Cheap talk in the chicken game

Cheap Talk in the Chicken Game

By: Marc Kramer

Student Number: 11131829

Supervisor: Arthur Schram

University of Amsterdam

Economics: Game Theory & Behavioural Economics

15 ECTS

August 2017

Abstract

An experiment is performed where subjects play a one-shot, two-player Chicken game. In some of the games, one of the subjects is allowed to use one-shot, one-way cheap talk. The goal of this paper is to find out if signallers of cheap talk are at an advantage compared to receivers of cheap talk or players who play the game without any form of communication. This paper finds that there is no statistically significant benefit (or cost) to being a signaller, receiver or neutral player in the Chicken game. However, by loosening the strict

communication rules practised in this experiment, findings might turn out differently. However, it should be noted that the origin of this effect might be found outside the field of Game Theory.

Table of Contents

5.1 Mean and Standard Deviation 13

5.2 OLS Regressions 14

6.0 Conclusion 19

6.1 Discussion 19

7.0 Bibliography 21

8.0 Appendices 23

8.1 Appendix I: Control group form 23

8.2 Appendix II: Test group form 24

This document is written by Student Marc Kramer, who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

1.1 Motivation

Every single day individuals are faced with strategic situations out of which they want to gain the optimal outcome for themselves. Think of political parties negotiating about the policy for the coming term, a manager and an employee discussing terms of an employment contract, or even countries who are in peace negotiations. In our everyday life, the right use of communication may prove a powerful weapon in these strategic situations with conflicting interests. Understanding how messages sent by individuals in these strategic situations could influence the choices made by the receivers of these messages is understandably of crucial importance for previously mentioned economical and political issues. This is why in this paper I will try to find out if sending messages benefits the sender in a strategic situation with conflicting interests.

1.2 Research Question

From a theoretic point of view, cheap talk should not have any effect in an anti-coordination game (Farrell & Rabin, 1996). Since the signal might be as well false as it may be true, the receiver of the signal should not place any value on it and act as if the signal was never sent. In practise, however, it might be sometimes that using cheap talk benefits the sender in an anti-coordination game. Think of the goalkeeper in soccer who points to which side he is going to dive during a penalty kick, or the politician who threatens to alter his vote if some of her criteria are not met, or a firm threatening to take their business to a competitor of the firm they are bargaining with. The variance between theory and some strategic

situations with conflicting interests observed in reality is the motivation for this paper. In this paper I will be testing if cheap talk benefits the sender in an anti-coordination game during a laboratory experiment. My experimental design will be using the Chicken game, because in this game there are two pure Nash equilibria that yield unequal payoffs for the two players and there is no Nash equilibrium with equal payoffs for both players. The cheap talk used in the experiment will be one-shot, so if cheap talk is used it will be only used by one player per game. The receiver of this cheap talk does not get an opportunity to communicate an answer. The focus will be on the effect this one-shot cheap talk has on the payoffs of both signaller and receiver, which will also be compared to the payoffs players receive from playing the Chicken game without any cheap talk or other communication whatsoever. The research question of this paper is thus constructed as follows:

To answer this question an experiment will be run with both a treatment group, as well as a control group. The results of this analysis will be used in order to answer the research question; consequently the conclusions made will be used as a setup for future research.

Besides this main research question, this paper will also study the different results between the different signals subjects can receive. Thus, the sub-research question of this paper is:

Does it matter what claims the signaller makes in his or her cheap talk in the Chicken Game?

This paper tries to answer these questions via the experiment, which is elaborately described in the Experimental Design chapter. This is how this paper tries to shine a new light on the effect of cheap talk in a game with conflicting interests and possibly open doors to future research.

1.3 Thesis Setup

To answer the research questions of this paper, first I will have a short literature overview. In this overview, I will have a look at existing literature and pinpoint this paper’s contribution to the field. After the literature overview I will present my hypotheses on the possible answers of the research questions. Then, the experimental setup will be described, after which the statistical results will be shown and I will draw my conclusions. Consequently I provide a discussion section in which limitations of this research will be discussed; also some possible openings for future research will be coined. Finally you can find the bibliography and two appendices, which respectively are the instructional forms both the control groups as well as the treatment groups received.

