the trustees’ side. With respect to investors’ behavior, the average amount sent was 5.2, with only 2 of 320 investors sending zero, but the amount sent varied substantially across subjects. The trustees returned on average 4.6 (about 1/3 of the tripled amount), and the amount repaid was highly heterogeneous also, with 50% of trustees returning less than $1. The trust game has been often replicated with different subjects and with a fair amount of variation in experimental procedures, either to test for the robustness of the Berg et al. results or to infer about the effectiveness and nature of trust and trustworthiness in different settings with incomplete contracts.

While the first player’s behavior in the trust game allows us to infer whether the investor trusts his or her experimental partner, it is less obvious what we can infer about the trustworthiness of the second player. If the trustee chooses to send a positive amount back, it can signal either altruism (a pure concern for the investor’s payoffs) or positive reciprocity (the desire to be kind to someone who was kind). Comparing the amount sent back in the trust game with dictator allocations of comparable size endowments, James Cox (2004) found support for both altruism and intention-based positive reciprocity. Further experiments have been conducted on the importance of intentions using a similar game to the trust game, called the “gift exchange” game.

As in the trust game, the gift exchange game (Fehr, Kirchsteiger, & Riedl, 1993) represents a situation where contracts are incomplete. It represents the interaction between an employer and a worker, and it was designed to test efficiency wage theory (Akerlof, 1982), according to which firms will offer higher than market clearing wages, expecting that workers will work harder in return. Workers then compare the wage received with a norm they consider fair and choose whether to increase their effort or not, resulting in a positive wage effort relationship.

In the most basic version of the gift exchange game, the employer first decides on an unconditional wage transfer. After observing the wage that he or she will earn, the worker subsequently decides how much effort to supply. Effort increases the profits to the employer, but is also (increasingly) costly to the worker. As an example, suppose that firms earn (q – w)e and workers earn w – c(e), where c(e) is a convex effort cost function over effort levels that range from 0.1 to 1.0. Fehr et al. (1993) implement this experiment in a labor market where there is an excess supply of workers (eight workers for six employers). The market is organized as a one-sided posted offer, in which workers accept or reject offers in a random order. Therefore, if the worker is entirely selfish, he or she will not supply any effort at all, irrespective of the actual wage offered. Anticipating this entirely flat wage effort schedule, the employer offers the lowest possible wage that satisfies the worker’s participation constraint.

Experimental findings sharply contrast these theoretical predictions. Workers are typically willing to supply more effort when a higher wage is offered, yielding a significantly positive correlation between wages and effort, which can be interpreted as positively reciprocal behavior. Results in these games have been highly replicated and appear to be robust and found in various versions of the standard gift exchange game.

In general, the robustness of laboratory results on social preferences has led to the extrapolation of prosocial behavior to a large number of real-world situations. As pointed out by Levitt and List (2007), however, behavior in the laboratory may differ from that observed in real economic environments, depending on the presence of moral and ethical considerations, the nature and extent of scrutiny of one’s actions by others, the stakes of the game, self-selection of the individuals making the decisions, and the context in which the decision is embedded. In general, real-world situations tend to be far more complex than those created in simple laboratory games—for example, workers in reality work in large firms with many other workers, at different hierarchical levels and within a complex payment structure. The possible dependence of fairness and reciprocity considerations on the features of such complex environments could invalidate the direct generalizability of laboratory experimental results to natural markets.

More recently, field experiments have been conducted to address such issues of generalizability and to explore the external validity of laboratory experiments on social preferences. One important example of this endeavor is field experiments on gift exchange. Findings from these field experiments appear to be mixed in terms of their support for social preferences. To capture gift exchange in the field, List (2006) conducted a series of field experiments in a sports card fair, where buyers and sellers of sports cards interacted. The buyer asked the seller for a card of a certain (unverifiable) quality and offered either a high or a low price, and the seller chose the quality of card to provide after seeing this offer. When there was no possibility of reputation building, List did not find reciprocal behavior by sellers: The average card quality provided by nonlocal dealers (who had no incentives to build reputation) was the same for both the buyers who offered a high price and the buyers who offered a low price. On the other hand, in a charity donation context, Falk (2007) found that a small gift included in solicitation mails significantly increases charitable donations.

Another issue that has been explored by field experiments is the time dimension of social behavior, which is a dimension of labor interactions largely ignored in laboratory experiments. Gneezy and List (2006) tested the gift exchange hypothesis in two actual spot labor markets, a library data entry job and a door-to-door fundraising job, and investigated the effect of the duration of the task. Both experiments consisted of a control group that was paid according to the promised wage and a treatment group that was told after recruitment that the wage was raised to a higher level than promised. The authors found that paying more than market-clearing wages had a positive effect on the effort exerted but that the effect was short-lived in both tasks. In particular, treated workers logged more books and raised more money in the initial hours of the tasks, but no significant differences were observed in later hours. Placing the labor relation within a firm instead of a spot labor market, however, Bellemare and Shearer (2009) got different results. They conducted a field gift exchange experiment within a tree-planting firm where workers received a surprise bonus. Here, reciprocity seems to play a role, as workers’ average daily productivity increased by 10%. Moreover, workers’ reciprocal behavior persisted several days after the gift. The mixed results obtained from gift exchange field experiments suggest that reciprocal behavior may be less important in spot markets and more relevant in economic settings where norms of giving apply such as employment relationships and charitable giving.

