Strategic Guilt Induction�

Eric Cardellay

April 23, 2012

Abstract

Guilt averse agents are motivated to meet the expectations of others, evenat the expense of their own material payo¤. Several experimental studieshave found results consistent with agents exhibiting guilt averse motivationsin games. However, there are strategic implications of guilt aversion, which canimpact economic outcomes in important ways, that have yet to be explored. Iintroduce a game that admits the possibility for agents to induce guilt uponothers in a manner consistent with the method posited by Baumeister, Stillwell,and Heatherton (1994). This game enables me to experimentally test whetheragents attempt to strategically inducing guilt upon others, and whether guiltinduction is an e¤ective mechanism for in�uencing the behavior of others. Ad-ditionally, the design enables me to test whether agents exhibit higher degreesof trust when they are given an opportunity to induce guilt. Furthermore,I appeal to the Battigalli and Dufwenberg (2007) model of simple guilt andshow that e¤ective guilt induction can be supported as an equilibrium of thepsychological game considered.

Keywords: guilt aversion, trust, psychological game theory, experiment

JEL Codes: C72, C91, D03, D80

�I would like to thank my dissertation advisor Martin Dufwenberg for helpful comments. Iwould also like to thank Pierpaolo Battigalli, Anna Breman, Gary Charness, Price Fishback, PaulHeidhues, Tamar Kugler, John Wooders, participants at the ESA conference in Tucson, and seminarparticipants at the University of Arizona, Rochester Institute of Technology, and Bowdoin Collegefor additional helpful comments. This project was funded by the Economic Science Laboratory andthe National Science Foundation (grant # SES-0819780).

yDepartment of Economics, University of Arizona, McClelland Hall 401, PO Box 210108, Tucson,AZ 85721-0108, Telephone: (858) 395-6699; Email: ecardell@email.arizona.edu.

1

1 Introduction

The results from a growing body of experimental literature suggests that economicagents making strategic decisions may not be solely motivated to maximize their ownmaterial payo¤s. One example of such �non-sel�sh�behavioral motivations is guiltaversion.1 The general idea of guilt aversion is that agents su¤er disutility, in the formof guilt, from letting others down. Thus, guilt averse agents may be motivated tocomply with the expectations of other agents, even at the expense of their own ma-terial payo¤. Behavior consistent with guilt aversion has been documented in severalexperimental studies.2 In light of the results from these experimental studies, it isimportant to consider the richer set of interpersonal strategic implications that canarise if agents are motivated by guilt aversion.

In particular, the guilt aversion of one agent can in�uence the behavior of otheragents in important ways. In certain strategic settings, the possibility may arise foragents to behave opportunistically and exploit the guilt aversion of others. Charnessand Dufwenberg (2006), in their concluding remarks, point to such a possibility byraising the question, �do people manipulate the guilt aversion of others in self-servingways� (p. 1595)?3 Given the opportunity, and incentive, agents could attempt toin�uence the behavior of guilt averse others by strategically inducing guilt upon thoseagents. Consequently, guilt averse agents may be more motivated to respond in kindto meet the expectations of those who induce guilt upon them. These interpersonalimplications of guilt aversion can impact strategic decision making and, consequently,economic outcomes in important ways that have yet to be explored.

The goal of this study is to explore these interpersonal implications of guilt aver-sion in a strategic economic setting. Speci�cally, this paper investigates the followingthree questions: First, do economic agents attempt to exploit the guilt aversion ofother agents in self-serving ways by strategically inducing guilt upon those otheragents? Second, is strategic guilt induction an e¤ective mechanism for in�uencing thebehavior of other agents? Third, is the strategic behavior of agents a¤ected by havingthe opportunity to induce guilt upon another agent?

Although previously unexplored in the economics literature, the interpersonal im-plications of guilt aversion have been recognized and documented in the psychologyliterature (Vangelisti et al., 1991; Baumeister et al., 1994, 1995 (BSH henceforth);

1Formal game theoretic models of guilt aversion have been developed by Dufwenberg (2002) fora speci�c game, and later by Battigalli and Dufwenberg (2007) for a general class of games.

2Examples include Dufwenberg and Gneezy (2000), Charness and Dufwenberg (2006, 2010),Bacharach et al. (2007), and Reuben et al. (2009) who consider variations of two-player experimental�trust� games (Berg et al. 1995) and �nd experimental evidence consistent with guilt aversion.Similarly, Dufwenberg et al. (2011) �nd evidence consistent with guilt aversion in an experimentalpublic goods game. On the contrary, Ellingsen et al. (2010) also consider various 2-player trustgames, but their study reveals little experimental support for guilt aversion.

3Rabin (1993) raises a similar question, albeit in a more general context and not speci�cally inreference to guilt, about whether players may be able to �force�emotions in a sequential move game.

2

Tangney and Fischer, 1995). In particular, BSH (1994) argue that one of the primaryfunctions of guilt is to motivate others to behave more pleasingly. In their study, BSH(1994) note that �we observed ample evidence of the hypothesized function of guiltas an interpersonal in�uence technique: People induced guilt to get another personto comply with their wishes�(p. 249). Similarly, Vangelisti et al. argue that peopleinduce guilt �primarily to achieve their own end-to persuade their listeners to do ornot to do something� (p. 33). These psychology studies provide valuable insightsregarding the functions of guilt in social relationships by drawing insights from non-incentivized personal narratives and surveys. However, the interpersonal functions ofguilt may not be restricted to social interactions; guilt may also function as an �in-terpersonal in�uence technique� in strategic economic interactions. An incentivizedexperimental game provides a suitable platform for investigating these interpersonalimplications of guilt aversion in economic settings.

The following example illustrates an economic setting, perhaps familiar to some,where strategic guilt induction could be employed as a means of in�uencing the be-havior of another in a self-serving way.

Example 1 You are a homeowner and you hire a private contractor to completean addition on your home by a pre-speci�ed date. Shortly thereafter, the contractorhas an opportunity to take-on another well-paying job. In order for the contractor tocomplete the new job on time, he must delay the completion of your addition by severalweeks. The contractor, however, is unaware that you had planned for your in-laws tocome stay in the new addition shortly after the original agreed upon completion date.Hence, the contractor is unaware of how much you would su¤er if the addition is notcompleted on time and your in-laws are not able to visit and stay in the new addition.Anticipating the guilt aversion of the contractor, you decide to inform the contractorthat your in-laws are coming into town and how much you would su¤er as a result ofthe addition not being completed by the original agreed upon date. To avoid the guiltthe contractor would feel by delaying the projects completion, the contractor decidesto work nights and weekends to complete the addition on time. As a result of yourstrategic guilt induction, the addition is completed on time and you get to spend a fewwonderful weeks with your in-laws!

This example highlights a speci�c setting where strategic guilt induction couldbe used to in�uence the behavior of a guilt averse other. However, there are othermore general economic settings where strategic guilt induction could be used to in-�uence the behavior of others and, consequently, impact outcomes. In the workplace,managers could induce guilt upon employees to mitigate shirking by conveying toemployees how their sub-standard e¤ort adversely a¤ects other employees.4 In con-tracting environments, guilt induction may allow a disadvantaged party to in�uence

4Sub-standard e¤ort by employees is likely to result in lower pro�ts for a �rm. Assuming thatbonuses are increasing in �rm pro�ts, then lower pro�ts would lead to lower bonuses for all employees.Thus, shirking by one employee could adversely a¤ect the well-being of other emplyees.

3

the behavior of an advantaged counter-party. As an example, a contracted �rm thathad made speci�c investments could possibly thwart opportunist re-contracting andhold-up by conveying to the counter-party �rm the loss in pro�ts associated with sucha hold-up. In academia, guilt induction by an assistant professor could lead to moretimely review decision on a submitted paper. Speci�cally, the assistant professor couldpossibly motivate a guilt averse journal editor into making a more prompt review de-cision by gently informing the editor, at the time of submission, that his/her tenurereview is rapidly approaching and a lengthy review period could hinder his/her tenureprospects.5 Many economic settings, like these mentioned, permit the possibility toinduce guilt upon others. Hence, a deeper understanding of the strategic interpersonalimplications of guilt aversion is required to accurately predict how the guilt aversionof agents impacts outcomes in such settings; the insights gleaned from this paper areintended to help with this understanding.

In order to investigate whether agents strategically induce guilt upon others, andthe e¤ectiveness of such strategic guilt induction, it is crucial to �rst identify howagents attempt to induce guilt upon others. For this, I draw valuable insights fromBSH (1994) who posit the following method for how people induce guilt in others:�If Person A wants Person B to do something, A may induce guilt in B by conveyinghow A su¤ers over B�s failure to act in the desired fashion�(p. 247). This methodfor inducing guilt implicitly requires that (i) Person A has private information abouthis own well-being, and (ii) has the possibility to convey such private informationto Person B. Therefore, the chosen strategic setting must feature both private pay-o¤ information and the ability to convey the private payo¤ information. However,the previous studies that have investigated guilt aversion in strategic settings mostlyconsider variations of 2-player �trust� games that do not feature private payo¤ in-formation or the possibility to convey the private payo¤ information.6 Hence, a newgame is warranted that provides a rich enough strategic structure to allow agents theopportunity to induce guilt upon others.

I employ a novel experimental design that uses a 2-player, binary choice trust gamefeaturing both private payo¤ information and an opportunity to convey this privateinformation. In the game, the privately informed �rst-mover (Player A) is given anopportunity to convey to the second-mover (Player B) how low his/her payo¤ wouldbe as a result of Player B choosing an action that is undesirable for Player A. Thise¤ectively provides Player A an opportunity to convey to Player B how much Player Awould �su¤er�over Player B�s failure to act in the desired fashion, which is consistentwith the BSH (1994) method of guilt induction.7 This game allows me to then derive

5This example was inspired by an editor who revealed to me that some assistant professors doactually inform the editor, at the time of submission, about such an upcoming tenure review!

6A study by Fong et al. (2007) uses a trust game with private information. However, theauthors incorporate private information as a means of testing their model of guilt driven reciprocity.Additionally, their trust game features private information for the second mover, while I will considera trust game with private information for the �rst mover, as will be shown in the next section.

