Measuring Motivations for the Reciprocal Responses Observed in a Simple Dilemma Game
Gary E Bolton, Jordi Brandts and Axel Ockenfels†,*
August 1997
We report on an experiment that uses a simple dilemma game to compare two characterizations of reciprocal behavior that differ on whether it is necessary to account for intentions in order to accurately predict and measure reciprocal actions. The design of the experiment allows us to study both positive and negative reciprocity in a single framework. We find little evidence that intentions matter. Knowing the preferences for relative payoffs is sufficient to predict nearly all the reciprocity we observe.
†
Bolton: Department of Management Science and Information Systems, 303 Beam, Penn State University, University Park, PA 16802, USA;
[email protected]. Brandts: Institut d'Anàlisi Econòmica (CSIC), Campus UAB, 08193 Bellaterra, Barcelona, Spain;
[email protected]. Ockenfels: Faculty of Economics and Management, University of Magdeburg, Postfach 4120 D-39016 Magdeburg, Germany;
[email protected]. * The project was funded by the Spanish DGCICYT (PB92-0593, PB93-0679 and PB94-0663-C03-01). Bolton gratefully acknowledges the Institut d'Anàlisi Econòmica’s financial support that made this collaborative effort possible. We thank Carlos Solà for his able assistance in running the experiment.
1. Introduction: When measuring reciprocity, do we need to account for intentions? There is now wide agreement that reciprocity, of a kind that differs from the standard strategic conception, is an important motivation for the behavior we observe in many laboratory games. One can distinguish two categories. Positive reciprocity, doing something beneficial for someone after being beneficially treated, is perhaps most clearly observed in the gift-exchange game, essentially a prisoner’s dilemma played sequentially. Second movers tend to cooperate after cooperative play by the first mover (e.g., Fehr, Kirchsteiger and Riedl, 1993). Negative reciprocity, doing something harmful to someone after receiving poor treatment, is observed in the ultimatum game, in which a first mover proposes a division of the bargaining pie to a second mover. Offers are sometimes rejected in favor of an outcome in which both bargainers receive nothing (e.g., Güth, Schmittberger and Schwarze, 1982). How this behavior should be modeled is currently unsettled.
Textbook theory
successfully describes reciprocity when it is possible to obtain a future favor from the person at whom the reciprocity is directed. But the kind of reciprocity found in ultimatum and gift exchange experiments, as well as many other lab games where reciprocity plays a role, cannot be explained in this way. In these games, reciprocity is observed even when subjects play with anonymous partners for just one game (and all know they will meet to play for just one game). Under such conditions, there can be no reasonable hope of having a favor returned. What then characterizes the reciprocity we see in these experiments? There are two key issues. First, what triggers the reciprocal action? That is, what factors induce a reciprocator to pass up strictly self-interested action in favor of action that helps or hinders another person, at a cost to oneself? Second, what factors determine the size of the reciprocal response? That is, what do we need to know to accurately measure „reciprocal returns“? In this paper, we present an experiment aimed at comparing two characterizations of reciprocity that are distinct with respect to both triggering and size issues.
Roughly, one
characterization says that well or ill-intended actions trigger reciprocity. Once triggered, the reciprocator awards what he or she considers to be a fair return for those intentions. The other 1
characterization is simpler. It says that the reciprocator just implements her preferences over payoff allocations, independent of any assessment of intentionality. We take as a given that intentions do, at least some of the time, play a role in the cognitive process involved in reciprocating. But we hasten to note that this does not imply that it is necessary to account for intentions in order to predict the reciprocal act. The reciprocator may well think in terms of rewarding the kindness of another, but he may nevertheless assign the same payoff distribution he would in a situation where intentions are irrelevant. The issue we examine is whether, for the purpose of predicting reciprocal action, one need account for intentions, or whether it is sufficient to simply consider preferences over payoff allocations. Our experiment builds on a handful of laboratory studies done by others. Each of the past studies focused exclusively on either positive or negative reciprocity. As we explain in the next section, all of these studies suggest a role for distributional considerations, independent of intentions. Specifically, all the studies find reciprocal behavior that cannot be explained as rewarding intentions. There is some evidence for intentions too, although as we explain, we think this evidence is harder to read. Several features distinguish our experiment. For one, the test of intentions is very direct. We compare reciprocal responses with an otherwise comparable situation, in which one player simply picks a payoff distribution for self and another. This allows us to separate the portion of reciprocal behavior that can be attributed to distributional considerations alone, from the portion that requires an intentionality explanation. A second innovative feature of our experiment is that it allows us to examine both positive and negative reciprocity in a single framework. While positive and negative reciprocity are conceivably distinct phenomena, the two characterizations of reciprocity that we consider treat them as flip sides of the same coin. Our results speak to whether this is reasonable. Third, the test is posed in the context of a simple dilemma game especially designed to study reciprocity. The design minimizes the strategic confounding that might otherwise occur in a regular dilemma game.
