IDENTIFYING SOCIAL NORMS USING COORDINATION GAMES:WHY DOES DICTATOR GAME SHARING VARY? Erin L.Krupka Roberto A.Weber University of Michigan University of Zurich Abstract we intr changes in behavior across seve eral variants of the dictator game.including data from a nove experiment and from prior published laboratory studies.that are unaccounted for by most curren theories of social pre eover,we find that norm compliance and willingness to sacrifice money to take behaviors that are socially appropriate.(JEL:C91.C72. D64) 1.Introduction Social norms have long been recognized as an important influence on behavior in social sciences such as social psychology (Sherif 1936;Cialdini.Reno,and Kallgren 9)and sociology(Merton 7:Coleman9).However in ec onomics norms have received significant attention only relatively recently,mainly as a tool fo explaining seemingly anomalous behavior such as involuntary unemployment(Akerlot 1980),conformity (Bernheim 1994),costly punishment (Fehr and Gachter 2000). tipping(Conlin,Lynn,and O'Donoghue 2003),and macroeconomic phenomena,such why consumption may track income even when wealth levels remain unaffected of2007 One possible reason for the relative absence of social norms in economics is that they are difficult to measure or quantify,making it hard to predict the precise influence they will exert on behavior.As a result,social norms are usually incorporated into economic research as post-hoc interpretations for behavior or outcomes that are The editor in charge of this paper was Stefano DellaVigna iversity's Cente ns of Human Soc e D: DellaVigna.Ulrike Malmendier.an mous referee and part nts a cral wor .seminars.and s for helpful co pka is a ch Fellow at IZA. E-mail:ekrupka@umich.edu (Krupka);roberto.weber@econ.uzh.ch (Weber) D010.1111jcea.12006
IDENTIFYING SOCIAL NORMS USING COORDINATION GAMES: WHY DOES DICTATOR GAME SHARING VARY? Erin L. Krupka University of Michigan Roberto A. Weber University of Zurich Abstract We introduce an incentivized elicitation method for identifying social norms that uses simple coordination games. We demonstrate that concern for the norms we elicit and for money predict changes in behavior across several variants of the dictator game, including data from a novel experiment and from prior published laboratory studies, that are unaccounted for by most current theories of social preferences. Moreover, we find that the importance of social norm compliance and of monetary considerations is fairly constant across different experiments. This consistency allows prediction of treatment effects across experiments, and implies that subjects have a generally stable willingness to sacrifice money to take behaviors that are socially appropriate. (JEL: C91, C72, D64) 1. Introduction Social norms have long been recognized as an important influence on behavior in social sciences such as social psychology (Sherif 1936; Cialdini, Reno, and Kallgren 1990) and sociology (Merton 1957; Coleman 1990). However, in economics social norms have received significant attention only relatively recently, mainly as a tool for explaining seemingly anomalous behavior such as involuntary unemployment (Akerlof 1980), conformity (Bernheim 1994), costly punishment (Fehr and Gachter 2000), ¨ tipping (Conlin, Lynn, and O’Donoghue 2003), and macroeconomic phenomena, such as why consumption may track income even when wealth levels remain unaffected (Akerlof 2007). One possible reason for the relative absence of social norms in economics is that they are difficult to measure or quantify, making it hard to predict the precise influence they will exert on behavior. As a result, social norms are usually incorporated into economic research as post-hoc interpretations for behavior or outcomes that are The editor in charge of this paper was Stefano DellaVigna. Acknowledgments: The authors thank the Ford Foundation, IZA, and Carnegie Mellon University’s Center for Behavioral Decision Research for financial support. We also gratefully acknowledge support from the research priority program at the University of Zurich “Foundations of Human Social Behavior.” We thank Nicholas Bardsley, Colin Camerer, Cristina Bicchieri, Ted O’Donoghue, Tanga McDaniel, Stefano DellaVigna, Ulrike Malmendier, anonymous referees, and participants at several workshops, seminars, and conferences for helpful comments and suggestions. Krupka is a Research Fellow at IZA. E-mail: ekrupka@umich.edu (Krupka); roberto.weber@econ.uzh.ch (Weber) Journal of the European Economic Association June 2013 11(3):495–524 c 2013 by the European Economic Association DOI: 10.1111/jeea.12006
496 Journal of the European Economic Association otherwise difficult to explain (Fehr and Gachter 2000:Ostrom 2000),and they are tified primarily by r eoretically related to the norm (Fehr and Fischbacher 2004:Camerer and Fehr 2004).Because norms are usually studied indirectly in economics,they are rarely used to form precise predictions about behavior I In this paper,we aim to put the horse (norm)before the cart(behavior),by introd ed method for identifyings ial behavior separately use this method to measure social norms in several economi choice contexts,and then use these elicited norms to predict behavior a priori.We do so in the context of other-regarding behavior in variants of the "dictator game",where recent laboratory experiments demonstrate that minor contextual features of a choice environment lead to substantially different choices and outcomes.We show that such changes in behavior are ent ly cons sistent with varying social norms and with a stabl preference for complying with social norms Rather than attempting to develop a theory of norm compliance based on underlying preferences,as in prior research.3 we start with the assumption that individuals care about behaving in a manner consistent with social norms.More precisely,we assum ion makers'utility is ha ased on the mo they obtain an on the degree to which their actions comply with social nomms,in the form of taking actions generally viewed as socially appropriate and avoiding those viewed as socially inappropriate.We then show that these two considerations-combined with a novel. incentivized method for identifying social norms that uses coordination games explain significant beha iation in dictator g contribut games.Thus our primary the cal,one We present and sulting changes in behavior that are consistent with the apreference for complying alizin m such labora t leas tiall or com orm ior in and to ryin For the labora s are identi he reason why ome e might sh are in one dictator e nt but not i e same as why one tips at a cof shop but not at a f d restauran Fot examnle Andreoni and Bemnheim (2009) plore situations in which one individual unilaterally avior that Csntstromindividualscarigaboutom 15 ously defined al c(r). n which there is im plicitly a ent that it is the edge that x ma that norms change ac ss contexts.and ount for changes in behavior