7I use the term �guilt induction�when referring to Player A�s strategic attempt to in�uence the

4

testable hypotheses regarding whether Player As attempt to strategically induce guiltupon Player Bs, and whether Player Bs are susceptible to strategic guilt induction.The experimental design also includes a second, related, trust game that does notfeature an opportunity for Player A to induce guilt upon Player B. This second gameprovides a baseline trust measure for Player As, which then allows me to derive atestable hypothesis regarding whether Player As are more trusting of Player Bs whenthey have an opportunity to induce guilt upon them.

Furthermore, I theoretically explore the connection between the psychological in-sights of BSH (1994) regarding the interpersonal implications of guilt and the model of�simple�guilt developed by Battigalli and Dufwenberg (2007) (B&D henceforth).8 Indoing so, I apply the B&D model of guilt to the proposed experimental trust game. Iderive conditions on beliefs under which the method for guilt induction, as posited byBSH (1994), is consistent with B&D. That is, I derive conditions on beliefs for whichthe B&D model predicts that Person A can induce guilt in Person B by conveyinghow low his payo¤ would be as a result of Person B choosing an undesirable action.Subsequently, I show that e¤ective guilt induction can be supported as a sequentialequilibrium of the proposed game under the guilt framework of B&D.

2 Hypothesis Development

I begin this section by �rst introducing the two trust games from which the researchhypotheses are developed. I refer to both games as trust games because they feature apayo¤ structure indicative of the broader class of trust games. Namely, a game wherethe �rst mover has an opportunity to choose an action that creates the possibility ofmutual bene�t, if the other person cooperates, but a risk of lower payo¤s to oneselfif the other person defects. Such an action taken by the �rst mover is referred to asa trusting action (Cox, 2004). Trust games, in general, allow for the possibility ofguilty feelings, which make them a suitable platform for developing and testing thehypotheses of this study relating to the strategic implications of guilt aversion.

behavior of Player B by conveying how low his payo¤ would be given a future undesirable action ofPlayer B. I do this to remain consistent with the psychological intuition and terminology outlined byBSH (1994). However, in relation to the Battigalli and Dufwenberg (2007) model of guilt, it is morepedagogical to think of this strategic behavior from Player A as �counterfactual� guilt induction.Essentially, Player A is trying to increase the amount of guilt that Player B would feel as a result ofchoosing an action that is undesirable for Player A. This makes the guilt counterfactual in the sensethat Player B may never experience the guilt if he chooses an action that complies with Player A�sdesires.

8B&D also model a second form of guilt, �guilt from blame�. However, in this paper I will onlyconsider simple guilt and, therefore, for the remainder of the paper when I refer to the guilt modelof B&D, I am implicitly refering to the model of simple guilt. The B&D model is an applicationof the authors�more general theoretical framework developed in Battigalli and Dufwenberg (2009),which extends the psychological game theory framework pioneered by Geanakoplos et al. (1989).

5

2.1 Uncertain Payo¤Trust Game ��UPT

�UPT is a 2-player, sequential move game. �UPT begins with the �rst mover, PlayerA, choosing between In or Out. If Player A chooses Out, then the game ends; PlayerA receives a payo¤ of 6, and Player B receives a payo¤ of 2. If Player A chooses In,then Player B is called upon to move. Player B must choose between Left or Right.If Player B chooses Right, then the game ends; Player A receives a payo¤ of 10, andPlayer B receives a payo¤ of 4. If Player B chooses Left, then the game ends; PlayerA receives a payo¤ of X; and Player B receives a payo¤ of 6. X is a random variablewhere prob(X = 0) = 1=2 and the prob(X = 6) = 1=2: At the start of the game, thedistribution of X is known to both players. The extensive form of �UPT is depictedbelow in Figure 1.

Figure 1: Extensive Form of �UPT

2.2 Private Payo¤Trust Game ��PPT

�PPT features a similar strategic structure and payo¤ structure to those of �UPT ;with two important di¤erences. First, �PPT features an opportunity for Player A tobecome privately informed about the value of X. Second, �PPT features an additionalstage where Player A has the opportunity to credibly convey his private informationabout the value of X to Player B.

6

�PPT begins analogously to �UPT with Player A �rst choosing between In or Out.If Player A chooses Out, the game ends; Player A receives a payo¤ of 6, and Player Breceiving a payo¤ of 2. If Player A choose In, Nature then decides whether Player Abecomes privately informed about the value of X: With prob = 4=5; Nature Reveals(Rev) the value of X to Player A, and with prob = 1=5; Nature does Not Reveal (NotRev) the value of X to Player A.

If the value of X is revealed to Player A, then an additional stage arises wherePlayer A must decide whether to credibly Convey (C) on Not Convey (NC) the valueof X to Player B before Player B gets the move. Upon getting the move, Player Bmust then decide between Left or Right. Analogous to �UPT , if Player B choosesRight, then the game ends; Player A receives a payo¤ of 10, and Player B receivesa payo¤ of 4. If Player B chooses Left, the game ends; Player A receives a payo¤of X; and Player B receives a payo¤ of 6, where, again, prob(X = 0) = 1=2 andthe prob(X = 6) = 1=2. The extensive form of �PPT is depicted in Figure 2. Tosimplify the extensive form, the two moves by Nature, determining the value of Xand determining whether the value of X is revealed to Player A, have been combinedinto one move.

Figure 2: Extensive Form of �PPT

7

If players are �sel�sh�, i.e., act to maximize their own material payo¤, then theunique equilibrium outcome in both �UPT and �PPT is Player A chooses Out. To seethis, note that it is dominant for Player B to choose Left in both �UPT and �PPT .As a result, it is sequentially rational for a sel�sh Player A to choose Out in both�UPT and �PPT ; and the unique equilibrium outcome is Player A chooses Out.9 Theinclusion of private information and the additional conveyance stage for Player A in�PPT has no impact on the equilibrium outcome assuming sel�sh players. However, itis exactly these two features of �PPT that will allow me to derive testable hypothesesregarding whether agents attempt to induce guilt, and its subsequent e¤ectiveness.

Before I proceed, I �rst highlight two important features of �PPT that will berelevant for the upcoming derivation of the research hypotheses and the applicationof the B&D model of guilt: First, Player A must choose between In or Out beforepossibly becoming informed about the value of X. This eliminates any possible sig-naling value regarding the value of X inferred from Player A�s decision of In or Out.Additionally, the value of X is only revealed to a Player A who chooses In withprob = 4=5. Therefore, if the value of X is not conveyed to Player B, then Player Bis unable to perfectly distinguish between whether the value of X was not revealedto Player A, or the value of X was revealed and Player A chose to Not Convey. As aresult, if the value of X is not conveyed to Player B, then Player B�s expectation ofthe value of X will be strictly greater than zero and strictly less than six.

More precisely, bmA 2 [1; 5] where bmA denotes Player B�s expectation of X condi-tional on the value of X not being conveyed.10 To see this, note that in calculatingbmA; Player B must think about the relative probability that (i) Player A learnedX = 0 and chose Not Convey, (ii) Player A learned X = 6 and chose Not Convey,and (iii) Player A did not learn the value of X. Although these probabilities areunobservable to the researcher and, thus, the exact value of bmA is unobservable, it ispossible to identify bounds on bmA: Speci�cally, the largest expectation that Player Bcould hold regarding the value of X is if he thinks that only a Player A who learnedthat X = 6 would choose to Not Convey. In this case, bmA is bounded above by13�E[X] + 2

3� 6 = 1

3� 3 + 2

3� 6 = 5: Here, 1

3and 2

3represent the updated probabilities,

via Bayes�rule, that Player A did not learn the value of X and Player A learned thatthe value of X = 6, respectively. By a similar argument, the smallest expectationthat Player B could hold regarding the value of X is if he thinks only a Player A wholearned X = 0 would choose to Not Convey. In this case, bmA is bounded below by13� E[X] + 2

3� 0 = 1

3� 3 + 2

3� 0 = 1: Therefore, bmA 2 [1; 5]:

9In �PPT ; there is multiplicity of Perfect Baysian Equilibria for sel�sh players that depend on thespeci�cation of Player B�s beliefs at the information set where no information is conveyed regardingthe value of X. However, regardless of Player B�s beliefs, it is rational for him to choose Left andsubsequently, it is sequentially rational for Player A to choose Out. Therefore, the unique equilibriumoutcome of �PPT is the game ending with Player A choosing Out.10The notational choice of bmA anticipates the upcoming application of the B&D model of guilt to

�PPT in Section 5.

8

2.3 Research Hypotheses

The �rst motivation of this study is to investigate whether agents attempt to induceguilt upon others. Recall that BSH (1994) posit that a person can induce guilt uponanother by conveying to that person how one su¤ers over that person�s failure to actin the desired fashion. Let us consider how this method of guilt induction appliesto �PPT : Conditional on choosing In, a Player A would �desire�Player B to chooseRight as it yields him a payo¤ of 10 compared to a payo¤ of X < 10 if Player Bwere to choose Left. The extent to which Player A would �su¤er� from Player B�sfailure to choose Right depends on the value of X; the lower X the lower Player A�spayo¤ and the more Player A would su¤er. Given the opportunity, the BSH (1994)method for inducing guilt would prescribe that Player A induce guilt upon PlayerB by �conveying�to Player B a low value of X, i.e., by conveying to Player B howmuch he would su¤er if Player B were to choose Left.