2.0 Literature Overview

Farrell & Rabin (1996) provide a clear definition cheap talk: Cheap talk consists of messages that are costless and non-binding to the sender, and not verifiable to the receiver. Nevertheless, these messages may affect the receiver's beliefs about the planned actions of the sender. Importantly, and key to this paper, cheap talk might influence choices made by

players. Thus cheap talk can possibly alter choices, but it is important to emphasize that the addition of cheap talk to a game does not exclude any of the equilibrium outcomes of the original game (Blume & Ortmann (2007). This is because any message sent via the cheap talk can simply be ignored. This follows directly from the above-mentioned definition of cheap talk by Farrell & Rabin (1996): Such messages are costless and non-binding to the sender, as well as non-verifiable to the receiver. Croson, Boles & Murnighan (2003) found, however, that cheap talk, in contrast with common theory, can influence bargaining situations.

Blume & Ortmann (2007) also conclude that repeated interactions with cheap talk can help surmount issues such as strategic uncertainty, equilibrium selection, and coordination failure in games. Bernheim, Peleg & Whinston (1987) already described how cheap talk could overcome strategic uncertainty, as they stressed the potential importance of pre-game, non-binding communication. They did so by hypothesizing about a game where two players can either choose "heads" or "tails". If their answers accord they both receive a payoff of one, if their answers do not, however, they both gain a payoff of zero. In this setup, pre-game communication can be non-binding, but aiding in reaching one of the two possible pure Nash equilibria, maximizing both players' payoff. This state can be reached because, even though the pre-game communication is non-binding, it is self-enforcing. Neither player gains any payoff by deviating from the reached agreement.

The game discussed by Bernheim, Peleg & Whinston (1987) fits perfectly into the kind of games that Charness (2000) reviews: Coordination games featuring multiple Nash equilibria. These games frequently have an efficient outcome, which usually only can be achieved via coordination. A great example Charness (2000) discusses is the classical Stag-Hunt game. A game of which Duffy & Feltovich (2002) also stated that cheap talk could prove a powerful tool, since messages are usually thought to be credible. In his experiment Charness (2000) sets up a Stag-Hunt game where one of the players either sends a one-way signal before making a choice, after which the other players choose an action. Or one of the players chooses an action, consequently sends a one-way signal to the other players, who in their turn choose an action. Charness (2000) finds that if the signal precedes the action, coordination and efficiency are rather high. However, the latter are considerably lower when

the order of signal-action is reversed. He hypothesizes this might be because the self-interest concern players have in the Stag-Hunt game might be more highlighted, causing a reduced credibility of the signal. But either way, both combinations generate higher coordination and efficiency than games without any communication.

Crawford & Sobel (1982) set up a model, which found that in a two-player game, direct communication plays a more essential role, the more closely related are if players' objectives. Their model also suggests that perfect communication is not to be expected unless players' preferences completely correspond. In a more recent paper, Crawford (1997) states that despite the term "cheap talk", messages without any direct payoff implications can possess interesting information to the other player when both players' interests are not too far apart. Interestingly enough, Cooper et al. (1989) found that non-binding pre-play

communication in the Battle of the Sexes game could improve the average payoff of both players. What is even more interesting is that one-way communication was found to be more efficient than two-way communication. Although the side note should be made, that one-way pre-play communication before the classical Battle of the Sexes game almost turns the game into an Ultimatum game, where the signaller simply gives his or her opponent the opportunity to choose for the option where the receiver at least gets some payoff, instead of nothing at all. The Chicken game used for this paper is a game with complete information. The only two outcomes that are pure Nash equilibria are if one player chooses Yield and the other player chooses Continue. When subjects make either the decision Yield or Continue, they basically decide between a 'safe' option and an 'all-or-nothing' option. In the Chicken game set up for this experiment, only the option where both subjects play Continue is Pareto

dominated; all other outcomes are not Pareto dominated; hence they are Pareto optimal (Voorneveld, 2003). Cooper et al. (1992) found that one-way pre-play communication (i.e. cheap talk) increases the play of the Pareto-dominant equilibrium strategies in coordination games. The latter means that one player cannot increase his or her own payoff, without lowering the payoff of the opposing player. This is an essential characteristic of the Chicken Game, which makes it so useful for this research. Since it really puts focus on the conflicting interests of both players.