496 Journal of the European Economic Association otherwise difficult to explain (Fehr and Gachter 2000; Ostrom 2000), and they are ¨ identified primarily by measuring behaviors that are theoretically related to the norm (Fehr and Fischbacher 2004; Camerer and Fehr 2004). Because norms are usually studied indirectly in economics, they are rarely used to form precise predictions about behavior.1 In this paper, we aim to put the horse (norm) before the cart (behavior), by introducing a novel incentivized method for identifying social norms separately from behavior. We use this method to measure social norms in several economic choice contexts, and then use these elicited norms to predict behavior a priori. We do so in the context of other-regarding behavior in variants of the “dictator game”, where recent laboratory experiments demonstrate that minor contextual features of a choice environment lead to substantially different choices and outcomes.2 We show that such changes in behavior are entirely consistent with varying social norms and with a stable preference for complying with social norms. Rather than attempting to develop a theory of norm compliance based on underlying preferences, as in prior research,3 we start with the assumption that individuals care about behaving in a manner consistent with social norms. More precisely, we assume decision makers’ utility is based on the money they obtain and on the degree to which their actions comply with social norms, in the form of taking actions generally viewed as socially appropriate and avoiding those viewed as socially inappropriate. We then show that these two considerations—combined with a novel, incentivized method for identifying social norms that uses coordination games— can explain significant behavioral variation in dictator games. Thus, our primary contribution is an empirical, rather than theoretical, one. We present and demonstrate 1. A few experimental studies manipulate the likely presence or strength of a social norm by varying features of a choice context (Krupka and Weber 2009; Andreoni and Bernheim 2009) and demonstrate resulting changes in behavior that are consistent with the influence of a norm or a preference for complying with a norm. 2. This apparent “instability” in behavior has led some researchers to question the value of generalizing from such laboratory experiments to the field (Levitt and List 2007). Our work at least partially addresses this concern by demonstrating how such behavior corresponds to varying and identifiable social norms. The sensitivity of behavior in the laboratory to the context of the experiment can be interpreted in a manner similar to how behavior in the field is sensitive to context and to varying social norms. For the laboratory data analyzed in this paper, we demonstrate that such sensitivity may be explained once varying social norms are identified. Therefore, the reason why someone might share in one dictator experiment but not in another very similar one might be the same as why one tips at a coffee shop but not at a fast-food restaurant, or in the United States but not in Europe. 3. For example, Andreoni and Bernheim (2009) explore situations in which one individual unilaterally shares wealth with another, and model a norm as the behavior that results from individuals caring about own wealth, intrinsically about fairness, and about how others perceive their concern for fairness. Their paper assumes an exogenously defined alternative (xF), on which there is implicitly agreement that it is the “fair” action for the decision maker to take, and they follow prior research in assuming that the equal (50–50) division of wealth is a natural reference point. They then show why pooling may occur at this alternative (as well as at other alternatives, under changing conditions). Andreoni and Bernheim acknowledge that xF may differ across contexts, and may thus account for varying behavior. This is similar to our main argument, that norms change across contexts, and can therefore account for changes in behavior, and suggests that our identification of what actions people agree upon as “appropriate” or “inappropriate” might provide an empirical basis for something like xF in their model
Krupka and Weber Identifying Social Norms Using Coordination Games 497 the usefulness of a novel method for norm elicitation by showing that the elicited social norms-when included as a component of utility in a conditional logit choice model- can account rather well for behavioral changes in experimental data.Moreover,we also show that the weights placed on money and norm compliance.in the estimated utility tant willingness to trade off roughly $5 of wealth in order to take actions that are socially appropriate. rather than socially inappropriate.The stability of these preferences thus allows a- priori predictions to be generated for new dictator-game contexts,once one obtains measures of the social appropriateness of different actions. We be defining cial no and mple utility fra ework for understanding their potential influence on choice.We then demonstrate how one can use coordination games to identify the social norms that make up one source of utility Using our utility framework,we show how the social norms we elicit from one set of individuals with the coordination games yield precise and testable predictions regarding the behavior of a new pe of particip nts,which we evaluate both with novel data from ew experi ment a d also using data from previously published experiment We find that the observed sensitivity of behavior to several surprising experimental treatments can be almost entirely explained by variations in social norms. We choose to study behavior in dictator games primarily for two reasons.First the simplicity and non- -strategic nature of the dictator game make it easy to establish altemative enviro nts in hich ca hold timportant tures of the choice faced by a decision maker(such as the set of possible payoffs and experimenta subjects'understanding of how actions map into payoffs),while varying the context in a way that is likely to influence norms.The nonstrategic nature of the dictator game also allows us to rule out the possibility that changes in behavior are due to changes in subjects'expectations about h oponents will behave.as would potentially be the case in a public goods.prisoner's dilemma,or trust games. Second,a primary motivation in our research is to provide an interpretation for several recent experiments in economics(Dana,Weber,and Kuang 2007;List 2007: Bardsley 2008:Lazear,Malmendier,and Weber 2012),all of which use variants of the dictator game.These experiments show that alternative treatments that make seemingly trivial or irrelevant ch s to the hoices ailable to a dictator evertheles surprising changes in behavior.We provide at least one possible account for these changes.Indeed,part of the value in our approach is that it can provide an explanation for why behavior changes across dictator-game variants in a manner not adequately accounted for by most current theories of social preferences.Thus,rather than adding to the literatur on social pr b ndu be differen leadin models.we puposely study simple decision contexts between whichthes oftheccount for all be variaon in bhavior acro the variants of the dictator game that we study here