Conditional on having chosen In in �PPT and having the value of X is revealed,Player A must decide whether to convey the value of X to Player B. In the case whereX = 0 is revealed, if Player A chooses to Convey X = 0; then Player B will knowthat if he chooses Left, Player A will receive a payo¤ of X = 0. Whereas, if Player Achooses to Not Convey X = 0, then Player B will think that if he chooses Left, PlayerA will receive a payo¤ of bmA 2 [1; 5]: Analogously, consider the case where the valueof X = 6 is revealed. If Player A chooses to Convey X = 6, then Player B will knowthat if he choose Left, Player A will receive a payo¤ of X = 6. Whereas, if Player Achooses to Not Convey X = 6; Player B will think that if he chooses Left, Player Awill receive a payo¤ of bmA 2 [1; 5]. Hence, Player B will know that Player A su¤ersmore from his choice of Left when X = 0 is conveyed, compared to when the value ofX is not conveyed. Similarly, Player B will know that Player A su¤ers more from hischoice of Left if the value of X is not conveyed, compared to if X = 6 is conveyed.Therefore, an opportunistic Player A who is attempting to strategically induce guiltupon Player B, as prescribed by the BSH (1994) method, would Convey X = 0; andNot Convey X = 6: This leads to the �rst testable hypothesis:

H1: The proportion of Player As who Convey X = 0 in �PPT is larger that theproportion of Player As who Convey X = 6:11

Later, I will apply the B&D model of guilt to �PPT and show that under certainconditions on the second-order beliefs of Player B, the B&D model posits that PlayerB will su¤er more guilt from choosing Left if X = 0 is conveyed, compared to if thevalue of X is not conveyed; similarly, Player B will su¤er more guilt from choosing

11BSH (1994) note that guilt results from hurting others. Thus a person may su¤er guilt frominducing guilt. The authors refer to this type of guilt from inducing guilt as �metaguilt�. It ispossible that Player A could therefore su¤er metaguilt from conveying X = 0 and not conveyingX = 6. These metaguilty feelings could then motivate Player A to Not Convey X = 0 and toConvey X = 6: However, this potential metaguilt e¤ect is in the opposite direction of H1, whichwould not give rise to a possible confounding e¤ect.

9

Left if the value of X is not conveyed, compared to if X = 6 is conveyed. Hence, thehypothesis that Player A can attempt to induce guilt upon Player B by choosing toConvey X = 0 and Not Convey X = 6 in �PPT is consistent with the B&D model ofguilt. I will further show that inducing guilt by choosing to Convey X = 0 and NotConvey X = 6 can be supported in an equilibrium of �PPT if Player B is su¢ cientlysensitive to feeling guilty.

The second motivation of this study is to investigate whether agents are susceptibleto guilt induction. That is, are agents more motivated to respond kindly after guilthas been induced upon them? BSH (1994) posit that after Person A has induced guiltin Person B, �Person B �nds the guilt aversive and, to escape from guilt, complieswith A�s wishes� (pp. 247). It is also possible, however, that a guilt averse PlayerB will recognize that Player A is trying to manipulate his behavior by �guilting�him, which can result in Player B being more motivated to choose the unkind actionof Left in response to Player A�s attempted guilt induction. BSH (1994) (1995)document this potential �cost� of guilt induction by arguing that target of guiltinduction (Player B) might feel resentment and be motivated to respond negativelytoward the guilt inducer (Player A). Hence, attempted guilt induction by Player Amay be counterproductive as a means of motivating Player B to choose Right as itmay foster more sel�sh behavior and motivate Player B to choose Left, contrary toPlayer A�s intended motivation.

In relation to �PPT , Player B complying with Player A�s wishes corresponds toPlayer B choosing Right. Therefore, if guilt induction by Player A is an e¤ectivein�uence mechanism, then Player B would by more motivated to choose Right afterPlayer A induces guilt by choosing to Convey X = 0 and Not Convey X = 6. Thisleads to the following testable hypothesis:

H2: The proportion of Player Bs choosing Right in �PPT after X = 0 is conveyedis larger than when the value of X is not conveyed, which is larger than whenX = 6 is conveyed.12

The third motivation of this study is to investigate whether having the opportunityto induce guilt fosters more trusting behavior. Comparing �UPT and �PPT ; we cansee that the di¤erences between �UPT and �PPT are the possibility for Player Ato become privately informed about the value of X; and the ability to convey thelearned value of X to Player B. As I have demonstrated, it is these di¤erences thatprovide the opportunity to Player A to induce guilt in Player B. Therefore, if havingan opportunity to induce guilt fosters more trusting behavior, then Player As would

12Note, if all Player As attempt to induce guilt, then the value of X = 6 would never actually beconveyed to Player B and, thus, there would exist not data on the proportion on Player Bs choosingRight after X = 6 was conveyed. In this case, H2 would just reduce down to the binary comaprisonof the proportion of Player Bs choosing Right after X = 0 was conveyed to when the value of Xwas not conveyed. However, as we will see in the results section, some Player As do convey X = 6so there will exist the necessary data to test H2 as it is stated.

10

be more motivated to choose In in �PPT ; compared to �UPT :13 This leads to thefollowing testable hypothesis:

H3: The proportion of Player As who choose In when playing �PPT is larger thanthe proportion of Player As who choose In when playing �UPT .

2.4 Possible Confounds

I conclude this section by addressing two plausible, and widely known, alternativemotivation that can possibly in�uence the decision making of second movers in trustgames: (i) inequality aversion, and (ii) reciprocity. In regards to inequality aversion,the idea is that agents are averse to unequal outcomes, both advantageous and disad-vantageous (Fehr and Schmidt, 1999; Bolton and Ockenfels, 2000). Applied to trustgames, inequality aversion can possibly motivate the second mover to choose the morekind action (Right), as it generally results in a more equal outcome. However, theexperimental design anticipates and controls for possible inequality aversion of PlayerBs in �PPT . Speci�cally, an inequality averse Player B playing �PPT always prefersto choose Left. The intuition is relatively straightforward. When Player B choosesLeft, his payo¤ increases and the outcome becomes less unequal, regardless of thevalue of X, therefore, it is optimal for an inequality averse Player B to always chooseLeft.14 It follows that Player A�s conveyance decision regarding the value of X in�PPT cannot be explained by any strategic considerations regarding the inequalityaversion of Player B.

In regards to reciprocity motivations, the general idea is that agents are motivatedto respond kindly to agent who are kind to them, and respond unkindly to agentswho are unkind to them (see Dufwenberg and Kirchsteiger, 2004 for a formal modelof reciprocity). In the context of �PPT ; Player B may be motivated to reciprocate thekind action of Player A choosing In by choosing Right (see Cox, 2004 for a generaldiscussion of reciprocity motivations in trust games). This motivation to positivelyreciprocate may be more prevalent if Player B knows that X = 0.

To control for this possible reciprocity confound, I consider a third game, which isa �dictator�version of �PPT : In this modi�ed dictator version of �PPT , which I referto as the private payo¤dictator game and denote as �PPD, the initial In/Out decision

13The idea that In constitutes a trusting action for Player A in �UPT and �PPT is consistent withthe behavioral de�nitions of trust presented in Cox (2004) and Fehr (2009).14Formally, if you apply the Fehr and Schmidt (1999) model, an inequality averse Player B playing

�PPT always prefers the payo¤ vector (X; 6) to the payo¤ vector (10; 4) 8X 2 [0; 6]: To see this notethat for the extreme case where X = 0; we have that 6 � �(6 � 0) > 4 � �(10 � 4) 8� 2 [0; 1) and� � �; where the LHS represents an inequality averse Player B�s utility from choosing Left and theRHS represnts the utiility from choosing Right. For the other extreme case where X = 6; we havethat 6 > 4 � �(10 � 4) 8� � 0; where the LHS represents an inequality averse Player B�s utilityfrom choosing Left and the RHS represnts the utiility from choosing Right. The constraints that� 2 [0; 1) and � � � in the above inequalities are assumed a priori in the FS model.

11

of Player A is eliminated. �PPD begins with Player A choosing whether to conveythe value of X to Player B, conditional on the value of X being revealed to Player A.Then, Player B decides whether to choose Right or Left. Hence, the extensive form of�PPD is simply the subgame of �PPT that begins with Nature�s move. Because thereis no initial In/Out decision in �PPD, Player Bs Right decision cannot be a result ofwanting to reciprocate Player A�s kind decision of choosing In. Thus, comparing thedecision making of Player B in �PPD with that in �PPT allows me test for and, ifnecessary, control for reciprocity motivations of Player B.15

The parameterization of �PPT ; as well as the use of �PPD; allows me to testin Player Bs are susceptible to guilt induction (H2) while controlling for possibleconfounds that can result from Player B preferences for payo¤ equality or reciprocity.Subsequently, this allows me to test in Player As attempt to strategically induce guilt(H1) while controlling for any possible confounds resulting from Player As strategicconsiderations of Player Bs preferences for payo¤ equality or reciprocity.

3 Experimental Design

I test the three main research hypotheses of this paper (H1-H3) using the followingthree experimental treatments:

UPT Treatment Subjects played �UPT one time as either the role of Player A orPlayer B, where the payo¤s from �UPT corresponded 1:1 with the monetarypayo¤s in the experiment.

PPT Treatment Subjects played �PPT one time as either the role of Player A orPlayer B, where the payo¤s from �PPT corresponded 1:1 with the monetarypayo¤s in the experiment.

PPD Treatment Subjects played �PPD one time as either the role of Player A orPlayer B, where the payo¤s from �PPD corresponded 1:1 with the monetarypayo¤s in the experiment

All experimental sessions were conducted in the Economic Science Laboratory(ESL) at the University of Arizona in April 2011 and October of 2011. The sessionswere computerized and the software was programmed using Z-tree.16 The subject poolconsisted of undergraduates who were recruited via an online database. In total, 22sessions were conducted using 444 subjects comprising 222 groups. Of the 222 groups,111 were assigned to the PPTTreatment, 45 were assigned to the UPTTreatment, and

15This approach of controlling for the possible reciprocity concerns of Player B in the trust game(�PPT ) by using the corresponding dictator game (�PPD) was inspired by the triadic design approachdeveloped and implemented in Cox (2004).16I thank Urs Fischbacher (Fischbacher, 2007) for providing this software.

12

66 were assigned the PPD Treatment. Each session lasted approximately 25 minutesand subjects earned an average of $9.68 USD (including a $5 show-up payment).

Subjects were randomly assigned to their player role, then randomly and anony-mously matched with a subject of the opposite player role, and then played theircorresponding game one time. A copy of the experimental instructions are presentedin the Appendix. Upon the completion of the game, the decisions of each player, thecorresponding outcome, the pro�t to each player, and the value of X were displayedto both players. All subjects were informed in the instructions that the value of Xwould be revealed to both players upon completion of the task, irrespective of thedecisions made in the task. Revealing the value of X to all players ensures that PlayerAs were not motivated to choose In (or Conveying X) just so Player A (Player B)could learn the value of X. This design feature eliminates any curiosity biases thatmay arise from the uncertainty regarding the value of X; and the consequent payo¤sto the other player.

