How norms can generate conflict

Transcription

1 How norms can generate conflict approx. 11,231 words (all incl.), 11 figures, 2 tables Fabian Winter, 1,2, Heiko Rauhut, 3 Dirk Helbing 3 1 University of Leipzig, Department of Sociology Beethovenstraße 15, Leipzig, Germany 2 Max Planck Institute of Economics Kahlaische Straße 10, Jena, Germany 3 ETH Zurich, Swiss Federal Institute of Technology UNO D11, Universitätstraße 41, 8092 Zurich, Switzerland To whom correspondence should be addressed: winter@econ.mpg.de May 22, 2009 Abstract Norms are known to play an important role in establishing social order. The current literature focuses on the emergence, maintenance and impact of norms with regard to coordination and cooperation among individuals. However, the issue of norm-related conflict deserves more attention. Our first aim is the development of a general theory of normative conflict by differentiating between inter-norm and intra-norm conflict. The first kind of conflict results from distinct expectations of which means should be chosen to fulfil the norm. The second one results from distinct expectations of how strong the normative obligation should restrain the self-interest. Our second aim is to demonstrate the empirical relevance of normative conflict. We confirm the existence of normative conflicts with an experiment that applies a variant of the strategy method to the ultimatum game. Our experimental data reveal prevalent normative conflict among different types of actors, in particular among egoistic, equity, equality 1

2 and cherry picker types. We believe that our new focus on the double-edge of social norms in promoting cooperation on one side but conflict on the other is becoming increasingly important in a globalized world, in which behavioral expectations of interacting members of different cultures or religions may clash. Keywords: Social norms, normative conflict, cooperation, ultimatum game, strategy method, equity, intra-norm conflict, inter-norm conflict 2

3 Many of our daily activities are governed by social norms, which set the rules of how we ought to behave. Often, we are not even aware of how societal rules influence the way we speak, greet, dress, eat, or express gratitude or love. By simplifying the complexity of social life, norms serve as a lubricant of social order and facilitate social interaction. It is quite remarkable, moreover, that normative restrictions and constraints can have the paradoxical effect of allowing more freedom of action. Social norms can enhance the welfare of a group by proscribing the contribution to collective goods such as a clean environment, a safe neighborhood, public infrastructure facilities, trust in business relationships, reciprocal social relations or conflict resolution in general. In the recent discussion, the problem of norm emergence has garnered the lion s share of attention (Ullmann-Margalit, 1977; Bicchieri, 1990; Coleman, 1990; Voss, 2001). The topical argument explains the emergence of social norms by means of the shadow of the future. The expectation of future interactions can outweigh the temptation of one-sided, singular exploitation (Taylor, 1976; Raub and Voss, 1986; Fudenberg and Maskin, 1986; Ellickson, 1991). In addition, other mechanisms have been proposed, such as reputation (Raub and Weesie, 1990), signaling (Spence, 1974; Molm et al., 2000) and altruistic punishment (Ostrom et al., 1992; Fehr and Gintis, 2007). Our intention here is not to challenge the prominence of these contributions, but rather enrich the literature by demonstrating that all these routes to cooperation require a necessary precondition: Actors have to share equivalent social norms. The fulfilment of this requirement is not trivial, since there are numerous possible normative alternatives, which solve the same cooperation problem in different ways. Members of the same group can hold quite distinct behavioral expectations of how cooperation should be achieved. This phenomenon, referred to as normative conflict, does not generate cooperation, but conflict. If actors have internalized a different norm than their interaction partners, they can have the best intentions and do their best, but nevertheless, their behavior would be perceived as improper. They fall apart in conflict, despite both sides are convinced of having behaved in an adequate way. In the following, we develop a classification of normative conflict. This classification considers the inter-norm conflict on the one hand, which results from mutually exclusive behavioral expectations of how one ought to behave. On the other hand, our classification distinguishes the intra-norm conflict, which results from distinct expectations concerning how strong a 3