2
Our experiment provides a baseline
demonstration of the method; the approach can be extended to other games where reciprocity is an issue.
2. The two hypotheses and the literature We begin this section by stating the precise hypotheses to be tested. We then briefly review the evidence from previous experimental work. 2.1 The intentionality and distributional hypotheses One way to characterize reciprocity is „helping those who help you, and hurting those who hurt you.“1 Because the trigger is helpful or hurtful acts, the reciprocal agent must make an assessment of intentions. That is, the reciprocal agent must interpret whether an action was intended as a helpful or a hurtful gesture. Once determined, the reciprocator responds with what he or she considers a fair return. We call this characterization of reciprocity the intentionality hypothesis. Rabin’s (1993) model is an example of this type of explanation. The model can be applied to various normal form games to explain both positive and negative reciprocal actions. Reciprocity can alternatively be characterized as „balancing accounts.“
By this
characterization, the reciprocator simply acts to implement his or her preferences over payoff distributions given the set of feasible alternatives. This is how Bolton (1991) models negative reciprocity for alternating bargaining games. Bolton and Ockenfels (1997) extend the approach to games involving positive reciprocity.
We call this characterization the distributional
hypothesis. Note that, according to this hypothesis, a player gives or takes from another in accordance to fixed payoff preferences – and regardless of whether the other player has done anything at all. Hence the behavior implied by the distributional hypothesis extends beyond the bounds of what is normally thought of as reciprocal action, to situations in which there is no prior action to respond to. Our experiment exploits this last observation. Of course, there may be some truth in both hypotheses.
It may be the case that
reciprocity is driven in part by distributional preferences, and in part by the desire to reward or 1
We are paraphrasing Rabin (1993), p. 1281. 3
punish intentions. Our experiment provides a measure of the proportion of reciprocal rewards that can be attributed to each motive. 2.2 Previous experimental studies Three previous studies have looked at the nature-of-reciprocity issue. Blount’s (1995) experiment looked for evidence of intentionality in two ways. First, second mover rejections in a standard ultimatum game were compared with rejections in a treatment in which an outside party, receiving no payoff for the game, named the proposal. Second, rejections in the standard treatment were compared with rejections in a treatment in which the proposal was randomly selected. Note that the first mover’s intentions cannot be an issue in either the outside party or the random treatment.
Nevertheless, only the random
treatment was found to have an influence on rejection rates; they were lower than in the standard game. Applied in the most straightforward manner, the intentionality hypothesis would predict that the rejection rate would be lower in both outside party and random treatments, and the distributional hypothesis would predict that the rejection rate would be the same in all three treatments. The experiment contradicts both of these predictions. Hence, Blount’s experiment tells us something important about the limits of both arguments. On the other hand, the experimental conditions add a sufficient number of features – an outside party, a random draw – that both hypotheses can be „extended“ to fit the experimental result without fatal sacrifice. For the intentionality hypothesis, one can relax the definition to allow punishment of one party for the intentions of another.