After the subjects �nished the game, they were asked to �ll out a short question-naire. In all treatments, the questionnaire contained 8 general demographic questions.In the PPT and PPD treatments, two additional questions were asked to both PlayerA and Player B that related to guilt feelings in the task. The speci�c questionsthat were asked to each player, the corresponding responses, and a discussion of thepossible insights that can be gleaned are presented later in Section 4.2.

4 Results

I will �rst present the aggregate decision data from each of the three treatmentsand the corresponding tests of the three research hypotheses. I will then presentsome results from the post-decision questionnaire to shed more light on the guiltperceptions of subjects. I conclude this section with some discussion and speculativeremarks about some of the observed patterns in the data.

4.1 Aggregate Data and Hypothesis Testing

I begin by comparing the aggregate Player B data from the PPT and PPD treat-ments in order to test for any possible reciprocity motivations. Speci�cally, I comparethe frequency of Player Bs choosing Right at each of the three possible conveyancestates: (i) X = 0 was conveyed (RightjX = 0), (ii) the value of X was not conveyed(RightjX = NC), and (iii) X = 6 was conveyed (RightjX = 6). Figure 3 displaysthe histogram of the relevant Player B data for the two treatments.

From Figure 3, we can see that the relative frequency of Right choices at each ofthe three conveyance states is very similar across the two treatments. If fact, a 2-sidedFisher�s Exact test does not yield a signi�cant di¤erence between the proportion ofPlayer Bs who choose Right jX = 0, Right jX = NC, and Right jX = 6 between the

13

two treatments (p = 1.000, p = 1.000, and p = 0.542, respectively). Thus, we canrule out reciprocity as a possible confounding motivation for Player B�s Right/Leftdecision in �PPT . As a result, I pool the data from the PPT and PPD treatments forthe remainder of the data analysis.

Figure 3: Relative Frequency of Right Rates by Treatment

I proceed with the hypothesis testing, which I present in the order that correspondsto working backwards through the game. Speci�cally, I �rst test H2, whether PlayerBs susceptible to guilt induction by Player As? In terms of testing H2, I compare theproportion of Player Bs who choose Right jX = 0, Right jX = NC, and Right jX = 6:Table 1 presents the relevant data. From Table 1, we can see that 12/30 (40%)Player Bs chose Right jX = 0, 7/61 (11%) chose Right jX = NC, and 2/33 (6%)choose Right jX = 6. Using a Jonckheere-Terpstra non-parametric test for orderedalternatives, we can reject the null that these three proportions are equal in favor ofthe ordered alternative (p < 0.001).17 The data suggests that Player Bs are susceptibleto guilt induction by Player A, in the sense that they are more likely to chooseRight jX = 0; compared to Right jX = NC; compared to Right jX = 6, which supportsH2. An alternative interpretation is that guilt induction by Player A, in the formof conveying X = 0 and not conveying X = 6, would be e¤ective at increasing thelikelihood that Player B chooses Right.

17The result of this hypothesis test is robust if a Cuzick�s non-parametric test is alternatively used.

14

Table 1: Comparison of Player B Right Rates by Conveyance State

Player B Data (PPT and PPD Treatments)

Right jX = 0 Right jX = NC Right jX = 6 (p-value)

12/30 7/61 2/33 (< 0.001)(40%) (11%) (6%)

Notes: reported p-value is from a Jonckheere-Terpstra non-parametric test forordered alternatives.

I now turn to testing H1, namely, are Players As attempting to induce guilt uponPlayer Bs? Recall that if Player As are attempting to induce guilt, then a largerproportion of Player As will Convey X = 0 compared to Convey X = 6 in �PPTand �PPD: To test this hypothesis, I compare the conveyance decisions of Player Asfor whom X = 0 was revealed with those for whom X = 6 was revealed. Table 2presents the aggregate pooled Player A conveyance data from the PPT and PPDTreatments. From Table 2 we can see that 30/43 (70%) Player As chose to ConveyX = 0 and 33/55 (60%) Player As chose to Convey X = 6. Although the proportionof Player As who Convey X = 0 is larger than the proportion who Convey X = 6, thisdi¤erence is not signi�cantly di¤erent using a 1-sided Fisher�s exact test (p = 0.215).The conveyance data suggests that Player As are not attempting to strategicallyinduce guilt upon Player Bs, in the sense that there is no signi�cant di¤erence in theaggregate proportions of Player As who Convey X = 0 and Convey X = 6, whichfails to support H1.

Table 2: Comparison of Player A Conveyance Rates

Player A Data (PPT and PPD Treatments)

Convey X = 0 Convey X = 6 (p-value)

30/43 33/55 (0.215)(70%) (60%)

Notes: p-value is reported for a 1-sided Fisher�s Exact test ofconveyance rates

15

H3 stated that if having an opportunity to induce guilt fosters more trustingbehavior, then more Player As would choose In when playing �PPT compared to�UPT : To test this hypothesis, I compare the aggregate Player A In rate in the PPTTreatment with the UPT Treatment. From Table 3, we can see that 58/111 (52%) ofPlayer As chose In in the PPT Treatment and 17/45 (38%) of Player As chose In inthe UPT treatment, which is marginally signi�cant using a 1-sided Fisher�s Exact test(p = 0.071). The data suggests that strategic settings that provide an opportunityfor agents to induce guilt may foster more trusting behavior by those agents.

Table 3: Comparison of Player A In Rates

UPT Treatment PPT Treatment (p-value)

In Rate 17/45 58/111 (0.071)(38%) (52%)

Notes: p-value is reported from a 1-sided Fisher�s Exact test of conveyance rates

4.2 Questionnaire Results

Next, I present results from two of the questions, which were asked to both Player Asand Player Bs in the PPT and PPD Treatments, from the post-decision questionnaire.The motivation of these two questions was to gain additional insights regarding PlayerB�s feelings of guilt, and Player A�s perceptions of Player B�s feelings of guilt. Thisquestionnaire was not incentivized and did not impact monetary earnings and, as aresult, the natural amount of discretion must be used in evaluating the resulting data.At the same time, there is really no scope for any type of gains from strategic falsereporting and, thus, no obvious motivation to not report truthfully. In addition, Icompare the responses of the two questions using matched samples, which controlsfor scaling di¤erences that could exist with unmatched samples.

For Player As, the �rst question asked how much guilt they thought Player Bfelt (would have felt) from choosing Left if Player B knew (would have known) thetrue value of X: The second question asked Player As how much guilt they thoughtPlayer B felt (would have felt) from choosing Left if Player B did not know (wouldnot have known) the true value of X: Responses were ranked on a 5-point scale with5 being a Very High amount of guilt and 1 being a Very Low amount of guilt. Forthe analysis, I only consider the Player As who had the value of X revealed to themand, thus, had an opportunity to convey the value of X to Player B. Table 4 presentsthe aggregate response data, pooled from the PPT and PPD Treatments, for these

16

Player As.18 Table 4 is divided into two panels that correspond to whether X = 0 orX = 6 was revealed to Player A. Within each panel, the average reported perceptionsof Player B�s guilt feeling from choosing Left is separately presented for those PlayerAs who conveyed the corresponding value of X and those who did not.

Table 4: Player A�s Perceptions of Player B�s Guilt Feelings fromChoosing Left

Panel 1 �Player As for whom X = 0 was Revealed

Guilt if Guilt ifPlayer B Player B did

knew X = 0 not know X = 0 (p-value)

Player As who 2.46 1.78 (0.008)conveyed X = 0

Player As who did 2.36 1.91 (0.524)not convey X = 0

Panel 2 �Player As for whom X = 6 was Revealed

Guilt if Guilt ifPlayer B Player B did

knew X = 6 not know X = 6 (p-value)

Player As who 1.72 2.21 (0.061)conveyed X = 6

Player As who did 1.48 2.49 (0.002)Not Convey X = 6

Notes: Reported p-values are from a Wilcoxon Signed-Rank test for matched samples

18Because of the way the questionnaire was programmed in z-tree and administered, it was possiblefor subjects to fail to submit an answer for each question. A total of 5 of 98 Player As who had thevalue of X revealed to them failed to answer at least one of the questions related to their beliefs ofPlayer Bs guilt feelings. Therefore, the aggregate data in Table 4 re�ects the responses of 93 PlayerAs who did answer both questions.

17

From Panel 1 of Table 4, we can see that the Player As who chose to ConveyX = 0 perceived that Player B would have felt signi�cantly more guilt from choosingLeft jX = 0 compared to Left jX = NC (p = 0.008). However, the Player As whochose toNot Convey X = 0 did not perceive that Player B would have felt signi�cantlymore guilt from choosing Left jX = 0 compared to LeftjX = 0 (p = 0.524). Similarly,from Panel 2, we see that the Player As who chose to Not Convey X = 6 perceivedthat Player B would have felt signi�cantly more guilt from choosing Left jX = NCcompared to Left jX = 6 (p = 0.002). While Player As who chose to Convey X = 6perceived that Player B would have felt only marginally more guilt from choosingLeft jX = NC compared to Left jX = 6 (p = 0.061).

What we see from the Player A questionnaire response data is that the Player Aswhose conveyance decisions were consistent with attempted guilt induction, i.e., thosePlayer As who conveyed X = 0 and did not convey X = 6; perceived that doing sowould induce more guilt upon Player B from choosing Left. While the Player As whoseconveyance decisions where not consistent with attempted guilt induction, i.e., thosePlayer As who did not convey X = 0; did not perceive that doing so would inducemore guilt upon Player B from choosing Left. This suggests that there may be twotypes of Player As. The �rst type is those who think that guilt can be induced uponPlayer B by conveying X = 0 and not conveying X = 6; and therefore, attempted todo so. The second type is those who do not think guilt can be induced upon PlayerB in this manner, and therefore, did not attempt to do so.