Krupka and Weber Identifying Social Norms Using Coordination Games 497 the usefulness of a novel method for norm elicitation by showing that the elicited social norms—when included as a component of utility in a conditional logit choice model— can account rather well for behavioral changes in experimental data. Moreover, we also show that the weights placed on money and norm compliance, in the estimated utility parameters in the conditional logit model, demonstrate a fairly constant willingness to trade off roughly $5 of wealth in order to take actions that are socially appropriate, rather than socially inappropriate. The stability of these preferences thus allows apriori predictions to be generated for new dictator-game contexts, once one obtains measures of the social appropriateness of different actions. We begin by defining social norms and presenting a simple utility framework for understanding their potential influence on choice. We then demonstrate how one can use coordination games to identify the social norms that make up one source of utility. Using our utility framework, we show how the social norms we elicit from one set of individuals with the coordination games yield precise and testable predictions regarding the behavior of a new sample of participants, which we evaluate both with novel data from a new experiment and also using data from previously published experiments. We find that the observed sensitivity of behavior to several surprising experimental treatments can be almost entirely explained by variations in social norms. We choose to study behavior in dictator games primarily for two reasons. First, the simplicity and non-strategic nature of the dictator game make it easy to establish alternative environments in which we can hold constant important features of the choice faced by a decision maker (such as the set of possible payoffs and experimental subjects’ understanding of how actions map into payoffs), while varying the context in a way that is likely to influence norms. The nonstrategic nature of the dictator game also allows us to rule out the possibility that changes in behavior are due to changes in subjects’ expectations about how opponents will behave, as would potentially be the case in a public goods, prisoner’s dilemma, or trust games. Second, a primary motivation in our research is to provide an interpretation for several recent experiments in economics (Dana, Weber, and Kuang 2007; List 2007; Bardsley 2008; Lazear, Malmendier, and Weber 2012), all of which use variants of the dictator game. These experiments show that alternative treatments that make seemingly trivial or irrelevant changes to the choices available to a dictator nevertheless produce surprising changes in behavior. We provide at least one possible account for these changes. Indeed, part of the value in our approach is that it can provide an explanation for why behavior changes across dictator-game variants in a manner not adequately accounted for by most current theories of social preferences.4 Thus, rather than adding to the literature on social preferences by conducting a “horse race” between different leading models, we purposely study simple decision contexts between which these 4. In the Online Appendix, we consider several leading models of social preferences, and show that none of them can directly account for all the variation in behavior across the variants of the dictator game that we study here
498 Journal of the European Economic Association models often fail to discriminate,and show that empirically-measured differences in social norms across these co By applying our norm elicitation method to our own and others'dictator game data,we offer a unified interpretation of the behavioral changes observed across several experiments while simultaneously demonstrating the usefulness of the elicitation technique that we introduce.We also highlight the benefits of re-examining data from test novel the and interpreta as opposed to uniquely cons ering novel data generated solely for the purposes of a current test. We should also note that,while this paper is the first to introduce the coordination game-based elicitation method for identifying social norms,the approach has already been used in other papers.For example,Burks and Krupka(2012)ap ply the method in a real firm,to study social norms regarding behavior tow ard cli sers an es in perceived socia norms correlate with job dissatisfaction.In another example,Gachter,Nosenzo,and Sefton (2012)use our method to elicit social norms in a setting where two workers provide costly effort to a firm,in a "gift-exchange"setting.They elicit norms to demonstrate that it is more socially ap ropriate to work hard when the other worker exerts high effort,and compare the predictive ability of social norms to othe social preferences. The next section presents our method for identifying social norms.Sections 3-5 demonstrate the usefulness of this method for predicting and explaining behavior in nonstrategic choice environments.Finally.in the conclusion.we discuss related work that validates the orm elicitation method. 2.Defining and Identifying Social Norms Following Elster(1989),we note two important features of social norms.First,socia norms generally prescribe or proscribe behaviors or actions,rather than outcomes As Elster notes,"The simplest social norms are of the type:Do X,or:Don't do X.' (p.99).Allowing norms to govern actions.rather than outcomes.suggests that two actions that produce the same outcome,but differ in other respects,may be governed cial econd,the cial" element of no orms req t they be jointly recognized,or collectively perceived,by members of a population.These two features of social norms -that they apply to actions rather than outcomes and that they irical input that can imp ofsoarTm omo to provide a basis for While singl data is valuable for the purpos
498 Journal of the European Economic Association models often fail to discriminate, and show that empirically-measured differences in social norms across these contexts correspond to observed differences in behavior.5 By applying our norm elicitation method to our own and others’ dictator game data, we offer a unified interpretation of the behavioral changes observed across several experiments while simultaneously demonstrating the usefulness of the elicitation technique that we introduce. We also highlight the benefits of re-examining data from prior experiments to test novel theories and interpretations, as opposed to uniquely considering novel data generated solely for the purposes of a current test.6 We should also note that, while this paper is the first to introduce the coordination game-based elicitation method for identifying social norms, the approach has already been used in other papers. For example, Burks and Krupka (2012) apply the method in a real firm, to study how social norms regarding behavior toward clients differ between financial advisers and their supervisors, and show that mismatches in perceived social norms correlate with job dissatisfaction. In another example, Gachter, Nosenzo, and ¨ Sefton (2012) use our method to elicit social norms in a setting where two workers provide costly effort to a firm, in a “gift-exchange” setting. They elicit norms to demonstrate that it is more socially appropriate to work hard when the other worker also exerts high effort, and compare the predictive ability of social norms to other social preferences. The next section presents our method for identifying social norms. Sections 3–5 demonstrate the usefulness of this method for predicting and explaining behavior in nonstrategic choice environments. Finally, in the conclusion, we discuss related work that validates the norm elicitation method. 2. Defining and Identifying Social Norms Following Elster (1989), we note two important features of social norms. First, social norms generally prescribe or proscribe behaviors or actions, rather than outcomes. As Elster notes, “The simplest social norms are of the type: Do X, or: Don’t do X.” (p. 99). Allowing norms to govern actions, rather than outcomes, suggests that two actions that produce the same outcome, but differ in other respects, may be governed by different social norms. Second, the “social” element of norms requires that they be jointly recognized, or collectively perceived, by members of a population. These two features of social norms—that they apply to actions rather than outcomes and that they 5. Our approach might also serve as a valuable complement to existing theoretical approaches, providing a useful empirical input that can improve their ability to distinguish between environments that differ in social norms. For example, the norms that we empirically elicit could be incorporated into existing models of social norms, such as those by Lopez-Perez (2008) or Andreoni and Bernheim (2009) to provide a basis for why certain actions are considered “fair” or “compliant” with a norm, or to provide a basis for expectations in non-strategic settings (Battigali and Dufwenberg 2007). 6. While most experimental studies in economics (and psychology) limit their attention to a single dataset (usually generated by the authors), using additional, pre-existing data is valuable for the purposes of demonstrating robustness and generalizing findings beyond a particular experimental study. For other papers that employ this approach, see Camerer, Ho, and Chong (2004) and Hyndman et al. (2012)