In summary, the existing literature mainly finds that cheap talk could improve efficiency in coordination games. But there is very little experimental material, which tests what happens when subjects use cheap talk while facing a game with conflicting interests. Thus, where existing literature mainly focuses on cheap talk in coordination games, this paper

will focus on the effect of one-shot, one-way cheap talk in a game with conflicting interests, by setting up an experiment.

3.0 Hypotheses

Before I will state my hypotheses, it is

important to first formally introduce the Chicken Game that will be used during the experimental sessions. The

Chicken Game is a one-shot, two-player game in which the two players have conflicting interests. The payoffs of the game are shown in Figure 1. As can be seen there, the game offers the following possibilities:

i. If both players choose Yield, both players get a payoff of 2.

ii. If one player chooses Continue and the opposing player chooses Yield, the former player gets a payoff of 3; the latter player gets a payoff of 1.

iii. If both players choose Continue, both players get a payoff of 0.

Observing the possible outcomes of the game, it can be concluded that the only two pure Nash equilibria are when one player chooses Continue whilst the opposing player chooses Yield.

Neither the payoffs, nor the pure Nash equilibria alter when cheap talk is introduced to the Chicken Game. Focus will be merely on whether or not the cheap talk influences the payoffs of senders and receivers respectively.

It will be interesting to see what results come out of the experimental sessions, but based on existing literature I do not expect to find a significant effect of the cheap talk that will benefit the sender. This hypothesis is mainly based on the model of Crawford & Sobel (1982), which found that the more closely related the goals of players are in a game, the more efficient communication could be to reach a more optimal goal for them. It is also backed by the findings of Farrell (1987), who found that even minor conflicting interests prevent players from achieving perfect coordination. Thus, the first hypothesis reads:

Hypothesis I: There will be no statistically significant beneficial effect for signallers during the Chicken game.

Since in the Chicken game set up for this experiment, the interests of the players are quite far apart with the only two pure Nash equilibria being conflicting, I do not expect statistically significant effects. Considering the impact of the claims a signaller makes within the cheap talk he or she uses, I again do not expect any statistically significant effect. Hence, the second hypothesis states:

Chicken Game Yield Continue

Yield 2,2 1,3

Continue 3,1 0,0

Hypothesis II: The claim a signaller makes in his or her cheap talk does not have a statistically significant effect on the payoff he or she receives.

The second hypothesis is mainly based on the same literature as the first hypothesis is. As Crawford & Sobel (1982) find with their model that communication get less effective when interest are less aligned, what is said within this communication should not matter anymore, as communication itself is losing its effect.

4.0 Experimental Design

Subjects were gathered from the (wider) social circle of the author. They were family members, friends and co-workers; I deliberately chose to not use any fellow students, since their history with game theory might cause noise in the results of the experiment. The subjects participated voluntarily and it was critical that they had no previous knowledge about the goal of the experiment. Once the needed amount of participants could be gathered at the same time, they were placed in a secluded room, where there was no communication prior or during the experiment between each other.

This experimental design uses a one-shot Chicken game. Seven rounds are played with random re-pairing after each round. A between subject design is used. In half of the sessions one of the subjects in each game is randomly selected to send a costless message (i.e. use cheap talk). In this message a subject states a claim with respect to the action he or she is going to play in the next game. This analysis focuses not only on the effect the mere use of cheap talk has on the payoff of the sender. Focus will be also on what subjects say with their cheap talk, and consequently how the receiver responds on either possible message.