Blount’s explanation of the
experiment is along these lines. For the distributional hypothesis, one can argue that an act of nature, such as a random draw, changes what people consider to be a fair distribution. A recent paper by Charness (1996), however, complicates matters further. He applies Blount’s framework to study positive reciprocity in the gift exchange game, and finds no difference between outside party and random treatments, and only a mild difference between these and the standard game. Perhaps most importantly, strong evidence for what is usually thought to be the tell-tale sign of reciprocity in gift exchange games, positive correlation between 4
offers and second mover actions, is found in all three treatments. This last result is consistent with the distributional hypothesis but not with the intentionality hypothesis.2 Kagel, Kim and Moser (1996) studied an ultimatum game in which the pie to be divided consists of 100 chips. The information one player had about the other player’s chip value varied from treatment to treatment. The study produced strong evidence that relative payoffs are a factor in second mover rejections. In two treatments, second movers knew both chip values, but first movers knew only their own. (Assigned chip values differed across the two treatments.) The incomplete information makes it more difficult to attribute unfair intentions to the first mover. Hence, by the intentionality hypothesis, we would expect lower rejection rates in the incomplete information treatments than in their complete information counterparts. Rejection rates in one of the incomplete information treatments was indeed statistically lower than in the corresponding full information treatment. In the other case, however, rejection rates were statistically higher. The investigators note that one interpretation of the latter result is that second movers punished first movers for intentionally taking advantage of their own ignorance. But the investigators also note that the money offers rejected tended to be relatively equitable.
This last observation suggests alternative interpretations.
Perhaps, for example,
rejections were an attempt to mislead first movers into believing that chip values were less favorable to second movers than was in fact the case.
Hence, the results might reflect
considerations having to do with the repeated interaction. Taken together, these studies provide substantial evidence that distributional considerations are an important influence, independent of intentionality considerations.
All
provide some evidence that intentions matter too, although we find this evidence less conclusive. The Blount experiment is not easily explained by the standard intentionality argument, and at least some of the evidence from the Kagel et al. experiment is open to a strategic interpretation. Both experiments focus exclusively on negative reciprocity. Charness’s experiment concerns
2
The correlation is not only positive but also very similar in all three treatments. The range of the (highly significant) Spearman rank correlation coefficient between wages and effort is 0.404 (random) to 0.491 (standard game), and between wages and average effort is 0.905 (random) to 1 (third party). 5
positive reciprocity, and there is some evidence for intentionality, although the size of the influence is not entirely clear.
3. Design and procedures for the new experiment Intentions are not directly observable, and hence evidence on intentions is particularly susceptible to confounding with other strategic issues.
We minimize the openings for
confounding interpretations of our experiment by tailoring a dilemma game to the purpose of the study, and by otherwise keeping things very simple. The design provides a direct measure of player preferences over payoff distribution in a setting that is independent of the intentions issue. The game is transparent, and the frame of the experiment, shown to be important to behavior in many lab settings, is held constant across all treatments up to the necessary changes in the payoff matrix. In fact, the sole difference across treatments is one row of the payoff matrix. The data can be fully evaluated with elementary statistics. The design also allows us to investigate both positive and negative reciprocity in a single framework.