Market Experiments

Market experiments, for which Vernon Smith received his Nobel Prize, have been very important in the development of experimental economics since these marked one of the first rigorous uses of experiments to understand economic behavior. One main motivation for running market experiments is to test the predictions of competitive equilibrium (the equilibrium price realizes where the demand curve intersects the supply curve), the equilibration process and speed, and market efficiency (the question of whether all potential gains from trade can actually be exploited). Another motivation is to study different “market institutions,” which are specifications of the rules of trading. For example, a retail market could be organized as a “posted-offer” market where one side of the market (e.g., sellers) posts prices. An alternative market institution is a “double-auction mechanism,” where both buyers and sellers could post bids and asks, and all participants can see the highest outstanding bid and the lowest outstanding ask. The design of a typical market experiment provides a direct illustration of induced values. In these experiments, subjects are assigned the roles of buyers and sellers and trade a virtual object. Buyers are assigned “valuations” for the object, whereas sellers are assigned “costs” of selling the object. A buyer’s profit, if he or she buys, is his or her valuation minus the transactions price, and a seller’s profit, if he or she sells, is the price minus his or her cost. The major finding from market experiments is that competitive equilibrium works (i.e., the trade price realizes at the intersection of the induced demand and supply curves) even with only a few buyers and sellers.

Another sales mechanism that has been analyzed extensively using experiments is auctions. These experiments test bidding behavior under different auction formats such as first-price, second-price, English, and Dutch auctions. The usual experimental design involves assignment of valuations to bidders randomly and the bidders submitting bids to win an experimental object in the auction. The profit of the winner is determined according to the auction format— for example, when the auction is a first-price auction, the winner gets an amount equal to his or her valuation minus his or her bid. The experimental auction literature to date has found some robust results regarding bidding behavior, such as bidders overbidding with respect to the risk-neutral Nash equilibrium prediction in first-price auctions, and different models of behavior have been considered to explain these results, such as risk-averse bidding, or bidders who obtain additional utility when they win (“joy of winning,” see Kagel and Levin [2008] for a review). Auction experiments also highlight the feedback between theory and experiments: Experiments test existing theories of bidding, and the results obtained could suggest new models that can more appropriately predict behavior in auctions.

It should also be noted that experiments are also being conducted in fields such as development economics and macroeconomics. Randomized field experiments in development economics have been very useful to test the effectiveness of different policies, such as the use of monetary incentives to increase school performance and attendance (for a review, see Duflo, 2006). Likewise, experimental macroeconomics research has yielded some important insights about the process of equilibration, equilibrium selection, and coordination problems (for a review, see Duffy, 2008).

Years of experimental economics research have also uncovered some differences in behavior driven by individual characteristics. One of these factors is gender. Although men and women behave similarly in many settings, there are some contexts that produce interesting and gender differences. One important context is decision making under risk: Women are found in several studies to be more risk averse than men (Croson & Gneezy, 2008). Likewise, there is some evidence that women are more averse to competitive situations (such as working under a tournament incentive scheme as opposed to an incentive scheme that only depends on one’s own performance) than men (Niederle & Vesterlund, 2007), but whether these differences come from biological (nature explanation) or social factors (nurture explanation) is not determined conclusively yet, and this has been an active area of research in recent years (Gneezy, Leonard, & List, 2009).

Another exciting field of research that adds one other layer to experimental methods in economics is neuroeconomics. Neuroeconomic research allows economists to get at the neural basis of economic decision making, which is especially helpful when there are competing models of behavior that make different assumptions about the decision-making process. In this interdisciplinary field that is rapidly growing, subjects’ brain activity is measured while they make decisions during an experiment. The predominant method of measurement is functional magnetic resonance imaging (fMRI), although positron emission tomography (PET) and electro-encephalography (EEG) have also been employed. The regions of the brain that are “activated” while making a certain choice can provide economists with direct data that can help them understand the motivations behind the observed behavior. An example of this type of result that has received much interest is one that establishes the source of “non-selfish” preferences in the brain. In an ultimatum game study by Sanfey, Rilling, Aronson, Nystrom, and Cohen (2003), subjects presented with unfair offers have been found to have different activation patterns in their brain, depending on whether the offer is coming from another human or a computer. In particular, brain areas associated with negative emotion such as anger “light up” when one faces an unfair offer, and this activation correlates with the subsequent decision to reject the offer. Neuroeconomics research can therefore shed light onto what kinds of decision processes are involved in economic choice and can help economists build models that have assumptions that have a “basis” in the brain.

Conclusion

Given that naturally occurring data are not always sufficient for measuring economic behavior or inference about a variable of interest, controlled experiments provide an invaluable tool for the economist for gathering data. Experimental data can be used for testing existing economic theories, and the results can direct the creation of new ones. For example, many theories of social preferences today have been built to explain experimental findings, and these new models are being increasingly widely applied in areas such as organizational and personnel economics. With the use of field experiments in conjunction with laboratory experiments, it is possible to test economic behavior in a realistic way, obtain insights that can tell us something about the economic behavior of real agents interacting in natural settings, and evaluate the potential effectiveness of different economic policies. We therefore believe that experimentation is likely to continue to establish itself as a standard method of empirical economic research in the years to come.