I turn now to the Player B questionnaire data. For Player Bs, the �rst questionasked how much guilt he/she felt (would have felt) from choosing Left (if he hadchosen Left) if he/she did not know the value of X: The second question asked PlayerB how much guilt he/she felt (would have felt) from choosing Left (if he had chosenLeft) if he/she knew the value of X: Again, responses were ranked on a 5-point scalewith 5 being a Very High amount of guilt and 1 being a Very Low amount of guilt.For the analysis, I consider the Player Bs who actually made a Left/Right decision ineither �PPT or �PPD. Table 5 presents the aggregate response data, pooled from thePPT and PPD Treatments, for these Player Bs.19 Table 5 is divided into two panelsthat correspond to the actual value X, and each panel shows the average reportedPlayer B guilt feelings from choosing Left.

From Panel 1 of Table 5 we see that when X = 0; Player Bs reported thatthey would have felt signi�cantly more guilt from choosing Left jX = 0 comparedto Left jX = NC (p < 0.001). Similarly, from Panel 2 we see that when X = 6,Player Bs reported that they would have felt signi�cantly more guilt from choosingLeft jX = NC compared to Left jX = 6 (p < 0.001). These reported perceived guiltfeelings by Player Bs from choosing Left at each of the conveyance states reinforce

19A total of 7 of 124 of the Player Bs did not answer at least one of the questions related to theirguilt feelings. Therefore, the aggregate data in Table 5 re�ects the responses of 117 Player Bs whodid answer both questions.

18

the observed behavioral �ndings that Player Bs are susceptible to guilt induction byPlayer A. In the sense that, Player Bs reported that they would feel more guilt fromchoosing Left jX = 0; compared to Left jX = NC; compared to Left jX = 6; therefore,Player Bs are more likely to choose Right jX = 0, compared to Right jX = NC,compared to Right jX = 6; which is what we observe in the data.

Table 5: Player B�s Reported Guilt Feelings from Choosing Left

Panel 1 �Player Bs for whom X = 0

Guilt if Guilt ifX = 0 Known X = 0 Unknown (p-Value)

2.53 1.73 (< 0.001)

Panel 2 �Player Bs for whom X = 6

Guilt if Guilt ifX = 6 Known X = 6 Unknown (p-Value)

1.42 2.15 (< 0.001)

Notes: Reported p-values are from a Wilcoxon Signed-Rank testfor matched samples

4.3 Discussion

Although the motivation of this study is to test H1-H3, I take this time to make afew speculative remarks regarding some of the observed patterns in the data. I beginby proposing some plausible explanations of why Player As do not attempt to induceguilt upon Player Bs. Failure by Player A to attempt to induce guilt in �PPT (and�PPD) can result from Player A either (i) not conveying X = 0, and/or (ii) conveyingX = 6. The data revealed that several Player As did not convey X = 0 (30%) anddid convey X = 6 (60%), which suggests that failure to support H1 is primarily aresult of a combination of both (i) and (ii).

One possibility is that some Player As just don�t realize or perceive that guiltcould have been induced by the method proposed. Consequently, this type of PlayerA would not necessarily be motivated to convey X = 0 and/or not convey X = 6 asa means of inducing guilt. The questionnaire data presented in the previous sectionprovided some evidence consitent with there being this type of Player A; namely, we

19

saw that the Player As who made conveyance decisions that were inconsistent withattempted guilt induction did not perceive that guilt could have been induced byconveying X = 0 and not conveying X = 6: A second possibility is that some PlayerAs feel guilt over inducing guilt, an idea that BSH (1994) refer to as metaguilt. PlayerAs who are averse enough to these metaguilty feelings would then be motivated to notinduce guilt by not conveying X = 0 and conveying X = 6: A third possibility is thatPlayer As may have some aversion to being deceptive (see Gneezy 2005 and Erat andGneezy 2011 for studies relating to deception in games). Intentionally not conveyingknown payo¤ relevant information to Player B as a means of trying to manipulate thebehavior of Player B could be perceived as being deceptive. Hence, not conveying thevalue of X to Player B could be perceived as being deceptive, which would motivatedeception averse Player As to convey X, even when X = 6.

Another pattern that emerges from the data is Player Bs propensity to chooseRight is not linearly increasing across the conveyance state. Player Bs are much morelikely to choose Right jX = 0 (40%) than Right jX = NC (11%), as compared toRight jX = 6 (6%). BSH (1994) posit that guilt is induced by conveying how onesu¤ers if another fails to act in a desired fashion. You can think of Player A choosingto convey X = 0 as an explicit attempt to induce guilt because Player A is actuallyconveying how he su¤ers. On the other hand, you can think of Player A choosingto not convey X = 6 as an implicit attempt to induce guilt because Player A isconveying how he su¤ers by not conveying how he doesn�t su¤er. This distinctionbetween what I refer to as explicit and implicit guilt induction might be importantin terms of the e¤ectiveness of guilt induction, in light of the fact that Player Bsare much more likely to choose Right after Player A had explicitly induced guilt byconveying X = 0 as compared to implicitly inducing guilt by not conveying X = 6:This suggests that guilt induction may be the most e¤ective when agents are ableto explicitly convey how they su¤er. That is, if an agent knows he will su¤er overanother�s failure to act in a desired fashion, then that agent may be able to moree¤ectively induce guilt by explicitly conveying that information to the other agent.

5 Guilt Aversion Theory

In this section, I explore the connection between the psychological insights of BSH(1994) regarding guilt induction and the formal model of guilt developed by B&D.In doing so, I �rst provide a brief outline of the B&D model of guilt, followed by itsapplication to �PPT . I then derive conditions under which the BSH (1994) method ofinducing guilt is consistent with predictions of the B&D model of guilt. I conclude thesection by showing that the method of guilt induction, as prescribed by BSH (1994),can be supported as an equilibrium in �PPT under the framework of the B&D model.

20

5.1 B&D Model of Guilt Applied to �PPTI begin by �rst providing a brief general outline of the model. B&D model two typesof guilt for a general class of extensive form games, simple guilt and guilt from blame.I only consider the former applied to �PPT , thus, I restrict my analysis to simpleguilt.20 What follows is only a simpli�ed outline of the model; Interested readersshould refer to B&D for the full, technical presentation of both models includingillustrative examples. Informally, the model posits that agents su¤er disutility, in theform of guilt, from failing to live up to others� expectations. This is captured bymodeling an agent�s utility as a function of his own material payo¤s, and the extentto which he let other agents down.

Formally, simple guilt is modeled by specifying a utility function for player i givenby:

uSGi = mi �Xj 6=i

�ij �Dj (Simple Guilt Utility)

In this expression, mi represents player i0s material payo¤, andPj 6=i�ij �Dj represents

player i0s disutility from simple guilt. The latter component is composed of two pieces.The �rst, �ij, is an exogenously given constant that measures player i0s sensitivity offeeling guilty toward player j: The second, Dj, represents the amount by which playeri lets player j down, as a result of player i0s strategy. Dj = Ej �mj is expressed asthe di¤erence between the material payo¤ that player j was expecting, Ej, and thematerial payo¤ that player j actually receives, mj:

21 Ej itself is a function of playerj0s strategy, and player j0s vector of ��rst-order�beliefs regarding the strategies ofthe other players. Note, player i does not actually observe Dj, as it is a functionof player j0s �rst-order beliefs. Therefore, it is assumed that player i maximizes theexpected value of uSGi , given player i0s �rst-order beliefs regarding player j0s strategy,and player i0s �second-order�belief regarding player i0s �rst-order beliefs.I proceed with the application of the B&D model of simple guilt to �PPT :22 In

20With simple guilt, an agent su¤ers disutility proportional to how much he/she lets down anotheragent. Whereas with guilt from blame, an agent su¤er disutility proportional to how much the otheragent blames him for being let down. Thus, the main di¤erence between the two models is the extentto which an agent can be blamed for letting down another agent. With respect to �PPT ; the twomodels are equivalent. Player A can unambiguously identify the action of Player B which impliesthat Player B will receive all the blame for letting Player A down. Although either of these modelsof guilt could be used for this analysis, I opt to apply the less complex model of simple guilt forclarity.21More precisely, B&D model the utility from simple guilt as uSGi = mi �

Pj 6=i�ij � Gij : In this

expression, Gij = Dj � minsi Dj measures the unavoidable amount by which player i lets downplayer j. However, in relation to �PPT ; Gij is equivalent to Dj so I opt to use the more simpli�ednotation.22The B&D model of simple guilt could similarly be applied to �UPT and �PPD. However, the

hypotheses that I develop regarding attempted guilt induction and the e¤ectiveness of guilt inductionare centered around �PPT : Therefore, for considerations of space, I only consider an application of

21

particular, I derive the guilt that Player B would experience from choosing Left in�PPT . When Player B is called upon to move, there are three possible histories thatPlayer B could observe. The �rst is that X = 0 was conveyed, which I denote C0.The second is that X = 6 was conveyed, which I denote C6. The third is that thevalue of X was not conveyed, which I denote CN . Henceforth, I refer to these threepossible histories as the three possible conveyance states. A strategy for Player B isa probability distribution over Player B�s possible actions, {Left, Right}, at each ofthe three possible conveyance states.

In deriving the guilt that Player B would su¤er from choosing Left, it is necessaryto �rst derive Player A�s material expectation. Before Player A makes her initialIn or Out decision, Player A forms an initial �rst-order belief regarding Player B�sstrategy. Player A�s initial �rst-order beliefs can be represented as the probabilitiesthat Player B would choose Right at each of the three conveyance states, which Idenote �A= (Pr(Right jC0);Pr(Right jCN);Pr(Right jC6)): For a given strategy, PlayerA forms an expectation,weighted over �A and moves by Nature, of his material payo¤.I denote Player A�s initial material expectation EA.