Table 1. Payoff rows for the three test matrices (payoffs in Spanish pesetas). C
t
R
m
b
c1
c2
c3
c4
c5
c6
C gets 2050
C gets 2000
C gets 1950
C gets 1900
C gets 1850
C gets 1800
R gets 800
R gets 1000
R gets 1200
R gets 1400
R gets 1600
R gets 1800
C gets 1650
C gets 1600
C gets 1550
C gets 1500
C gets 1450
C gets 1400
R gets 900
R gets 1100
R gets 1300
R gets 1500
R gets 1700
R gets 1900
C gets 1250
C gets 1200
C gets 1150
C gets 1100
C gets 1050
C gets 1000
R gets 1000
R gets 1200
R gets 1400
R gets 1600
R gets 1800
R gets 2000
3.1 The games and the test of the hypotheses The design extends the experiment of Bolton, Brandts and Katok (1996) to a test of the present hypotheses. 6
The experiment revolves around the rows of 2-person payoffs displayed in Table 1. The three treatments in the design involve three ‘adaptations’ of this table, not the table itself. The three adaptations create three different 2x6 matrix games played between the C[olumn] and the R[ow] player. All three games involve row m, and in all cases, playing c1 is a dominant strategy for player C. The game used in the free choice treatment consists of the simultaneous play of the matrix game in which both rows are identical to the m[iddle] row in Table 1. Because the two rows are identical, R’s choice is irrelevant to the final payoff. The game is similar to a dictator game, and shares two important features with dilemma games. First, C has a dominant strategy that corresponds to not contributing to the public good. Second, deviation from the dominant strategy increases the joint payoff. The game used in the positive reciprocity treatment consists of the matrix game with the m and b[ottom] rows of Table 1. In this treatment, the game is not played in the standard simultaneous fashion. Rather, C chooses contingent on R’s choice; that is, C makes two choices, one for each of R’s potential choices. R chooses unconditionally, not knowing C’s decision. The players’strategies combined determine payoffs in the canonical way.3 The negative reciprocity treatment is played in an analogous way to the positive reciprocity treatment. However, the game used in the negative reciprocity treatment is the matrix game consisting of the m and t[op] rows of Table 1. Our analysis centers on comparing C choices for row m across the three treatments. (R player data will be included as well.) The behavior of C in the free choice treatment provides the benchmark for the comparisons. In the free choice treatment, C’s payoff is unaffected by R’s choice. In the positive reciprocity treatment, R’s choosing row m puts C into a relatively better
3
Although standard theory says they should be equivalent, asking a subject for a conditional choice as opposed to asking after he sees his partner's move, can produce modest but significant differences in results. Why this is so is not well understood. A recent paper by Camerer, Knez and Weber (1996) discusses and tests some possibilities. One potential explanation is that people have a hot (emotional) reaction when reacting to a partner’s move, whereas they are more cool (calculating) when making a conditional choice. Assuming that hot reactions are relatively short lived, it is unclear whether hot or cool is the more economically relevant. Obviously our experiment is posed in a cool state. 7
position than if R had opted for row b. In the negative reciprocity treatment, R choosing m is less favorable to C than R choosing t. In sum, the design allows us to observe the distribution of C’s decision in row m while varying the intentions of R (helpful, irrelevant, or hurtful). Two specific features of the payoff table are worth noting. First, the equal split cell was deliberately located in the middle of row m, since C might conceivably choose the equal split in the free choice treatment, and want to add a ‘kindness bonus’ in the positive reciprocity treatment. To detect this behavior, the equal split need be somewhere other than the right-hand end of row m. Second, the matrix is designed so that the difference in C’s payoff when moving one row is large, 400, relative to the difference in R’s payoff when moving one column, 50. This affords C a fairly large variety of choices in rewarding R for a relatively generous act.4 The test for evidence of intentionality is very direct: C’s observed levels of contributions provide evidence for intentionality in positive reciprocity if contributions are larger in row m of the positive reciprocity treatment than in the free choice treatment.