The game is played with complete information. So in the experiment all subjects have full information of all possible outcomes and payoffs. The payoffs of the game are shown in Table 1. Note that in the classical chicken game the payoffs of (Continue, Continue) are usually negative, but in this experiment subjects obviously cannot be asked to pay the experimenter if they are selected for payment after the experiment. This is why the payoff of (Continue, Continue) is nonnegative in this experiment. All other possible payoffs are

adjusted accordingly in order to not alter the strategic situation the Chicken Game represents. This experimental design tries to control the environment as much as possible, in order to check for the difference in results of one-shot chicken games where some subjects are allowed to use cheap talk, and others are not. To achieve this, subjects are not allowed to look, talk, or communicate in any way with each other. All messages sent will go via the experimenter. To place the emphasis as much as possible on the effect of cheap talk, subjects are paired with a different other player in each round; this will be done anonymously

throughout (and after) the whole experiment. The games are anonymous. This is important, because subjects who have already interacted with each other for half an hour have

substantially higher odds to correctly predict what their opponent is going to play (Frank, Gilovich & Regan 1993). All interactions in the experiment were anonymous.

As an incentive, at the end of every session one participant is selected for monetary reward according to his/her payoffs summed up after the seven rounds all subjects have

played. The motivation behind the decision for this reward system will be discoursed in the Discussion section of this paper.

All participants are aware they play seven games against seven unique opponents. Subjects are not communicated what their payoff is after each round; instead they will get this information after they have played all of their seven rounds. Any time a costless message is sent, this will go via the experimenter. Hence, any subject chosen to act as signaller writes down on his or her form what message he or she wants to send to his or her opponent. Consequently, the experimenter writes down the chosen message on the form of the subject chosen to act as receiver. This message is either Yield or Continue. Communication goes via the experimenter so no non-verbal signals can be sent between subjects. Plus, subjects stay unaware of with whom they are playing. During the experiment the Chicken game will be referred to as 'Game' and the message-options 'Yield' and 'Continue' will be named 'A' and 'B'. This is done to keep the experimental design as neutral as possible. See Appendix I and Appendix II to see the exact instructions that the subjects received.

The total experiment exists of two sessions with fourteen participants each, where seven subjects are selected to send a cheap talk message to seven receivers. As a control group there were also two sessions with eight participants. In the latter two sessions no cheap talk was used whatsoever, so since subjects didn't have a particular role in the game (i.e. signaller or receiver), fewer participants were needed, while each participant still played seven games against seven unique opponents. The group of participants consists of both students and non-students. Ages ranged from 22 to 72 and both male and female subjects played in all experimental sessions. Participation is voluntary and after each session one of the subjects is randomly selected for cash payout. This payout is the sum of payoffs the subject generated during the seven games he or she has played. This means the total payoff a subject could possibly gather ranges from 0 to 21 euro.

An experimental session lasts about ten to fifteen minutes. During an experimental session, subject are not looking nor talking to one another. After they have been given an oral and written overview of the experimental procedure subjects are granted the opportunity to ask any questions they may still have. Besides this specific moment to ask questions, subjects are allowed to ask the experimenter for clarification throughout the whole experiment.

During each session each subject plays seven games of chicken against seven unique opponents. This generates a total of 49 games played per session with cheap talk and 28 games played per session without any cheap talk. This adds up to 154 games played during 4 sessions. In total this gives 308 payoffs, which are used as observations during the statistical

analysis. Concerning the 308 observations, it should obviously be noted that each subject makes seven consecutive decisions; hence those are not statistically independent. This is controlled for by clustering the standard errors. I.e. the seven choices made by a single subject are clustered together.

5.0 Results

The dataset consists of 308 observations. Each observation represents the payoff an individual subject gained from playing one game. This means each subject contributed to this research with seven observations. Of course, the latter had to be controlled for. Since, though the outcomes of other games played by a certain subject did not affect the payoff in the given game at all, the seven choices made by a subject cannot be viewed as statistically

independent. To control for this non-independence, the regressions performed in the statistical analysis were corrected for clustered standard errors.So during the regression analysis, when checking for the effect of the role of senders, receivers or neutral players could be, or what signal was sent during a game, the choices and options a subject had gotten during the six other games he or she played were taken into account.

5.1 Mean and Standard Deviation

Starting the statistical analysis, first the means and standard deviations of the main variables will be briefly discussed. Consequently, a number of regressions is run to check for any statistically significant effects of the role of subjects in the Chicken Game on their payoff, as well as the effect a certain signal has on the payoff of both sender as well as the receiver.