Player B experiences disutility from simple guilt when he fails to live up to PlayerA�s expectations i.e. chooses a strategy that yields a payo¤ to Player A that is lowerthan EA. However, when Player B is called upon to move, he does not observe EA:Player B must then form an expectation regarding EA; conditional on the conveyancestate: This conditional expectation is a function of Player B�s conditional second-orderbeliefs regarding �A; which I denote �B: I de�ne EBjh where h 2 fC0; CN ; C6g asPlayer B�s expectation of EA; conditional on the conveyance state: The guilt thatPlayer B would su¤er from choosing Left is proportional to the di¤erence betweenEBjh and mA; where mA the material payo¤ that Player A actually receives as aresult of Player B�s Left decision. Let �B � 0 denote Player B�s sensitivity to feelingguilty. Below, I derive Player B�s utility, including simple guilt, from choosing Leftat each of the three possible conveyance states.

X = 0 was conveyed: Because the conveyance is credible, Player B knows that if hechooses Left, Player A will receive a payo¤of mA = 0. Player B�s expectation ofPlayer A�s expectation is EBjC0: By choosing Left, Player B will su¤er disutilityfrom guilt equal to �B � (EBjC0�0): Thus, Player B�s utility from choosing Left,after X = 0 was conveyed, is equal to:

6� �B � (EBjC0 � 0)

the model to �PPT , and omit its application to �UPT and �PPD:

22

X = 6 was conveyed: Because the conveyance is credible, Player B knows that if hechooses Left, Player A will receive a payo¤of mA = 6. Player B�s expectation ofPlayer A�s expectation is EBjC6: By choosing Left, Player B will su¤er disutilityfrom guilt equal to �B � (EBjC6�6): Thus, Player B�s utility from choosing Left,after X = 0 was conveyed, is equal to:

6� �B � (EBjC6 � 6)

Value of X not conveyed If the value of X is not conveyed, then Player B mustthink about the expected material payo¤ that Player A would receive if hechooses Left. Let bmA = EB[mAjCN ] denote this expectation. As I have pre-viously shown (see Section 2.2), bmA 2 [1; 5]. Player B�s expectation of PlayerA�s expectation is EBjCN : By choosing Left, Player B will su¤er disutility fromguilt equal to: � � (EBjCN � bmA) where bmA 2 [1; 5]: Thus, Player B�s utilityfrom choosing Left, after the value of X was not conveyed, is equal to:

6� �B � (EBjCN � bmA) where bmA 2 [1; 5]

5.2 Guilt Induction via the B&D Model of Guilt

BSH (1994) argue that guilt is induced by conveying to another how one su¤ers whenthe other fails to act in a desired fashion. Is the BSH (1994) method of guilt inductionconsistent with the B&D model of simple guilt? That is, does the B&D model predictthat Player B would su¤er more guilt from choosing Left when X = 0 was conveyedcompared to when the value of X was not conveyed? Similarly, does the model predictthat Player B would su¤er more guilt from choosing Left when the value of X wasnot conveyed compared to when X = 6 was conveyed? These questions can both beconditionally answered in the a¢ rmative. In what follows, I derive conditions underwhich the BSH (1994) method for inducing guilt is consistent with the prediction ofthe B&D model of simple guilt.

In the previous section, I showed that the disutility, from guilt, that Player Bwould su¤er from choosing Left at each of the three possible conveyance states, aspredicted by the B&D model, is as follows:

� X = 0 was conveyed: �B � (EBjC0 � 0)

� X = 6 was conveyed: �B � (EBjC6 � 6)

� Value of X not conveyed: �B � (EBjCN � bmA) where bmA 2 [1; 5]

According to B&D, Player B would su¤er more disutility from choosing Left whenX = 0 was conveyed, compared to when the value of X was not conveyed, when:

23

�B � (EBjC0 � 0) � �B � (EBjCN � bmA) =) EBjC0 � EBjCN � bmA (Condition #1)

Similarly, Player B would su¤er more disutility from choosing Left when the value ofX was not conveyed, compared to when X = 6 was conveyed, when:

�B �(EBjCN� bmA) � �B �(EBjC6�6) =) EBjCN� bmA � EBjC6�6 (Condition #2)

Because bmA 2 [1; 5], we can derive a set of necessary and su¢ cient conditions onEBjh under which guilt induction, as prescribed by BSH (1995), is consistent with theprediction of the B&D model of simple guilt. The two conditions that are necessaryfor guilt induction by Player A upon Player B in �PPT are:EBjC0 � EBjCN � 5

EBjC0 � EBjCN � 5 (Necessary Condition #1)

EBjCN � EBjC6 � 5 (Necessary Condition #2)

The two conditions that are su¢ cient for guilt induction by Player A upon Player Bin �PPT are:

EBjC0 � EBjCN � 1 (Su¢ cent Condition #1)

EBjCN � EBjC6 � 1 (Su¢ cent Condition #2)

Essentially, Condition 1 states that Player B�s second-order belief of Player A�sexpectation after X = 0 is conveyed, EBjC0; is not too much lower than Player B�ssecond-order belief of Player A�s expectation after the value of X is not conveyed,EBjCN : Similarly, Condition 2 states that Player B�s second-order belief of PlayerA�s expectation after the value of X is not conveyed, EBjCN ; is not too much lowerthan Player B�s second-order belief of Player A�s expectation after X = 6 is conveyed,EBjC6: Conditions 1 and 2 would certainly be satis�ed if we assumed that Player Bdoes not update his belief of Player A�s expectation, i.e. EBjC0 = EBjCN = EBjC6:Such an assumption would be satis�ed in an equilibrium as I will discuss in thesubsequent section: However assuming that EBjC0 = EBjCN = EBjC6 is clearlystronger than is needed for Condition 1 and 2 to be satis�ed.

If Conditions 1 and 2 are satis�ed, then the B&D model predicts that Player Bwould feel more guilt from choosing left after X = 0 was conveyed and the value ofX was not conveyed, compared to when the value of X was not conveyed and X = 6was conveyed, respectively. Therefore, if Condition 1 and 2 hold, then according tothe B&D model, Player A can induce guilt upon Player B by choosing to ConveyX = 0 and Not Convey X = 6, which is consistent with the BSH (1994) method forinducing guilt.

24

5.3 Guilt Induction as an Equilibrium of �PPTThe primary motivation of the study is to experimentally investigate whether agentsstrategically induce guilt upon others, and whether agents respond in kind to strategicguilt induction. These are questions related to behavioral motivations in games thatare independent of any equilibrium supposition. Nevertheless, it is an importantquestion whether such behavior can be supported as an equilibrium of �PPT underthe formal guilt framework of B&D. It is the case that guilt induction by Player A,and a kind response to guilt induction by Player B, can be supported as sequentialequilibrium (SE) of �PPT (Battigalli and Dufwenberg 2009):23 The intuition behindthis rests in the fact that in a SE, an assessment (pro�le of behavioral strategies andconditional hierarchical beliefs) will be consistent. Battigalli and Dufwenberg showthat in equilibrium, �players never change their beliefs about the conditional beliefsthat the opponents would hold at each h (history)�(pp. 16). Thus, in equilibrium,there is no belief updating. It follows that EBjC0 = EBjCN = EBjC6, which impliesthat Conditions 1 and 2 from above would be satis�es in an equilibrium.

As I have shown, guilt induction by Player A in �PPT is characterized by thechoice to Convey X = 0 and Not Covey X = 6: Therefore, the strategy (In, ConveyX = 0; Not Convey X = 6) for Player A is consistent with attempted guilt induction.E¤ective guilt induction is characterized by Player B choosing Right as a responseto the guilt induction by Player A. The following two strategies for Player B areconsistent with responding in kind to Player A�s guilt induction: (RightjC0, RightjCN ,LeftjC6), and (RightjC0, LeftjCN , LeftjC6). Thus, the strategy pro�les ((In, ConveyX = 0; Not Convey X = 6), (RightjC0, RightjCN , LeftjC6)), and ((In, Convey X = 0;Not Convey X = 6), (RightjC0, LeftjCN , LeftjC6)) are the candidate equilibriumpro�les for e¤ective guilt induction in �PPT . In Claims 1 and 2 below, I show thateach of these strategy pro�les can be supported as an equilibrium of �PPT .

Claim 1 The strategy pro�le ((In, Convey X = 0; Not Convey X = 6), (RightjC0,RightjCN , LeftjC6)) can be supported as a SE of �PPT for �B 2 [25 ;

12]

To verify that this strategy pro�le is an equilibrium, we need to check that neitherplayer has a pro�table deviation. For Player A, this is rather trivial. By followingthe equilibrium strategy, Player A earns a payo¤ of 10, which is the highest payo¤ ofthe game. Therefore, Player A has no pro�table deviation. For Player B, we need toconsider deviations at each of the possible conveyance states. Given consistent beliefsin equilibrium, we have that �A= �B= (1; 1; 0); EA = EBjh = 10 8 h 2 fC0; CN ; C6g;and bmA = 5: Player B will not deviate to RightjC6 so long as: 6� �B � [10�6] � 4,�B � 1

2: Player B will not deviate to LeftjCN so long as: 4 � 6��B �[10�5], �B � 2

5:

Similarly, Player B will not deviate to LeftjC0 so long as: 4 � 6��B �[10�0], �B � 15

which is satis�ed if �B � 25:

23I refer the reader to the original paper for a formal equilibrium analysis of dynamic psychologicalgames. The authors extend the concept of sequential equilibrium by incorporating hierarchies ofconditional beliefs.

25

Claim 2 The strategy pro�le ((In, Convey X = 0; Not Convey X = 6), (RightjC0,LeftjCN , LeftjC6)) can be supported as a SE of �PPT for �B 2 [27 ; 1]

Again, to verify that this strategy pro�le is an equilibrium, we need to check thatneither player has a pro�table deviation. For Player A, playing the equilibrium strat-egy yields an expected payo¤ of 6.75, therefore Player A cannot pro�table deviateto Out. Player A would not deviate and Not Convey X = 0 which would result ina payo¤ of 0 compared to a payo¤ of 10 from following the equilibrium strategy toConvey X = 0: Player A is indi¤erent between Convey X = 6 and Not ConveyX = 6: Therefore, Player A has no pro�table deviation from the prescribed equilib-rium strategy. For Player B, we need to consider deviations at each of the possibleconveyance states. Given consistent beliefs in equilibrium, we have that �A= �B=(1; 0; 0); EA = EBjh = 7 8 h 2 fC0; CN ; C6g; and bmA = 5: Player B will not deviateto RightjC6 so long as: 6 � �B � [7 � 6] � 4 , �B � 2. Player B will not deviateto RightjCN so long as: 6 � �B � [7 � 5] � 4 , �B � 1: Similarly, Player B will notdeviate to LeftjC0 so long as: 4 � 6� �B � [7� 0], �B � 2

7.