Analogously, there is
evidence for intentionality in negative reciprocity if contributions are larger in the free choice treatment than in the m row of the negative reciprocity treatment. In both cases, we can measure the portion of reciprocity due to distributional considerations, and the portion due to intentions. If, however, contribution levels are positive but equal across the m rows of the three treatments, then we cannot reject the distributional hypothesis as a complete explanation for C contributions in these games. 3.2 Sample sizes, subject pool and procedures Each treatment had two sessions. Each session involved from 16 to 18 subjects (always an even number). There were a total of 34 subjects in the free choice treatment, and 36 in each of the other treatments. Sessions were run at the Universitat Autònoma de Barcelona. Subjects were recruited from undergraduate classes. Participation required appearing at a special place and time, and was restricted to one session. The chance to earn cash was the only incentive. 4
To see the point, suppose we had restricted C to just two choices, say 1 and 6. Then when R chooses m in the positive reciprocity treatment, we run the risk of observing all C's playing 1, simply because no C wants to reward as much as 250 to R. Our design minimizes the chance of this problem by affording C a fairly wide range of gifts. 8
Subjects were seated so that they could not observe other's choices. The lab protocol (Appendix A) includes all verbal and written communication between monitor and subjects save for individual subject questions-and-answers. The same experimenter was the communicating monitor for all sessions (others silently managed the paper flow). Instructions were read aloud. The payoff matrix was displayed on a blackboard, and the monitor explained each cell. Subjects played two games without knowing which one would be chosen for payment. Players were never matched with the same partner twice (and this was common knowledge). Otherwise, matching was anonymous and randomly selected. C and R roles were alternated between games. Once both games were complete, a coin flip chose one game for payoff. To preserve anonymity, those with the R role for the chosen game were taken to another room and paid separately. Subjects received no fee beyond earnings for the game. Average subject earnings were 1334 pesetas (about $9.34), and a standard deviation of 377 pesetas ($5.38).5 The experiment had a "no feedback" design: subjects were not informed of the outcome of the first game prior to playing the second. This arrangement has two advantageous features. First, order and learning effects are not an issue; in this sense each game is 'single play.' Second, roles are rotated between games, allowing us to gather player C data from each subject. These features were held constant across all treatments.6
5
All treatments were run in February and March of 1997. Each session lasted about 45 minutes. In the context of our game, the double blind hypothesis attributes subject contributions to experimenter observation. Laury, Walker and Williams (1995) performed a direct test of double blind in a public goods game, and found no supporting evidence. Evidence on double blind for the dictator game is mixed; Roth (1995) reviews the experiments and critiques the evidence in favor of the hypothesis. 6
9
Table 2. Data set from the experiment.
Positive C-m,1 C-b,2 1 1 1 1 1 1 6 6 4 2 1 1 1 1 6 6 4 2 6 6 1 1 4 2 6 3 1 1 1 1 1 1 4 4 1 1 1 1 1 1 4 2 1 1 4 2 3 6 1 1 1 1 1 1 1 1 1 1 1 1 4 3 1 1 1 1 1 1 3 2 1 1 mean std dev n
2.3 1.8
1.9 1.6
R 2 2 2 1 2 2 2 1 2 2 2 2 1 2 2 1 2 2 2 2 2 2 2 2 2 2 1 2 2 2 2 2 2 2 1 2
C-t,1 1 6 6 6 6 6 1 1 1 6 1 1 1 1 6 6 1 1 1 2 1 1 1 5 6 1 1 1 1 1 1 1 6 1 1 5
1.8 0.4 36
2.6 2.3
Negative C-m,2 R 1 2 6 2 6 1 6 2 1 2 6 2 1 2 1 2 1 2 4 2 1 1 1 1 1 2 1 2 1 2 4 2 1 2 1 2 1 2 1 2 1 2 1 2 2 2 4 2 6 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 4 2 1 2 1 2 5 2 2.2 1.9
1.9 0.3 36
Free C 3 1 1 1 1 1 1 6 1 1 4 1 1 6 2 6 1 1 5 2 6 1 1 1 4 2 4 5 1 4 1 5 4 1
R 1 1 2 1 2 1 2 1 1 1 2 2 1 1 2 1 2 1 1 2 2 2 1 1 2 2 2 2 1 1 2 2 1 1
2.5 1.9
1.5 0.5 34
Each row of each table displays responses for one subject, for both C and R roles. C-x refers to Cs response when R chooses the row labeled x in Table 1. R choice labels refer to the ‘1’ and ‘2’ in the C choice columns; e.g., in the positive reciprocity treatment, ‘1’refers to R choosing m.
10
4. Results from the new experiment Table 2 exhibits the entire data set, together with some descriptive statistics. We first demonstrate that the experiment produces reciprocal behavior. We then turn to a comparison of the two main hypotheses. Within the context of the present games, reciprocal behavior is commonly understood to mean that C contributes more when R contributes greater than the minimum. For instance, in the positive reciprocity treatment, players C should tend to contribute more when R chooses m than when R chooses b. Table 2 exhibits evidence for reciprocal behavior within both non-free choice treatments: many C’s contribute more when R contributes more than the minimum, and only rarely in this situation do C’s contribute less (only once in each treatment).