Krupka and Weber Identifying Social Norms Using Coordination Games 499 must be jointly recognized-are present in most researchers'definitions(Bettenhausen and Murnighan 1991;Fehr and Gachter 2000).For example,Ostrom (2000)defines social norms as"shared understandings about actions that are obligatory,permitted, or forbidden"(pp.143-144,emphasis added). Frther,we distinguish regarding what norms,from customs or actions that people regularly take,or descriptive norms.Both kinds of norms influence behavior(Cialdini,Reno,and Kallgren 1990:Krupka and Weber 2009).However,our focus here is on injunctive social norms-that is,those described by Elster as prescribing what one“should do'”or“should not do'”.Asw will show,(injun tive)social n the appropr eness of behavio r ar sufficient for explainingaconsiderable amount of variation in other-regarding behavior Therefore.following the literature.we define (injunctive)social norms as collective perceptions,among members of a population.regarding the appropriateness of different behaviors.They are things that people in the population jointly recognize one should or should not do,and people wh o belong to the populatio expec thers to be aware of and understand this agreement.The power of s cial norms col mes both from the willingness of people within the population to punish(or reward)others'deviation from(or adherence to)them and from the experience of positive or negative emotions produced by one's own adherence or deviation from a social norm (Elster 1989:Fehr nd Gachter 2000;Lopez-Perez2008). To formalize tion,let A=(a.....ax)repre ent a set of K action available to a decision maker.A social norm,N(ag)[-1,1],is an empirically measurable collective judgment that assigns to each action a degree of appropriateness or inappropriateness.Therefore,we assume that if for an action,ag,there is collective recognition that the action constitutes"appropriate."or socially prescribed.behavior V(ag)>0,while if the recogniti that an acti s"inapprop ate or socially proscribed, havior,N(ak)<0.Thu con tent with the above definitions of social norms,N(a)identifies the degree to which a specific action.a.is collectively perceived as one that should or should not be taken. An important feature of the above definition,and where we depart from some prior work,is that a norm is not necessarily a binary classification,such that a particular ction(the"norm”,eg "tip 20%" orthe 50-50 split")should be taken,by as sumptio leaving all remaining actions as those (equally inappropriate)actions that should no be taken (Lopez-Perez 2008).Instead,our definition of a social norm applies to the entire set of possible actions,and allows actions to vary in the degree to which they are perceived as appropriate.Thus,we can characterize a social norm by the profile of eness r er all the act ailable toa decision m r.Fo exampl ere may be social agreement that it is always appropriate to arrive on ti me i many Western cultures.there may be some instances in which amriving five minutes late is less socially inappropriate (meeting friends at a bar)than others(arriving at a funeral).We will also see that representing social norms as a profile of appropriateness igning N(a)=I to only one action
Krupka and Weber Identifying Social Norms Using Coordination Games 499 must be jointly recognized—are present in most researchers’ definitions (Bettenhausen and Murnighan 1991; Fehr and Gachter 2000). For example, Ostrom (2000) defines ¨ social norms as “shared understandings about actions that are obligatory, permitted, or forbidden” (pp. 143–144, emphasis added). Further, we distinguish norms regarding what one “ought” to do, or injunctive norms, from customs or actions that people regularly take, or descriptive norms. Both kinds of norms influence behavior (Cialdini, Reno, and Kallgren 1990; Krupka and Weber 2009). However, our focus here is on injunctive social norms—that is, those described by Elster as prescribing what one “should do” or “should not do”. As we will show, (injunctive) social norms concerning the appropriateness of behavior are sufficient for explaining a considerable amount of variation in other-regarding behavior. Therefore, following the literature, we define (injunctive) social norms as collective perceptions, among members of a population, regarding the appropriateness of different behaviors. They are things that people in the population jointly recognize one should or should not do, and people who belong to the population expect others to be aware of and understand this agreement. The power of social norms comes both from the willingness of people within the population to punish (or reward) others’ deviation from (or adherence to) them and from the experience of positive or negative emotions produced by one’s own adherence or deviation from a social norm (Elster 1989; Fehr and Gachter 2000; Lopez-Perez 2008). ¨ To formalize our definition, let A = {a1,..., aK } represent a set of K actions available to a decision maker. A social norm, N(ak ) ∈ [−1, 1], is an empirically measurable collective judgment that assigns to each action a degree of appropriateness or inappropriateness. Therefore, we assume that if for an action, ak , there is collective recognition that the action constitutes “appropriate,” or socially prescribed, behavior, N(ak ) > 0, while if there is joint recognition that an action constitutes “inappropriate”, or socially proscribed, behavior, N(ak ) < 0. Thus, consistent with the above definitions of social norms, N(ak ) identifies the degree to which a specific action, ak , is collectively perceived as one that should or should not be taken. An important feature of the above definition, and where we depart from some prior work, is that a norm is not necessarily a binary classification, such that a particular action (the “norm”, e.g., “tip 20%” or “the 50–50 split”) should be taken, by assumption leaving all remaining actions as those (equally inappropriate) actions that should not be taken (Lopez-Perez 2008).7 Instead, our definition of a social norm applies to the entire set of possible actions, and allows actions to vary in the degree to which they are perceived as appropriate. Thus, we can characterize a social norm by the profile of appropriateness ratings over all the actions available to a decision maker. For example, while there may be social agreement that it is always appropriate to arrive on time in many Western cultures, there may be some instances in which arriving five minutes late is less socially inappropriate (meeting friends at a bar) than others (arriving at a funeral). We will also see that representing social norms as a profile of appropriateness 7. Such a definition is possible in our framework by, for example, assigning N(ak ) = 1 to only one action (the “norm”) and letting all other actions have a constant value of N(ak ) = 0