Kind of Player Observations Mean Payoff Standard Deviation

Neutral Player 112 1.75 1.00

Signaller 98 1.55 1.14

Receiver 98 1.46 1.09

Total 308 1.59 1.08

Table 2: Statistics per kind of player

With 308 observations and, of course, payoffs ranging from 0 to 3, the mean payoff of all subjects in a single round game was 1.59, with a standard deviation of 1.08. As can be seen in Table 2.

During the experimental sessions with the neutral players (i.e. no cheap talk), the players gathered a mean payoff of 1.75 in a single round game played, with a standard deviation of exactly 1.00, which is also shown in Table 1.

Receivers got the lowest mean payoff per single round game played of the three divisions of players. Receivers got a mean payoff of 1.46, with a standard deviation of 1.09, which can be found in Table 1.

Looking at the three different kinds of players observed in the experimental sessions, it shows that the neutral players (i.e. the players without any cheap talk) have the highest mean

payoff, with the lowest standard deviation. Both signallers and receivers follow (in that order) on quite some distance. This suggests that if there even is an effect of using cheap talk during the Chicken game, it might be negative compared to not using cheap talk at all. To check if this is statistically true, section 5.2 will observe some regressions performed on the dataset. When subjects were allowed to use cheap talk, it is interesting to see that a (small) majority of the messages sent did not comply with the actual action played by the particular subject. 53.06 per cent of the signallers sent a message that did not comply with what they actually played, the other 46.94 per cent of the signallers played the same action as they messaged their opponent. 60.20 per cent of the signallers messaged they would play A, the other 39.80 per cent of the signallers messaged they would play B. Eventually, 47.96 per cent of them played A, where the other 52.04 per cent of the signallers played B. This got

signallers a mean payoff of 1.55 per round, with a standard deviation of 1.14.

5.2 OLS Regressions

For the regression analysis, the subjects’ payoff in a single round was used as the independent variable. Whether a subject played A (i.e. Yield) or B (i.e. Continue), whether a subject was a signaller, receiver or neutral player and what signal a subject either had sent or received were all regressed on the payoff the subjects collected. The results of these

regressions are discussed below.

What is played: A (i.e. Yield) or B (i.e. Continue)

As can be seen in Table 3, the OLS regression suggests that subjects who played B, independent of their specified role, gathered on average 0.10 more than the average of all subjects. Subjects who played A gathered, independent of their specified role, on average 0.10 less than the average of all players. Since there are only two possibilities to choose from in this Chicken Game, this means that the data suggest that playing B gains a subject 0.20 more payoff than playing A. However, it should be noted that these results are not significant on a 95 per cent level, so the null hypothesis that playing either A or B does not alter the payoff cannot be rejected at the 5%-level.

Signaller vs. Receiver

As can be seen in Table 4, the regression analysis suggests that signallers receive a 0.20 lower payoff than neutral players do. It also suggests that receivers receive a 0.29 lower payoff than neutral players do. These results would suggest that the mere use of cheap talk harms both signallers and receivers in the Chicken Game, compared to subjects who play the game without any form of cheap talk. It should be noted that only the negative effect on the payoff of receivers is significant on a 95 per cent level. This means that the null hypothesis that being a receiver of cheap talk in the Chicken Game does not alter the payoff of the subject, compared to neutral players, can be rejected at the 5%-level.

However, the effect of being a signaller in the Chicken Game is not significant on a 95 per cent level. Hence the null hypothesis that being a signaller of cheap talk in this game does not alter the subject's payoff, compared to neutral players, cannot be rejected at the 5%-level.

Table 2 and Table 4 do provide an answer to the main research question of this paper:

Does cheap talk benefit the signaller in the Chicken game?

Table 2 shows that the mean payoff of neutral players exceeds the payoff of signallers, which in its turn exceeds the payoff of receivers. The results in Table 4 show that whether a subject sends or receives a message, it both gathers him or her a lower payoff than subjects who are not confronted with cheap talk in any way during the game. It should be noted that the only effect from these regressions that is statistically significant on a 95 per cent level is the negative effect that the role of receiver has for a subject compared to subjects who are not receivers of cheap talk during the Chicken Game.