Within both
treatments, the larger contributions are statistically significant (matched pair test, one-tail p-value = 0.024 and 0.013, for positive and negative treatments respectively).
So the experiment
produces reciprocal behavior within both positive and negative reciprocity treatments.
Figure 1. Cumulative distributions showing Column player responses to Row playing m.
1 0,9 0,8 0,7 0,6 0,5 1
2
3
4
pos
free
5
6
neg
We next turn to the issue of whether the reciprocity we observe is best understood in terms of distribution, rewarding intentions, or some combination of the two. Note from Table 2
11
that a fair number of C’s contribute greater than the minimum even when R fails to do so. This indicates there is at least some distributional component to contributions. In order to determine what proportion of the behavior can be attributed to rewarding intentions, we turn to a comparison of the m row across the three treatments. Figure 1 compares the relevant cumulative distributions.
No sharp difference between the three treatments is
evident. A test for a locational difference using the Kruskal-Wallis test (essentially a three sample Mann-Whitney test) finds no evidence for a difference across treatments (p = 0.874). We then look for evidence of any general difference in distributions using an empirical χ2-test, and again we find no evidence for a difference (p = 0.348).7 So we cannot reject the hypothesis that contributions in the three m rows are drawn from the same distribution. Pairwise comparisons lead to the same conclusion. Hence our statistical tests imply that all of the observed reciprocity is attributable to the distributional hypothesis. Of course, finding for the null hypothesis always raises the issue of whether the test has sufficient power.
Look again at Figure 1 and note that the positive reciprocity treatment
dominates the free choice treatment – in the wrong direction relative to the intentionality hypothesis. So there is no evidence of any kind – descriptive or inferential – for intentionality in positive reciprocity.
On the other hand, average contributions for the negative reciprocity
treatment do go in the right direction (Table 2). There is, however, no dominance in the Figure 1 graph, and the size of the difference in averages is small. Specifically, the difference in average contributions is equal to 12 percent of the contributions in the free treatment. So even if we were to ignore the inference results and confine ourselves to descriptive statistics, we would still conclude that distribution is about nine times as important to negative reciprocity as is intentions.
5. Summary: Preferences over distribution are the major motivation for contributing We began with two hypotheses, each offering an account of reciprocal behavior. The intentionality hypothesis asserts that reciprocity is triggered by kind or unkind acts. The 7
The p-value is the average of five 20,000 trial samplings of the contingency table distribution. 12
distributional hypothesis says that the reciprocator simply acts to implement her preferences over payoff allocations. The design involved three 2x6 matrix games, each differing from one another by one row of payoffs, and all having one row in common. In all three games, the column player had the same dominant strategy. For the free choice game, the two rows were identical. This game provides a direct measure of column player preferences over payoff allocations. We gauged the residual influence of intentions by comparing free choices to choices made when the variable row is one that is more (less) favorable to the column player. Within each of the non-free choice treatments, we found evidence that column players are on average more generous when the row player is kind then when he is unkind. However, in the positive reciprocity treatment, in which the fixed row is the one most beneficial to the column player, the amount of this generosity was fully accounted for by distributional considerations. There is neither statistical nor descriptive evidence for intentionality playing any role beyond what distributional considerations can account. For the negative reciprocity treatment, in which the fixed row is the one least beneficial to the column player, contributions were slightly lower than in the free choice treatment, but we were unable to reject the hypothesis that distributional considerations accounted for all of the observed negative reciprocity. In sum, our experiment produces little evidence for the necessity of accounting for intentions in order to measure reciprocity. We find that distributional preferences are sufficient to predict reciprocal acts. Finally, subjects in our experiment each played with an anonymous partner, never the same partner more than a single time. Consequently, people reciprocate – award or punish others at a cost to themselves – without any reasonable chance of getting a return directly from the person at whom the reciprocity was directed. One might conjecture that this behavior is altruistic, that people are paying to improve (or diminish) the welfare of others. There is, however, evidence against this explanation; see for example, the experiment of Selten and Ockenfels (forthcoming).