Claims 1 and 2 show that strategic guilt induction can be supported as an equi-librium of �PPT under the formal guilt framework of B&D.24 I acknowledge thatan equilibrium supposition, especially when a game features multiple equilibria, is arather strong notion. However, an equilibrium supposition is su¢ cient, and not nec-essary, for the research hypotheses of this study to be consistent with predictions ofthe B&D model of guilt.

6 Conclusion

The motivation of this study is to investigate some of the interpersonal strategicimplications from guilt aversion. In doing so, I consider an experimental designcentered around a private information trust game ��PPT . This augmented trustgame features a strategic structure rich enough to allow the �rst mover (Player A) anopportunity to induce guilt upon the second mover (Player B) in a manner derivedfrom the psychological insights posited by BSH (1994). This game then enables me toexperimentally test whether Player As attempt to manipulate the behavior of PlayerBs by strategically induce guilt upon Player Bs, whether Player Bs are susceptibleto this type of guilt induction, and whether having the opportunity to induce guiltfosters more trusting behavior by Player As.

The experimental data reveals that Player Bs are susceptible to guilt induction byPlayer As. That is, after controlling for possible inequality aversion and reciprocitymotivations, Player Bs are more likely to choose the kind action toward Player A

24E¤ective guilt induction by Player A can similarly be supported in �PPD using the analagousstrategy pro�les (minus the In decision for Player A) speci�ed in Claim 1 and Claim 2, and thenapplying the B&D model, mutatis mutandis.

26

if Player A were to induce guilt upon Player B. This susceptibility of Player Bs toguilt induction is reinforced by data from a post-decision questionnaire where PlayerBs reported on their feelings of guilt. However, the data reveals that Player As donot attempt to induce guilt upon Player Bs, despite the fact that is would have beene¤ective. Although, the data from a post-decision questionnaire that asked PlayerAs to report on the guilt feelings of Player B reveals that there may be two types ofPlayer As: (i) Player As who think inducing guilt upon Player B would be e¤ectiveand, therefore, attempt to do it, and (ii) Player As who don�t think inducing guiltupon Player B would be e¤ective and, therefore, do not attempt to do it. Lastly, thedata reveals some evidence that Player As exhibit marginally more trusting behaviorwhen the strategic setting features and opportunity to induce guilt upon Player B.

Additionally, I show formally that the BSH (1994) method for inducing guilt,when applied to �PPT ; is consistent with predictions of the B&D model of simpleguilt. Speci�cally, I show that under certain reasonable conditions on Player B�ssecond-order beliefs, the B&D model predicts that Player B would su¤er more guiltfrom choosing the unkind action in �PPT after Player A had induced guilt uponPlayer B. Furthermore, it is the case that e¤ective guilt induction by Player A, i.e.,Player A attempting to induce guilt and Player B responding kindly after Player Ahad induced guilt, can be supported as equilibrium of �PPT under the B&D guiltframework.

The susceptibility of Player Bs to guilt induction that is observed in the datacan be viewed as additional experimental evidence consistent with the hypothesisthat agents are motivated by guilt aversion. Hence, the experimental design providesan alternative approach for investigating guilt aversion from those previously imple-mented. Namely, �PPT and �PPD provide a method for identifying behavior thatis consistent with guilt aversion without having to elicit beliefs, or convey elicitedbeliefs, both of which present previously established limitations (cf. Charness andDufwenberg, 2011).25 The ability to test for guilt aversion without eliciting beliefsis particularly relevant in light of the recent studies by Reuben et al. (2009) andEllingsen et al. (2010), which both test for the presence of guilt aversion using similarexperimental designs that feature belief elicitation, yet reach opposing conclusions.This paper joins Charness and Dufwenberg (2011), in its ability to test models ofbelief dependent utility without having to elicit or convey elicited beliefs.

BSH(1994) note that �guilt [induction] does not depend on formal power or in-

25Speci�cally, Dufwenberg and Gneezy (2000), Charness and Dufwenberg (2006), Bacharach et al.(2007), and Dufwenberg et al. (2011) elicited second order beliefs and test for a positive correlationbetween elicited second order expectations and actions. However, because these studies provide onlya correlation between elicited second order beliefs and actions, anchoring and false consensus e¤ectscannot be ruled out as possible explanations. Alternatively, Reuben et al. (2009) and Ellingsen et al.(2010) elicit �rst-order expectations of subjects, convey those expectations to the subject�s partner,and test for correlations between expectations and actions. However, the possibility of untruthfulreporting of beliefs and skepticism of conveyed beliefs arise with this approach, as noted by Reubenet al.

27

�uence and may even work best in the absence of such power, because one inducesguilt by depicting oneself as the helpless victim of another�s actions�(p. 247). Thissuggests that guilt induction could be particularly e¤ective in economies with lessdeveloped legal systems. In such economies guilt induction could serve as an infor-mal mechanism for enforcing contracts and mitigating corrupt behavior, that mightotherwise transpire in the absence of formal prohibitive legislation (Lee, 2010). Guiltinduction could also prove to be e¤ective at in�uencing behavior and impacting out-comes in credence goods markets (Dulleck and Kerschbamer, 2006; Beck et al., 2010;Dulleck et al., 2011; Balafoutas et al., 2011). In these credence goods markets, e.g.doctors, mechanics, taxis, or other expert services, the consumer is often the �helplessvictim�of the experts actions. Guilt induction by the consumer could be implementedto thwart opportunistic behavior by the expert, especially in developing economieswhere the incentives for opportunistic behavior are likely to be much stronger.26

Partnerships, principle-agent contracting, and employee-employer relationships,represent some of the many economic settings where trust is pivotal for successfuland e¢ cient relations. There is a growing body of literature that investigates theimportance of trust in social and economic settings, and how trust can be fostered(see Fehr, 2009; Charness et al., 2011 for reviews). Much of this literature focuseson the e¤ectiveness of reputation building in fostering trust.27 While there is oftenan incentive to trust in economic settings, this incentive is often o¤set by exposureto the risk of opportunistic behavior by the trusted agent. However, guilt inductionby the trusting agent can serve as a mechanism for thwarting such opportunisticbehavior, thus mitigating the risk associated with trusting actions. Therefore, havingan opportunity to induce guilt would then lead to more trusting behavior, which iswhat is observed in this paper. This might help explain why trust is so prevalent inmany economic interactions in our society today, where the strategic environmentsare often rich enough to allow the possibility to induce guilt.

I conclude by noting that the e¤ectiveness of guilt induction as an in�uence mech-anism in strategic settings may have limitations. In particular, repeated applicationsof guilt induction may become less e¤ective since the target of the guilt inductionwill likely become resentful or angered over its repeated application. This could ul-timately lead to less kind actions in response to guilt induction, which is counterto its intended purpose. BSH (1995) recognize this and argue that �although guiltmay often be an e¤ective way of getting one�s way, it appears to be costly and to

26Evidence of such opportunistic behavior by experts in credence goods has been found in a recent�eld experiment by Balafoutas et al. (2011). In particular, the authors �nd evidence of systematicover-charging and over-treating of passengers by taxi drivers in Greece.27That is, building a trustworthy reputation through prior trustworthy actions, that are observable

to other agents, induces agents to trust you in the future. Many experimental studies have foundevidence consistent with this �indirect reciprocity�including Bohnet and Huck (2004), Bolton et al.(2005), Greiner and Levati (2005), Seinen and Schram (2006), Du¤y et al. (2008), and Engelmannand Fischbacher (2009). Charness et al. (2010) provide a thorough review of these papers as wellas provide experimental evidence that a reputation of trusting behavior can foster trust.

28

carry some stigma. This suggests that inducing guilt may be a technique that hasto be used with caution and restraint�(p. 183). Perhaps guilt induction in strategiceconomic settings should be a mechanism that is reserved for instances when there islittle scope for reputational e¤ects, and when the payo¤ and potential risk associatedwith a trusting action are largest.

29

7 Appendix

Sample Instructions �PPT Treatment

Welcome and thank you for participating. Your participation is VOLUNTARY,and you may leave at any time. Feel free to raise your hand and ask questionsat any time, and you may refer back to these instructions at any time during thesession. Please remain seated and quiet for the remainder of the session. All decisionsare to be completed individually and interaction with other participants is strictlyPROHIBITED. Thank you for your cooperation.

Each person will receive a $5 show-up payment for participating. In addition, youcan receive additional compensation based on the decision(s) that are made in thedecision task described below. After the task is complete, you will be privately paidthe amount of money you have earned. Upon completions of the decision task, pleaseremain quietly seated in your carrel until you have been paid.

The Decision Task:

You will be participating in a 2-person decision task. Each person will be randomlyand anonymously paired with another person in the lab. In each of the 2-persondecision-making pairs, one person will be randomly assigned the role of PLAYERA and the other person will be randomly assigned the role of PLAYER B. You willremain in your assigned role for the entire session. The earnings of each Playerwill depend on the decision(s) he/she makes, and/or the decision(s) of the Playerwith whom they are paired. A brief outline of steps of the decision task will �rstbe provided, followed by a detailed description of each step and the correspondingearnings for each Player.

� Step 1: PLAYER A begins by �rst choosing IN or OUT.

� If PLAYER A chooses OUT, then the task ends.

� If PLAYER A chooses IN, then the task proceeds to Step 2.

� Step 2: PLAYER A might privately learn some payo¤ information that wasinitially unknown to both players. If PLAYER A does learn the information,the PLAYER A will then have an opportunity to convey the information toPLAYER B. The task will then proceed to Step 3.

� Step 3: PLAYER B chooses between RIGHT or LEFT, and the task ends.

30

Step 1: PLAYER A �rst chooses between IN or OUT.

� If PLAYER A chooses OUT, then the decision task ends. PLAYER A willreceive $6 and PLAYER B will receive $2.