My hypothesis for this main research question was:

Hypothesis I: There will be no statistically significant beneficial effect for signallers during the Chicken Game.

This hypothesis is confirmed by the data. The only statistically significant effect found is the negative effect the role of receiver has on the subject's payoff compared to subjects, who do not have the role of receiver in the Chicken Game. The data suggest that being a signaller harms the payoff of the subject, compared to the neutral players. However, this effect is not statistically significant on a 95 per cent level, au contraire to the negative effect of being a receiver.

In summary, there is no statistically significant beneficial effect on being a signaller during the Chicken Game; the data even suggest a statistically insignificant negative effect on being a signaller during the Chicken Game compared to neutral players. However, the data do suggest a statistically significant negative effect on payoff if the subject is a receiver.

Signaller staying true or deviate from signal

As can be seen in Table 5, the regression analysis suggests that playing A benefits the signaller both when he or she signalled A (+0.12 compared to playing B) as well as when he or she signalled B (+0.04 compared to playing B). These results suggest that playing A gains the signaller a higher payoff than playing B, both when A and when B are signalled. The positive effect of playing A is however stronger when it's true to the signal sent. However, none of these results are significant on a 95 per cent level. Hence, the null hypothesis that any combination of staying true (or deviating from) the signal sent changes the payoff of the sender cannot be rejected at the 5%-level.

Reaction receiver on given signal

As can be seen in Table 6, the regression analysis suggests that playing the same option as has been sent to the receiver generates a higher payoff for the receiver. If a receiver receives the signal A, playing A generates a higher payoff of 0.56 than playing B. If a receiver receives the signal B, playing B generates a higher payoff of 0.29 than playing A. It should be noted that these results are not significant on a 95 per cent level. Hence, the null hypothesis that copying or deviating the option played from the signal received changes the payoff cannot be rejected at the 5%-level.

Neither in Table 5, nor in Table 6 any statistically significant effect is found that would suggest that a certain reaction on a certain signal would have a statistically significant effect on the payoff of the subject in case. This strikes with the percentage of receivers that play A after receiving either the signal A or the signal B. From the 59 receivers that received signal A, 32 of them actually chose A, which comes down to 54.24 per cent. From the 39 receivers that received signal B, 21 of them actually chose A, which comes down to 53.85 per cent.

This answers the second research question of this paper:

Does it matter what claims the signaller makes in his or her cheap talk in the Chicken Game?

The answer to this question is: No, it does not matter what claims the signaller makes in his or her cheap talk in the Chicken Game. This confirms the second hypothesis of this paper:

Hypothesis II: The claim a signaller makes in his or her cheap talk does not have a statistically significant effect on the payoff he or she receives.

Not a single statistically significant effect was found from the possible claims signallers could make in their cheap talk. Even when looking at the sheer percentages of responses from receivers there is virtually no difference between the choices receivers make when they receive signal A or signal B.

6.0 Conclusion

Knowledge of how individuals make decisions in strategic situations with conflicting interests remains of vital importance considering numerous economical, political or social situations. Based on the results of the experiment run for this paper, there is however no statistical evidence that one-way, one-shot cheap talk benefits the sender, nor the receiver in any way. This answers the main research question of this paper: Does cheap talk benefit the sender in the Chicken game? The data do not suggest any benefit of sending cheap talk. It does suggest a negative effect of being a receiver of cheap talk in the Chicken Game. If and how this effect could have been made stronger will be further discussed in the Discussion section.

The following research question can also be answered: Does it matter what claims the sender makes in his/her cheap talk in the Chicken game? The data do not suggest any benefit (nor deficit) from what signal a subject either sends or receives on the payoff the particular subject gains in the played round. In fact, receivers responded with playing A with the same percentage when they received either signal A or signal B.

If and how the claims in the cheap talk could have benefited the sender will also be further reviewed in the Discussion section of this paper.