500 Journal of the European Economic Association ratings over all actions is validated in our analysis of the experimental data.We show that even though the mos socially appropriate behavior is the same across all the experiments we examine- share the experimental endowment equally(Andreoni and Bernheim 2009)-we also find that differences in the relative appropriateness of the other actions exert an important influence on behavior. To embed this definition of social norms in a simple utility framework-which will allow us to sub sequently estimate the hat individuals ha ve for compic relative to moneye assume thata both the monetary payoff produced by the selected action,(as),and the degree to which the action is collectively perceived as socially appropriate: u(a)=V(π(a)+yN(ag). (1) The function VO represents the value the individual places on the monetary payoff:we assume that this function is increasing in ()The parametery about to social n ays select th payoff-maximizing action.On the other hand,as y increases,an individual will derive greater utility from selecting actions that are socially appropriate relative to the utility from those that are not. It follows directly from these preferences that behavior may change substantially across choice en vironments in which the sets of payoffs are identic if the socia norms change.Consider two choice environments,A=(a.a2)and A'=(j) such that (a)=(a)>(a2)=(a2).Then,if there exist no social norms in either environment (N(ag)=N(a)=0,for k =1,2)the decision maker will choose a in the first environment and a:in the second.However.if social norms differ between the two choice er nte the individual may selec payoffs in the two environments.For example,if N()=N(N()N() then for some values of y a decision maker will select a in the first environment and a in the second environment.This is in spite of the fact that the most appropriate actions,az and a,are analogous across the two contexts. The above framework.while simple.presents a potentially useful approach for understanding how varying so cial no ms mig ht affect behavic when choice environments are payoff-equivalent.It also providesaprecise.and testab e,relationship between the degree of social approval of actions(N(a))and individuals'willingness to take those actions,provided one has a method for empirically measuring the"social appropriateness"of the different available actions. of this paper.we predict and explain behavior of social appropriateness(N().We first measure social appropriateness using a nove ocial,or derive utility from
500 Journal of the European Economic Association ratings over all actions is validated in our analysis of the experimental data. We show that even though the most socially appropriate behavior is the same across all the experiments we examine—share the experimental endowment equally (Andreoni and Bernheim 2009)—we also find that differences in the relative appropriateness of the other actions exert an important influence on behavior. To embed this definition of social norms in a simple utility framework – which will allow us to subsequently estimate the concern that individuals have for norm compliance, relative to money – we assume that a decision maker cares about both the monetary payoff produced by the selected action, π(ak ), and the degree to which the action is collectively perceived as socially appropriate: u(ak ) = V(π(ak )) + γ N(ak ). (1) The function V() represents the value the individual places on the monetary payoff; we assume that this function is increasing in π(ak ). The parameter γ ≥ 0 represents the degree to which the individual cares about adhering to social norms. An individual entirely unconcerned with social norms (γ = 0) will always select the payoff-maximizing action. On the other hand, as γ increases, an individual will derive greater utility from selecting actions that are socially appropriate relative to the utility from those that are not.8 It follows directly from these preferences that behavior may change substantially across choice environments in which the sets of payoffs are identical, if the social norms change. Consider two choice environments, A = {a1, a2} and A = {a 1, a 2}, such that π(a1) = π(a 1) > π(a2) = π(a 2). Then, if there exist no social norms in either environment (N(ak ) = N(a k ) = 0, for k = 1, 2) the decision maker will choose a1 in the first environment and a 1 in the second. However, if social norms differ between the two choice environments, the individual may select actions corresponding to different payoffs in the two environments. For example, if N(a1) = N(a 1) < N(a2) < N(a 2), then for some values of γ a decision maker will select a1 in the first environment and a 2 in the second environment. This is in spite of the fact that the most appropriate actions, a2 and a 2, are analogous across the two contexts. The above framework, while simple, presents a potentially useful approach for understanding how varying social norms might affect behavior even when choice environments are payoff-equivalent. It also provides a precise, and testable, relationship between the degree of social approval of actions (N(ak )) and individuals’ willingness to take those actions, provided one has a method for empirically measuring the “social appropriateness” of the different available actions. In the rest of this paper, we predict and explain behavior using elicited measures of social appropriateness (N(ak )). We first measure social appropriateness using a novel 8. Other researchers have noted that individuals care heterogeneously about norm compliance (Ostrom 2000, Andreoni and Bernheim 2009). Such heterogeneity in pro-social concern is also common in most models of social preferences (Fehr and Schmidt 1999; Andreoni and Miller 2002). Cases in which γ < 0, which we do not explore here, might correspond to individuals who are anti-social, or derive utility from violating norms
Krupka and Weber Identifying Social Norms Using Coordination Games 501 incentivized elicitation method.We elicit social norms over possible action choices across different contexts,from individuals who do not make choices in those contexts. We then observe how well the elicited social norms.when integrated into the above simple utility framework,explain the actual choices made by a separate group of individuals We measure the extent to which actions are socially appropriate or inappropriate by presenting respondents with a description of a choice environment,including all the possible available actions (i.e.,A=fa.....ax).We ask respondents to judge the sca appropriateness -that is,we elicit N(s),for all a four point that ran ges ov "very socially inappropria inappropriate"somewhat socially appropriate tovery socially appropiateW provide respondents with incentives not to reveal their own personal preferences but instead to match the responses of others.Thus,respondents play a pure matching coordination game(Schelling 1960:Mehta,Starmer,and Sugden 1994)in which their ipate the extent to which others will rate an action as socially appropriate and to respond accordingly. Because social