� If PLAYER A chooses IN, then the task proceeds to Step (2) wherePLAYER Amight privately learn the unknown information, and then havean opportunity to convey that information to PLAYER B. After Step (2),the task will proceed to Step (3) where PLAYER B will then be asked todecide between RIGHT or LEFT.

Step 2: I postpone the details about the information that PLAYER A can possiblelearn, and convey to PLAYER B until after Step (3) is described. DescribingStep (3) �rst will help clarify Step (2).

Step 3: If PLAYER A chooses IN, at Step (1), then PLAYER B must choose betweenRIGHT or LEFT.

� If PLAYER B chooses RIGHT, then the decision task ends. PLAYER Awill receive $10 and PLAYER B will receive $4.

� If PLAYER B chooses LEFT, then the decision task ends and PLAYERA will receive $X and PLAYER B will receive $6. There is a 50% chancethat X = $0 and a 50% chance that X = $6. That is, X = $0 and X = $6are equally likely.

NOTE: When the decision task begins, neither PLAYER A nor PLAYER B knowsthe value of X. Therefore, PLAYER A does not know the value of X when he/shedecides between IN or OUT in Step (1). Now we proceed with the description of Step(2):

Step 2: If PLAYER A chooses IN, there is an 80% chance that PLAYER A will privatelylearn the value of X, and a 20% chance that PLAYER A will not privately learnthe value of X.

� If PLAYER A does learn the value of X (80% chance), then PLAYER Amust then decide whether or not to convey the value of X to PLAYER Bbefore PLAYER B makes his/her decision in Step (3).

31

� If PLAYER A does convey the value of X, then PLAYER B will knowthe value of X before he/she decides between RIGHT or LEFT in Step(3).

� If PLAYER A does not convey the value of X, then PLAYER B willnot know the value of X before he/she decides between RIGHT orLEFT in Step (3).

� If PLAYER A does not learn the value of X (20% chance), then PLAYERA will not have an opportunity to convey the value of X to PLAYERB. The task will proceed to step (3) where PLAYER B will then choosebetween RIGHT or LEFT without knowing the value of X.

Payo¤Table:The table below summarizes the earnings of each Player for each of the possible

outcomes in the decision task:

Decision Outcome PLAYER A Earnings PLAYER B Earnings

PLAYER A chooses OUT $6 $2

PLAYER A chooses IN and then:PLAYER B chooses RIGHT $10 $4PLAYER B chooses LEFT $X $6

There is a 50% chance X = 0 and a 50% chance X = 6

Each person will participate in this decision making task ONE time. After thetask has ended, the decision(s) of each Player and the corresponding earnings of eachPlayer will be revealed to both Players. Additionally, the value of X will be revealedto both PLAYER A and PLAYER B regardless of the decisions made in the task. Youwill then be asked to �ll out a short questionnaire that will take about 3 minutes tocomplete. Your answers to the questionnaire are con�dential and will not be sharedwith any other participants. After completion of the questionnaire, an Experimenterwill then come by and privately pay you your total experimental earning which equalsyour earnings from the decision task PLUS the $5 show-up payment. After you havebeen paid, you may quietly exit the lab.

32

Sample Questionnaire - Player A

1. What is your Gender?

2. How old are you?

3. What is your major?

4. Have you ever taken an economics course?

5. Are you currently carrying more than $10 in cash?

6. Rate the amount of guilt you think Player B felt (would have felt) from choosingLeft when (if) Player B knew the value of X was [true value of X]

7. Rate the amount of guilt you think Player B felt (would have felt) from choosingLeft when (if) Player B did know the value of X.

8. How did you hear about the Economic Science Lab?

9. Have you ever referred a friend to the Economic Science Lab?

10. Is English your �rst language?

33

References

[1] Bacharach, M., Guerra, G., & Zizzo, D. (2007). �The Self-Ful�lling Property ofTrust: An Experimental Study.�Theory and Decision 63, 349-388.

[2] Balafoutas, L., Beck, A., Kerschbamer, R., & Sutter, M. (2011). �What DirvesTaxi Drivers? A Field Experiment on Fraud in a Market for Credence Goods.�IZA Discussion Paper.

[3] Battigalli, P., & Dufwenberg, M. (2007). �Guilt in Games.�American EconomicReview 97, 170-176.

[4] Battigalli, P., & Dufwenberg, M. (2009). �Dynamic Psychological Games.�Jour-nal of Economic Theory 144, 1-35.

[5] Baumeister, R., Stillwell, A., & Heatherton, T. (1994). �Guilt: An InterpersonalApproach.�Psychological Bulletin 115, 243-267.

[6] Baumeister, R., Stillwell, A., & Heatherton, T. (1995). �Personal NarrativesAbout Guilt: Role in Action Control and Interpersonal Relationships.� Basicand Applied Social Psychology 17, 173-198.

[7] Beck, A., Kerschbamer, R., Qiu, J., & Sutter, M. (2010). �Guilt from Promise-Breaking and Trust in Markets for Expert Services �Theory and Experiment.�University of Gothenburg Working Paper.

[8] Berg, J., Dickhaut, J. & McCabe, K. (1995). �Trust, Reciprocity and SocialHistory.�Games and Economic Behavior 10, 122-142.

[9] Bohnet, I., & Huck, S. (2004). �Repetition and Reputation: Implications forTrust and Trustworthiness when Institutions Change.�American Economic Re-view Papers and Proceedings 94, 362-366.

[10] Bolton, G., Katok, E., & Ockenfels, A. (2005). �Cooperation Among Strangerswith Limited Information About Reputation.�Journal of Public Economics 89,1457-1468.

[11] Bolton, G., & Ockenfels, A. (2000). �A Theory of Equity, Reciprocity, and Com-petition.�American Economic Review 90, 166-193.

[12] Charness, G., Du, N., & Yang, C. (2011). �Trust and Trustworthiness Reputa-tions in an Investment Game.�Games and Economic Behavior 72, 361-375.

[13] Charness, G., & Dufwenberg, M. (2006). �Promises and Partnership.�Econo-metrica 74, 1579�1601.

[14] Charness, G., & Dufwenberg, M. (2010). �Bare Promises: An Experiment.�Economics Letters 107, 281-283.

34

[15] Charness, G., & Dufwenberg, M. (2011). �Participation.�American EconomicReview 101, 1211-1237.

[16] Cox, J. (2004). �How to Identify Trust and Reciprocity.�Games and EconomicBehavior 46, 260-281.

[17] Du¤y, J., Lee, Y., & Xie, H. (2008). �Social Norms, Information, and TrustAmong Strangers: An Experimental Study.�Mimeo.

[18] Dufwenberg, M. (2002). �Marital investments, time consistency, and emotions.�Journal of Economic Behavior and Organization 48, 57-69.

[19] Dufwenberg, M., Gächter, S., & Hennig-Schmidt, H. (2011). �The Framing ofGames and the Psychology of Play.� Games and Economic Behavior (In Press).

[20] Dufwenberg, M. & Gneezy, U. (2000). �Measuring beliefs in an experimental lostwallet game.�Games and Economic Behavior 30, 163-182.

[21] Dufwenberg, M. & Kirchsteiger, G. (2004). �A Theory of Sequential Reciprocity.�Games and Economic Behavior 47, 268-298.

[22] Dulleck, U. & Kerschbamer, R. (2006). �On doctors, mechanics, and computerspecialists: The economics of credence goods.�Journal of Economic Literature44, 5-42.

[23] Dulleck, U., Kerschbamer, R., & Sutter, M. (2011). �The economics of credencegoods: On the role of liability, veriability, reputation and competition.�AmericanEconomic Review 101, 526-555.

[24] Ellingsen, T., Johannesson, M., Tjotta, S., & Torsvik, G. (2010). �Testing GuiltAversion.�Games and Economic Behavior, 68, 95-107.

[25] Engelmann, D., & Fischbacher, U. (2009). �Indirect Reciprocity and StrategicReputation Building in an Experimental Helping Game.�Games and EconomicBehavior 67, 399-407.

[26] Erat, S., & Gneezy, U. (2011). �White Lies.�Management Science, Articles inAdvance, 1-11.

[27] Fehr, E. (2009). �On the Economics and Biology of Trust.�Journal of the Eu-ropean Economic Association 7, 235-266.

[28] Fehr, E., & Schmidt, K. (1999). �A Theory of Fairness, Competition, and Coop-eration.�Quarterly Journal of Economics 114, 817-868.

[29] Fischbacher, U. (2007). �z-Tree, Toolbox for Readymade Economic Experi-ments.�Experimental Economics 10, 171�178.

35

[30] Fong, Y., Huang, C., & O¤erman, T. (2007). �Guilt Driven Reciprocity in aPsychological Signaling Game.�Working Paper.

[31] Geanakopolos, J., Pearce, D. & Stacchetti, E. (1989). �Psychological games andsequential rationality.�Games and Economic Behavior 1, 60-79.

[32] Gneezy, U. (2005). �Deception: The Role of Consequences.�American EconomicReview 95, 384-394.

[33] Greiner, B., & Levati, V. (2005). �Indirect Reciprocity in Cyclical Networks: AnExperimental Study.�Journal of Economic Psychology 26, 711-731.

[34] Lee, S.Y., (2010). �Economics of Guanxi as an Interpersonal Investment Game.�Review of Development Economics 14, 333-342.

[35] Rabin, M. (1993). �Incorporating fairness into game theory and economics.�American Economic Review 83, 1281-1302.

[36] Reuben, E., Sapienza, P., & Zingales, L. (2009). �Is Mistrust Self-Ful�lling?�Economic Letters 104, 89-91.

[37] Seinen, I., & Schram, A. (2006). �Social Status and Group Norms: IndirectReciprocity in a Helping Experiment.�European Economic Review 50, 581-602.

[38] Tangney, J., & Fischer, K. (1995). Self-conscious emotions: Shame, guilt, em-barrassment, and pride. New York: Guilford Press.

[39] Vangelisti, A., Daly, J., & Rudnick, J. (1991). �Making People Feel Guilty inConversations: Techniques and Correlates.� Human Communication Research18, 3-39.

36