6.1 Discussion

First of all, it would obviously have been preferred if the experiment could have been run with bigger groups or a larger amount of sessions. Under the current circumstances, the present experimental sessions could be seen as a sort of pilot.

Also, it should be noted that, due to a shortage of monetary funds, only one subject per session was selected to be paid out in cash. This was done so subjects would still have an economic incentive to think about their choices. It was, prior to starting the experiment, common knowledge for the subjects, that only one subject per session would be selected to actually be paid his or her payoff. Numerous subjects told the experimenter, after the experimental sessions had been run, that if they would get paid out their payoff for certain, they might have answered differently. Some others stated that higher payoff in general might have altered their choices. When these subjects were asked how certain or higher payoff would have altered their choices, all of them answered they would have put more thought into their decisions, or at least have a different thought process. However, if and how this would have altered their answers, remains unknown. This could be studied in a more qualitative research, which would probably be of a more psychological nature. Alterations in the thought

process subjects have during games with- and without cheap talk might make for some interesting research, but goes far beyond the scope of this paper.

Furthermore, it is possible that subjects approach the seven independent rounds as one extensive game, and hence adapt their strategy to this approach, despite the fact that they are fully aware they are playing a different opponent in each round. Subjects might, even if this is subconsciously, take their previous (or upcoming) decisions in other games into account whilst making up their mind about the game at hand. This is, amongst other reasons, why dummy variables were created to control for the seven decisions each and every subject made, which were all accounted for as individual observations.

Additionally, the effect of the cheap talk might have been different if the

communication methods of the subject were more abundant. Allowing for more flexible forms of communication, or even free-form communication could have a different effect on the choice of the receiver; keeping in mind that in this experiment there was only one-shot, one-way cheap talk. Loosening the latter restriction might gather different (statistically significant) results, however the research would quickly move to a more qualitative field and might get too rich to analyse. However, it has been a conscious choice to have strict

regulations considering communication within the experimental setup used for this paper. This is done in order to benefit the amount of control the experimenter has over the communication between subjects, so the focus could be solely on the sort of cheap talk designed for this paper.

In summary, after running a number of experimental sessions, there seems to be no evidence of any effect of cheap talk on the outcome of the Chicken game. At least, not in the one-way, one-shot method that was set up during the experimental sessions used for this paper. The effect of more elaborate cheap talk, as well as the cognitive action that is triggered within subjects after the introduction of cheap talk in games with conflicting interests still remains fascinating and might make up for more interesting research, albeit that some of this possible future research would be outside the field of Game Theory, and more towards the field of Psychology. The latter was however not the goal of this paper. This paper searched for any effect of one-shot, one-way cheap talk on either the sender or receiver of said cheap talk in the Chicken game. The gathered data nonetheless do not suggest any statistical significant effect.

7.0 References

Bernheim, B. D., Peleg, B., & Whinston, M. D. (1987). Coalition-proof Nash equilibria i. concepts. Journal of Economic Theory, 42(1), 1-12.

Blume, A., & Ortmann, A. (2007). The effects of costless pre-play communication: Experimental evidence from games with Pareto-ranked equilibria. Journal of Economic theory, 132(1), 274-290.

Charness, G. (2000). Self-serving cheap talk: A test of Aumann's conjecture. Games and Economic Behaviour, 33(2), 177-194.

Cooper, R., DeJong, D. V., Forsythe, R., & Ross, T. W. (1989). Communication in the battle of the sexes game: some experimental results. The RAND Journal of Economics, 568-587.

Cooper, R., DeJong, D. V., Forsythe, R., & Ross, T. W. (1992). Communication in coordination games. The Quarterly Journal of Economics, 107(2), 739-771.

Crawford, V. P., & Sobel, J. (1982). Strategic information

transmission. Econometrica: Journal of the Econometric Society, 1431-1451.

Croson, R., Boles, T., & Murnighan, J. K. (2003). Cheap talk in bargaining experiments: lying and threats in ultimatum games. Journal of Economic Behavior & Organization, 51(2), 143-159.

Duffy, J., & Feltovich, N. (2002). Do actions speak louder than words? An experimental comparison of observation and cheap talk. Games and Economic Behavior, 39(1), 1-27.