norms reflect"collective perceptions",coordination games present a useful incentivized way to identify such socially held judgments.From a game- theoretic point of view.pure matching games such as the one we use in our experiment have many equilibria and nothin intrinsic to the payoffs of the gan me make equilibrium fa the oth Sche ling (1960 theorized and Mehta.Stamer,and Sugden (1)and Sugden(195)demonstrated that prominence derived from common culture and shared experiences can create focal points.In our experiment,we allow collectively recognized social norms to create focal points in the matching game.Therefore,our elicitation method will yield a representation of a ocial norm if(a)there is gen al social greement that ome actions are more or less and if (b)respondents attemptng tacitly match others'responses rely on such shared perceptions to help them do so. We begin by focusing on two payoff-identical variants of the dictator game.In Experiment I subiects see a description of one of these two choice environments including all the possible choices available to the"dictator."From these subjects. who ctually play the dictator game des d to m or see the othe we elicit social norms over actions in the described choice environment using th incentives we describe above.We then use the social norms elicited in Experiment 1 to predict behavioral changes across the two environments,and we test these predicted considering the tradeoff )and hav ine too mar (in which case it might be too difficult for s naps from the social norm. and oue 2011). s can be used with economic incentives to reveal shared understanding(see also Xiao
Krupka and Weber Identifying Social Norms Using Coordination Games 501 incentivized elicitation method. We elicit social norms over possible action choices across different contexts, from individuals who do not make choices in those contexts. We then observe how well the elicited social norms, when integrated into the above simple utility framework, explain the actual choices made by a separate group of individuals. We measure the extent to which actions are socially appropriate or inappropriate by presenting respondents with a description of a choice environment, including all the possible available actions (i.e., A = {a1,..., aK }). We ask respondents to judge the social appropriateness of each action—that is, we elicit N(ak ), for all ak - on a four point scale that ranges over “very socially inappropriate”, “somewhat socially inappropriate”, “somewhat socially appropriate”, to “very socially appropriate”.9 We provide respondents with incentives not to reveal their own personal preferences but instead to match the responses of others. Thus, respondents play a pure matching coordination game (Schelling 1960; Mehta, Starmer, and Sugden 1994) in which their goal is to anticipate the extent to which others will rate an action as socially appropriate or inappropriate, and to respond accordingly. Because social norms reflect “collective perceptions”, coordination games present a useful incentivized way to identify such socially held judgments. From a gametheoretic point of view, pure matching games such as the one we use in our experiment have many equilibria and nothing intrinsic to the payoffs of the game makes one equilibrium favored (or focal) over the others. Schelling (1960) theorized and Mehta, Starmer, and Sugden (1994) and Sugden (1995) demonstrated that prominence derived from common culture and shared experiences can create focal points. In our experiment, we allow collectively recognized social norms to create focal points in the matching game. Therefore, our elicitation method will yield a representation of a social norm if (a) there is general social agreement that some actions are more or less socially appropriate, constituting the social norm, and if (b) respondents attempting to tacitly match others’ responses rely on such shared perceptions to help them do so.10 We begin by focusing on two payoff-identical variants of the dictator game. In Experiment 1 subjects see a description of one of these two choice environments, including all the possible choices available to the “dictator.” From these subjects, who never actually play the dictator game described to them or see the other variant, we elicit social norms over actions in the described choice environment using the incentives we describe above. We then use the social norms elicited in Experiment 1 to predict behavioral changes across the two environments, and we test these predicted 9. The decision to have only four appropriateness categories was made after considering the tradeoff between having too few (in which case it would be harder to discriminate between degrees of appropriateness) and having too many (in which case it might be too difficult for subjects to match on the social norm, perhaps leading them to attempt to match using other focal principles). Further, we omitted the “neutral” category, as this would have been a focal point separate from the focal point stemming from the social norm. 10. Many previous researchers have noted the important relationship between social norms and equilibrium selection in games (Kandori 1992; Young 1998). Camerer and Fehr (2004) note that coordination games can be used with economic incentives to reveal shared understanding (see also Xiao and Houser 2011)
502 Journal of the European Economic Association effects of social norms using data collected from a second,separate,group of subjects who make choices in one of the ofour analysis involves identifving social norms ment 2) in previously studied additional variants of the dictator game (Lazear.Malmendier.and Weber 2012;List 2007;Dana,Cain,and Dawes 2006).Therefore,as part of Experiment of diffe available in thes r experimen ts We te that the identified social normsexplain considerable variation across treatments in both our owr experiment(Experiment 2)and across these previously studied experiments.We also use a conditional logic choice model to obtain estimates of the weights that individuals place on complying with social norms(y)and on monetary payoffs in several of these showthata stable setof weights cnexp a considerable amoun of the variation in behavio or across these experiments 3.Identifying Social Norms in Payoff-Equivalent Environments(Experiment 1) Consider the following two choice environments.In a"standard"dictator game,a decision maker initially receives $10 while another person receives $0.The decision maker must then decide how much,between $0 and $10.in one-dollar increments. to give to the other person.In a"bully"variant of the game,the decision maker and othe eive $5 and the deciding individual can giv t between $0 and $5,agair n in one person.Both choice environments offer the decision maker exactly the same eleven choices over final wealth allocations ranging from($10,$0)to ($0,$10),but vary in the actions required to obtain those dollar allocations. While the two choice sets are identical in terms of final payoffs,they differ in contextual features o the actior required to achieve those payoffs.In the standar case,any outcome other than ($10.S0)involves"giving"money to the other person; in the bully variant all outcomes from($10,$0)to(S6,$4)involve the decision maker taking from the other person.Therefore.it is possible that social norms govering the two sets of behaviors might differ,even though the resulting outcomes do not.In njecture that s s will diffe ver actions that involveaking versus"giving",holding the res that involve "taking less socially appropriate. To identify social norms in the two choice environments,we applied our elicitation method to obtain ratings of the extent to which different actions in the two environments are collectively perceived as socially appropriate or inappropriate.Subjects providing 2031.0 r "bully" e dictaorgame地n dng options (sec