An alternative explanation is that the reciprocator expects his
contribution to be returned – to be in a sense reciprocated – sometime in the future. This would 13
represent a kind of indirect reciprocity, in which a person contributes to the welfare of ‘the group’ with the expectation that the group will contribute to her welfare sometime in the future. This line of reasoning naturally suggests an evolutionary foundation – in which negative reciprocity is the way nature mitigates the free rider problem inherent in an indirect reciprocity arrangement. Güth, (1995), Huck and Oechssler, (1995), and Kockesen, Ok and Sethi (1997) all take theoretical steps in this direction.
14
References
Blount, Sally (1995), „When Social Outcomes Aren’t Fair: The Effect of Causal Attributions on Preferences,“ Organizational Behavior and Human Decision Processes, 63, 131-144.
Bolton, Gary E (1991), „A Comparative Model of Bargaining: Theory and Evidence,“ American Economic Review, 81, 1096-1136.
Bolton, Gary E, Jordi Brandts and Elena Katok (1996), „A Simple Test of Explanations for Contributions in Dilemma Games,“ discussion paper 360.96, Institut d’Analisi Economica, Barcelona, revised April 1997.
Bolton, Gary E and Axel Ockenfels (1997), „ERC: A Theory of Equity, Reciprocity and Competition,“ working paper, Penn State University.
Camerer, Colin, Marc Knez, and Roberto Weber (1996), "Virtual Observability," working paper, Cal Tech.
Charness, Gary (1996), „Attribution and Reciprocity in a Simulated Labor Market: An Experimental Investigation,“ working paper, UC Berkeley.
Fehr, Ernst, Georg Kirchsteiger and Arno Riedl (1993), „Does Fairness Prevent Market Clearing: An Experimental Investigation,“ Quarterly Journal of Economics, 108, 437-459.
Güth, Werner (1995), „An Evolutionary Approach to Explaining Cooperative Behavior by Reciprocal Incentives,“ International Journal of Game Theory, 24, 323-344.
Güth, Werner, R. Schmittberger and B. Schwarze (1982), „An Experimental Analysis of Ultimatum Bargaining,“ Journal of Economic Behavior and Organization, 3, 367-88.
Huck, Steffen and Jörg Oechssler (1995), The Indirect Evolutionary Approach to Explaining Fair Allocations, working paper, Humboldt University. 15
Kagel, John, Chung Kim and Donald Moser (1996), „Fairness in Ultimatum Games with Asymmetric Information and Asymmetric Payoffs,“ Games and Economic Behavior, 13, 100-110.
Kockesen, Levent, Efe A. Ok, and Rajiv Sethi (1997), „Interdependent Preference Formation,“ working paper, Barnard College, Columbia University.
Laury, Susan K., James M. Walker and Arlington W. Williams (1995), „Anonymity and the Voluntary Provision of Public Goods,“ Journal of Economic Behavior and Organization, 27, 1995, 365-80.
Rabin, Matthew (1993), „Incorporating Fairness into Game Theory and Economics,“ American Economic Review, 83, 1281-1302.
Roth, Alvin E. (1995), Bargaining Experiments, in Handbook of Experimental Economics (J. Kagel and A. E. Roth, eds.), Princeton: Princeton University Press.
Selten, Reinhard, and Axel Ockenfels (forthcoming), „An Experimental Solidarity Game,“ Journal of Economic Behavior and Organization.