4 given norm should restrict the self-interest of the target actor. Subsequently, we outline our theory of normative conflict by exemplifying it with a simple formal model for the case of bargaining norms in the ultimatum game. Further, we present respective experimental data and study the empirical magnitude and significance of normative conflicts. 1 A perspective of normative conflict So far, we referred to social norms as a behavioral expectation regarding what ought to be or ought not to be done. For the establishment of our theory of normative conflict, we have to be more precise. First, let us refer to norm-relevant situations as situations in which social norms exist. In such situations, almost every member of a population believes that almost every other member has a certain behavioral expectation. These expectations are directed towards the so-called target actors of a norm, or shorter, norm targets. Moreover, norms are to the benefit of a certain group of actors, called beneficiaries of a norm. 1 We define social norm as a commonly known behavioral expectation among beneficiaries and target actors regarding how the target actors ought to behave, which is enforced by sanctions in case of violations. 2 Note that a norm target and a beneficiary can be the same person, but they can also belong to a different group of people. For the purpose of specifying the concept of normative conflict, we can subdivide two elements that build the structure of social norms. These two elements specify the factors that generate the behavioral expectation. We term these two elements the normative content and the level of normative commitment. This distinction enables us to classify two different types of normative conflict, as illustrated in the following. The normative content may be defined as the kind of behavior that is prescribed or proscribed in a norm-relevant situation. It can be understood as the method that ought to be considered by the norm targets to serve the beneficiaries interests. Figure 1 illustrates the interrelation of the normative expectation, normative content and level of normative commitment. In what 1 Note that the terms target actor and beneficiary stem from Coleman (1990, 247). 2 For a discussion of different definitions of social norms see Opp (2001) and Elster (1989), for a current review on social norms see Rauhut and Krumpal (2008), for literature on the punishment aspect see Yamagishi (1986); Heckathorn (1989), and for a microscopic foundation of coordination norms see Helbing (1992) and also Young (1993). 4

5 beneficiary normative expectations target content commitment Figure 1. Illustration of two different kinds of normative expectations in a normrelevant situation. The norm-relevant situation comprises of the beneficiary and the norm target, who both hold the expectation that the beneficiaries interests ought to be served by a certain normative behavior of the norm target. These expectations may be subdivided into the normative content, which denotes the means to accomplish the objective of the norm, and the level of normative commitment, which denotes how much the norm ought restrict the self-interest of the norm target. These two elements define two different classes of normative conflict. follows, we give two examples of norms, solving equivalent cooperation problems. These examples illustrate how norms with different normative contents may be regarded as equally valid to achieve cooperation; however, different actors may expect adherence to a different normative alternative, giving rise to normative conflict. Let us exemplify the normative content with the social norm to signal trustworthiness. Often there exist group norms that prescribe their members to signal their trustworthiness (cf. Raub, 2004). These signaling devices should be costly so that untrustworthy actors, who seek only short-term benefits, can be sorted out. It is, however, imaginable that different alternative norms are regarded as norm-relevant by different group members. One social group, for example, might demand the wearing of expensive suits from a certain brand, another group signifying piercings, and another tatoos. As a second example, consider reciprocity norms for the communication of affection. Let us imagine a man, who is dating a women and likes to disclose his affection and long-term interests for her. There may be a norm permitting the man to make her gifts, such as an invitation for dinner, handing over a ring, or writing a letter. However, some kinds of norms may be considered as appropriate, while others may not. The second element of norms considers that social norms imply obligations. As outlined above, the target actor is expected to restrict her selfinterest to enhance the beneficiaries well-being. Nevertheless, these restrictions can affect the self-interest in different ways and magnitudes. We will 5

6 term this relation the level of normative commitment. We may define the level of normative commitment as a function that specifies the tradeoffs between the normative objectives and the self-interest of the norm targets. This function prescribes a strong or weak restriction of the self-interest. Let us illustrate the concept of normative commitment by using the two examples from above. Regarding the norm to signal trustworthiness, the group members might be subject to severe restrictions, such as to place the tatoo on the face or the piercing in the lips. In contrast, the social norm could require less normative commitment, such as to mark the body at less visible and painful spots like the upper arm or the bum. Secondly, the reciprocity norm of communicating affection by gift giving can imply the normative commitment to invite the woman to a luxurious restaurant or rather to a cosy Italian restaurant nearby. Alternatively, a diamond or a plain ring can be presented. Or a love message may be transmitted in a short text message or a long letter. As revealed by the examples, actors can adhere to social norms, remain under the belief of behaving correctly and still have conflicts with each other. Consequently, we define normative conflict as the situation, in which the norm targets and the beneficiaries hold different behavioral expectations of how the targets ought to behave in a given norm-relevant situation. We can easily see from the examples above how normative conflict emerges due to the actors adherence to different normative alternatives or levels of normative commitment. The previous examples served to illustrate the general applicability of the concept of normative conflict to various norm-relevant situations. In the following, we introduce the subdivision into inter- and intra-norm conflicts, reflecting the distinction between normative content and normative commitment. We will exemplify this subdivision for the case of asymmetric bargaining norms, because these norms serve as real-life examples of our subsequent, more detailed discussion of the theory of normative conflict and its experimental measurement. We consider asymmetric situations, in which a powerful actor is the target actor who is expected to compensate the less powerful beneficiary. We believe that especially these asymmetric norms trigger a great extent of inter- and intra-norm conflicts and therewith enable a sound empirical demonstration of normative conflicts. We conjecture that asymmetric norms provoke intra-norm conflicts due to the differential power between beneficiaries and targets on the one hand and inter-norm conflicts on the other due to the different perceptions about the appropriateness of 6