502 Journal of the European Economic Association effects of social norms using data collected from a second, separate, group of subjects who make choices in one of the two environments (Experiment 2). A second part of our analysis involves identifying social norms governing behavior in previously studied additional variants of the dictator game (Lazear, Malmendier, and Weber 2012; List 2007; Dana, Cain, and Dawes 2006). Therefore, as part of Experiment 1, we also use our elicitation method to measure the degree of social appropriateness of different actions available in these particular experiments. We demonstrate that the identified social norms explain considerable variation across treatments in both our own experiment (Experiment 2) and across these previously studied experiments. We also use a conditional logic choice model to obtain estimates of the weights that individuals place on complying with social norms (γ ) and on monetary payoffs in several of these experiments, and show that a stable set of weights can explain a considerable amount of the variation in behavior across these experiments. 3. Identifying Social Norms in Payoff-Equivalent Environments (Experiment 1) Consider the following two choice environments. In a “standard” dictator game, a decision maker initially receives $10 while another person receives $0. The decision maker must then decide how much, between $0 and $10, in one-dollar increments, to give to the other person. In a “bully” variant of the game, the decision maker and other person both initially receive $5 and the deciding individual can give or take any amount between $0 and $5, again in one-dollar increments, to or from the other person. Both choice environments offer the decision maker exactly the same eleven choices over final wealth allocations ranging from ($10, $0) to ($0, $10), but vary in the actions required to obtain those dollar allocations.11 While the two choice sets are identical in terms of final payoffs, they differ in contextual features of the actions required to achieve those payoffs. In the standard case, any outcome other than ($10, $0) involves “giving” money to the other person; in the bully variant all outcomes from ($10, $0) to ($6, $4) involve the decision maker “taking” from the other person. Therefore, it is possible that social norms governing the two sets of behaviors might differ, even though the resulting outcomes do not. In particular, we conjecture that social norms will differ over actions that involve “taking” versus “giving”, holding the resulting payoffs constant, in a manner that makes actions that involve “taking” less socially appropriate. To identify social norms in the two choice environments, we applied our elicitation method to obtain ratings of the extent to which different actions in the two environments are collectively perceived as socially appropriate or inappropriate. Subjects providing 11. Our experiment joins other research that examines the effect of varying initial endowment levels, such that dictators may “take” money from the recipient (Cox, Friedman, and Gjerstad 2007, Swope et al. 2008). Our “bully” variant differs from the dictator games with taking options studied by List (2007) and Bardsley (2008), which modify the standard dictator game by introducing additional taking options (see Section 5.2)
Krupka and Weber Identifying Social Norms Using Coordination Games 503 the ratings saw only one of the two choice environments,and received incentives to match the modal response provided by others rating the same choice environment. 3.1.Experimental Design for Experiment 1 ms of expe partic pants at Car Participants received S7 for showing up to the experiment and could earn additional money from a task in which they attempted to match others'appropriateness ratings. Subject payment in the matching task was not tied to the hypothetical dictator games about which they read. The instru ions (see Online Appendix)explained that subjects would read descriptions of different situations in which a person("Individual A")faced a choice among several possible alternatives.For each situation,subjects were asked to rate the extent to which each alterative available to the person was "'socially appropriate' and 'consistent with moral or proper social behavioror'socially inappropriate'and inconsistent with mo ral or ocial behavior how they might indicate their ratings for each action in this situation.3 After subjects were led through the example situation,but before they began to fill out the tables for the actual situations,they were told that one of the situations for which they were to dhatoicehtobedmy e possible choices in this situation would also be e randomly as the modal response in the session,then that participant would receive an additional payment($5 in Pittsburgh,$10 in Michigan)at the conclusion of the session.Thus, participants were incentivized to match the modal rating in their session,for each Subjects then saw a description of the standard or bully variant of th dictator game.Subjects in Experiment 1 never actually played this game,but only read about the situation and were asked to consider all of the actions that Individual A(the dictator)could take.In each session,only one of these two variants was used,meaning that no subject read descriptions of both the bully and standard choice contexts.The descriptio of th on stated that the target individual (IndividualA)was matche with another random and anonymous person(Individual B)and that both people would th 20 and Mic akin the the wallet belonged to them.eavin the allet the store manager
Krupka and Weber Identifying Social Norms Using Coordination Games 503 the ratings saw only one of the two choice environments, and received incentives to match the modal response provided by others rating the same choice environment. 3.1. Experimental Design for Experiment 1 We recruited 199 subjects from populations of experimental participants at Carnegie Mellon University, the University of Pittsburgh, and the University of Michigan.12 Participants received $7 for showing up to the experiment and could earn additional money from a task in which they attempted to match others’ appropriateness ratings. Subject payment in the matching task was not tied to the hypothetical dictator games about which they read. The instructions (see Online Appendix) explained that subjects would read descriptions of different situations in which a person (“Individual A”) faced a choice among several possible alternatives. For each situation, subjects were asked to rate the extent to which each alternative available to the person was “ ‘socially appropriate’ and ‘consistent with moral or proper social behavior’ or ‘socially inappropriate’ and ‘inconsistent with moral or proper social behavior’.” Participants then read, as an example, a hypothetical situation and were shown how they might indicate their ratings for each action in this situation.13 After subjects were led through the example situation, but before they began to fill out the tables for the actual situations, they were told that one of the situations for which they were to provide appropriateness ratings would be selected at random at the end of the session, and