16
Appendix A: Laboratory Protocol This section contains a description of all procedures, as well as verbal and written instructions given to subjects for the free choice treatment. The monitor read all verbal instructions directly from the protocol. The only monitorsubject communication not included in the protocol are answers to individual subject questions. The positive and negative reciprocity treatments differed only as indicated on written instructions and task forms. The protocol has been translated from Spanish. Seating. Upon entering the room each participant is randomly directed to a seat. Half the seats have red folders and half have blue. Red folder seats on the right side of the room, and blue folder seats on the left side. May I have your attention please. We are ready to begin. Thank you for coming. With the exception of the folder, please remove all materials from your desk. Open your folder and take out the sheet marked 'Instructions'. At this time please read the instructions. Participants read silently. Written instructions begin here: Instructions General. The purpose of this session is to study how people make decisions in a particular situation. Feel free to ask a monitor questions as they arise. From now until the end of the session, unauthorized communication of any nature with other participants is prohibited. During the session you will make money. Upon completion of the session the amount you make will be paid to you in cash. Payments are confidential: no other participant will be told the amount of money you make. Half the participants were given blue folders and half red folders. During the session, you will be paired with a person having a different color folder than yours. No one, however, will know the identity of the person they are paired with. Nor will these identities be revealed after the session is complete. Decision Task. In each pair, one person will have the role of A, and the other will have the role of B. The amount of money you earn depends on the decision you make and the decision of the person you are paired with. The Earnings Table below describes the options available to each person and the associated earnings. You make your decision by choosing one of the options available to you and recording it on a paper form. The person in role A can choose from options A1 through A6. The person in role B can choose option B1 or B2. Each person makes their decision without knowing the decision of the other. [In the positive and negative reciprocity treatments: B decides unconditionally; that is, he simply chooses between B1 and B2. A chooses conditionally; that is, A indicates a decision for the case where B has chosen B1 and a decision for the case where B has chosen B2. The decision of A that counts is the one that corresponds to the decision of B.] Each person receives the earnings in the Earnings Table cell corresponding to the chosen options. (Earnings table adapted from Table 1 appeared here.) Conduct of the Session. You will participate in two decision tasks, called Task 1 and Task 2. Both tasks are identical to the description in the previous paragraph. For each task, you will be paired with a different person. You will have the role of A in one task, and the role of B in the other. In task 1, those with red folders will have the role
17
of A, and those with blue folders will have the role of B. In task 2, blue folders are A, and red folders are B. First, you will receive a decision form for the role you have in task 1. You will complete task 1, and all the forms will be collected. You will then receive a decision form for task 2 and complete task 2. The results for task 1 will not be revealed prior to completion of task 2. Payment. You will actually be paid your earnings for just one of the two tasks. The one for payment will be chosen by a coin flip after both tasks have been completed. In order to preserve anonymity, the participants who are B for the chosen task will be taken to a second room and paid separately. Once you are paid, you may leave. Written instructions end here. I will now read the instructions out loud. Read instructions. Are there any questions? Performing the decision task. Pass out decision forms for task 1. We are ready to begin decision task 1. You make your decision by filling out the form that is now being handed out. Those with red folders will receive the form for role A, and those with blue folders will receive the form for role B. Please review the form with me. Written decision form for A begins here (form for B was analogous). Distribution Task 1 You have the role of A. The person you have been paired with has the role of B. You must make your decision by choosing one of the options available to you and described in the earnings table below. (Earnings table adapted from Table 1 appeared here.) Use the information in the earnings table to make your decision by circling the number of your chosen option. A’s Decision: Option
A1 A2 A3 A4 A5 A6
[For pos. & neg. treatments substitute: A’s Decision if B chooses B1: Option A’s Decision if B chooses B2: Option
A1 A2 A3 A4 A5 A6 A1 A2 A3 A4 A5 A6 ]
When finished, please turn this sheet over and wait quietly. Written decision form ends here. Are there any questions?...Please fill out the form. Again, when you are finished, turn the form over and wait quietly. Wait until all are finished. A monitor will now come around to collect the forms. Collect the forms. We are now ready to begin decision task 2. Remember, you will be paired with a different person than you were in task 1. Hand out forms for task 2. Are there any questions?...Please fill out the form. Again, when you are finished, turn the form over and wait quietly. Wait until all are finished. A monitor will now come around to collect the forms. Collect the forms A coin is flipped to determine roles. Those in role B are taken to a different room and paid separately after those in role A.
18