Farrell, J. (1987). Cheap talk, coordination, and entry. The RAND Journal of Economics, 34-39.

Farrell, J., & Rabin, M. (1996). Cheap talk. The Journal of Economic Perspectives, 10(3), 103-118.

Frank, R. H., Gilovich, T., & Regan, D. T. (1993). The evolution of one-shot cooperation: An experiment. Ethology and sociobiology, 14(4), 247-256.

Voorneveld, M. (2003). Characterization of Pareto dominance. Operations Research Letters, 31(1), 7-11.

8.0 Appendices

In the first two appendices attached to this paper the instruction forms can be found that were used during the experimental sessions of this paper. In Appendix I you can find the instruction form the neutral players from the control group received; in Appendix II you can find the instruction form that both the senders and receivers of cheap talk received.

8.1 Appendix I: Experimental form control group

Dear participant,

Welcome to my experiment! Feel free to ask the experimenter any question you have at any moment during the experiment. Please refrain from communicating in any way at all times with other participants. Please do not use your phone during the experiment. You will be participating in an experiment in the economics of

decision-making. You will be playing 7 rounds of the same, one-shot game. You will be randomly matched with a different player in all 7 rounds. So you will never play the same opponent twice. You, nor your opponent, will know the identity of the participant you are playing in any round. This will stay anonymous even after the end of the experiment. The sum of your combined payoffs in all 7 rounds will be the amount you will be paid in euros if you get selected for payment after this session.

NB: Only one person will be randomly selected for payment.

The payoffs of the game can be found in Table 1. As you can see:

 If you both play 'A', you both get a payoff of 2 euro for the concerning round.

 If you play 'B' and your opponent plays 'A', you will get a payoff of 3 euro; in this case your opponent earns 1 euro.

 If you both play 'B', neither of you will earn anything in said round.

 If you play 'A' and your opponent plays 'B', you will earn 1 euro and your opponent will earn 3 euro.

Please wait for the sign of the experimenter before you fill in your answer. Please answer either 'A' or 'B' in each round. Remember: Each round you will be playing against a new opponent.

If you have any questions left, feel free to ask them now!

Table 1: Payoffs

Game A B A 2,2 1,3 B 3,1 0,0

8.2 Appendix II: Experimental form test group

Dear participant,

Welcome to my experiment! Feel free to ask the experimenter any question you have at any moment during the experiment. Please refrain from communicating in any way at all times with other participants. Please do not use your phone during the experiment.

You will be participating in an experiment in the economics of decision-making. You will be playing 7 rounds of the same game. You will be randomly matched with a different player in each of the 7 rounds. So you will never play the same opponent twice. You, nor your

opponent, will know the identity of the participant you are playing in any round. This will stay anonymous even after the end of the experiment. The sum of your combined payoffs in all 7 rounds will be the amount you will be paid in euros if you are selected for payment after this session.

NB: Only one person will be randomly selected for payment.

The payoffs of the game can be found in Table 1. As you can see:

 If you both play 'A', you both get a payoff of 2 euro for the round concerned.  If you play 'B' and your opponent plays 'A', you will get a

payoff of 3 euro; in this case your opponent earns 1 euro.

 If you both play 'B', neither of you will earn anything in said round.

 If you play 'A' and your opponent plays 'B', you will earn 1 euro and your opponent will earn 3 euro.

It is possible that you are asked to send a message to your opponent before you both decide what you want to play. The message indicates your intended choice.

 If you are asked to send a message, select your message choice on your form and wait until the experimenter has taken note of your message.

 If you are not asked to send a message, please wait until the experimenter communicates to you what message your opponent is sending you. It is important that you communicate your message clearly to the experimenter.

NB: Though the message that you send/receive indicates the choice the signaller intends to make, it does not have to be true.

Please wait for the sign of the experimenter before you fill in your message. Then, please wait for the sign of the experimenter before you fill in your actual answer. Please answer either 'A'

Table 1: Payoffs

Game A B

A 2,2 1,3

or 'B' in each round. Remember: Each round you will be playing against a new opponent. You will not get to know what your opponent chose after each round.