that one of the possible action choices in this situation would also be randomly selected. If, for this action choice, the participant’s appropriateness rating was the same as the modal response in the session, then that participant would receive an additional payment ($5 in Pittsburgh, $10 in Michigan) at the conclusion of the session. Thus, participants were incentivized to match the modal rating in their session, for each possible action. Subjects then saw a description of either the standard or bully variant of the dictator game. Subjects in Experiment 1 never actually played this game, but only read about the situation and were asked to consider all of the actions that Individual A (the dictator) could take. In each session, only one of these two variants was used, meaning that no subject read descriptions of both the bully and standard choice contexts. The description of the situation stated that the target individual (Individual A) was matched with another random and anonymous person (Individual B) and that both people would 12. Sessions conducted in Pittsburgh used 115 subjects and were conducted using pen and paper, while the sessions conducted at Michigan used 84 subjects and were conducted using the software z-tree (Fischbacher 2007). We find virtually no difference between the ratings obtained in Pittsburgh and in Michigan, despite there being differences in procedures (choices collected on paper in Pittsburgh versus through a computer in Michigan; slightly different payoffs; sessions conducted years apart), indicating robustness of the social norms we elicit in these populations. We therefore pool across the two locations in the analyses. 13. In the example, the decision maker found a wallet at a coffee shop and faced four alternatives: taking the wallet, asking others if the wallet belonged to them, leaving the wallet alone, or giving it to the store manager
504 Journal of the European Economic Association receive a"small participation fee"as well as any money produced by Individual A's The description then listed the eleven action choices available to Individual A. The labels associated with these action choices varied depending on which dictator game variant subjects were asked to consider(see Table 1).Subjects were also shown hoy paymens to each indv y evry lised acton ice.For ea possible bject had to ra the choice as either"very socially inappropriate""somewhat socially inappropriate" “somewhat socially appropriate”,or“very socially appropriate'”,with the goal of matching this rating to the modal response in the session. In the above manner.each subiect provided social appropriateness ratings for all actions available in either the standard or bully variant of the dictator g me.This vields our primary outcome measured abjects”elicite ratings of social appropriateness N()for the bully and standard choice environments After rating all actions in either the bully or standard dictator variants,subjects then saw descriptions of either four(Pittsburgh)or five (Michigan)additional variants of the dictator game.Each situation corresponded to a variant of the dictator game used mental Malmendier and Weber 2012:List 2007 Dana.Cain.and Dawes 2006).We discuss these variants in more detail in Section 5 After subjects indicated social appropriateness ratings in all choice scenarios,the experimenter randomly selected one scenario and one possible action choice in that scenario.The experimenter computed the modal response for that choice and privately ect s of whethe r no tthei modal rating Subjects were then paid privately in cash.rec an additional payment if they had selected the modal appropriateness rating for the selected scenario and action. 3.2.Results of Experiment 1 Recall that we conjectured that"taking"would generally be considered less socially n“giving",even when t hey produc cediden The or thah that eft the dictator dndivdunl cal with more money than the recipient,the corresponding actions would be generally considered less socially appropriate in the bully variant than in the standard variant of the dictator game We c verted subjectsresponses into numerical scores.Arating of "very socially appropriate "received a score of-1."somewhat socially inappropriate"a score o rtTblemy Each row corresponds to one possible action choice that Individual A could take and fencorcsof-l
504 Journal of the European Economic Association receive a “small participation fee” as well as any money produced by Individual A’s choice. The description then listed the eleven action choices available to Individual A. The labels associated with these action choices varied depending on which dictator game variant subjects were asked to consider (see Table 1). Subjects were also shown the monetary payments to each individual (A and B) produced by every listed action choice. For each possible action choice available to Individual A, a subject had to rate the choice as either “very socially inappropriate”, “somewhat socially inappropriate”, “somewhat socially appropriate”, or “very socially appropriate”, with the goal of matching this rating to the modal response in the session. In the above manner, each subject provided social appropriateness ratings for all actions available in either the standard or bully variant of the dictator game. This yields our primary outcome measured in Experiment 1—the “between-subjects” elicited ratings of social appropriateness N(ak ) for the bully and standard choice environments. After rating all actions in either the bully or standard dictator variants, subjects then saw descriptions of either four (Pittsburgh) or five (Michigan) additional variants of the dictator game. Each situation corresponded to a variant of the dictator game used in previous experimental research (Lazear, Malmendier, and Weber 2012; List 2007; Dana, Cain, and Dawes 2006). We discuss these variants in more detail in Section 5. After subjects indicated social appropriateness ratings in all choice scenarios, the experimenter randomly selected one scenario and one possible action choice in that scenario. The experimenter computed the modal response for that choice and privately informed subjects of whether or not their appropriateness rating matched the modal rating. Subjects were then paid privately in cash, receiving a $7 participation fee and an additional payment if they had selected the modal appropriateness rating for the selected scenario and action. 3.2. Results of Experiment 1 Recall that we conjectured that “taking” would generally be considered less socially appropriate than “giving”, even when they produced identical outcomes. Therefore, we expected that, for those wealth allocations that left the dictator (Individual A) with more money than the recipient, the corresponding actions would be generally considered less socially appropriate in the bully variant than in the standard variant of the dictator game. We converted subjects’ responses into numerical scores. A rating of “very socially inappropriate” received a score of –1, “somewhat socially inappropriate” a score of –1/3, “somewhat socially appropriate” a score of 1/3, and “very socially appropriate” a score of 1.14 Table 1 presents subjects’ social appropriateness ratings by condition. Each row corresponds to one possible action choice that Individual A could take and 14. We chose this particular scoring because it is intuitive (the least and most appropriate possible ratings receive scores of –1 and 1, respectively) and simple (possible ratings are evenly spaced over the –1 to 1 interval)