7 different means to serve the beneficiaries interests Inter-norm conflict Normative expectations can clash if individual actors live up to different moral standards. We may define the case of inter-norm conflict as the situation in which the difference in the behavioral expectations is due to the disagreement between norm targets and beneficiaries about the appropriateness of the normative content. The case of inter-norm conflict is relevant for situations in which different means are regarded as valid for serving the beneficiaries interests. Consider the social norm to offer a person in need a place in a public bus. Both seniority and pregnancy can be regarded as criteria for justifying the claim for a seat. An inter-norm conflict emerges, if at least one of these criteria is not shared among the norm target and the beneficiary. For instance, a teenager, the norm target, might not accept pregnancy as a criterion when asked for his seat by a benefitting pregnant woman although both may agree that seniority would be a valid criterion. Another example would be the case of divorce in which both partners agree that one party has to be compensated for having taken the burden of managing the household, bringing up the children and having to face a lower employment perspective after the divorce. An inter-norm conflict emerges if there is disagreement about the particular compensation norm. One party may prefer to retain the initial ratio of assets prior to marriage and the other party to consider only those assets that have been generated during the marriage. For example, if the wife was four times richer prior to her marriage, she may claim that she should be four times richer afterwards. In contrast, the husband may claim that the assets jointly generated during the period of the marriage shall be divided equally, regardless of the initial ratio of assets. 3 The advanced reader might note the analogy to the concepts conjoint and disjoint norm, which were introduced by Coleman (1990, 245) to differentiate between symmetric and asymmetric norms. Here, we make use of these ideas and consider disjoint norms to achieve a better measurement of normative conflict. 7

8 1.2 Intra-norm conflict Normative conflict can be invoked by different conceptions about the relevance and weight of the respective normative alternative. To be more specific, we define intra-norm conflict as the situation in which the difference in the behavioral expectations is due to the disagreement between the norm targets and the beneficiaries about the level of normative commitment. Let us consider our previous example of the norm in a public bus, that an old person should be the beneficiary of being offered a seat by a younger and therefore more powerful target actor. Imagine a young man, sitting on one seat and occupying one additional seat with his bag. When two elder people enter the bus, he removes the bag and offers the freed seat to one of the seniors. However, when being asked to stand up, so that the second senior can sit down, he refuses. This illustrates the agreement among the young man and the seniors that elder people should be supported, but a disagreement with regard to the extent. In the case of compensation after divorce, both partners may agree on the similar kind of compensation norm. Let us assume, both agree that the partner who has interrupted his or her work should be compensated based only on those assets that have been generated during the time of the marriage. However, the beneficiary of the compensation norm, who took care of the household, may expect to receive half of the common assets, while the target actor, who had been working, might perceive that a smaller fraction would be adequate. Despite the beneficiary and the norm target may agree on the equivalent normative content, they may disagree regarding the level of the normative commitment. We visualize our classification of normative conflict in figure 2. Note that the disagreement in the case of the inter-norm and the intra-norm conflict can occur only among the norm targets, only among the beneficiaries or between norm targets on the one hand and beneficiaries on the other. Further, one could extent the discussion and imagine combinations of intra- and internorm conflicts, which is denoted as multi-dimensional normative conflict in the upper right cell of figure 2. For example, there can be norm-relevant situations, in which the same group of beneficiaries favor different (asymmetric) norms to be imposed on the target actors. In the following, however, we concentrate on the pure cases of intra- and inter-norm conflict for the case of bargaining norms. 8

9 normative commitment different similar intra-norm conflict normative consensus multidimensional normative conflict inter-norm conflict similar different normative content Figure 2. A classification of normative conflict. The x-axis denotes whether actors adhere to norms with different contents of how to serve the beneficiaries interests. The y-axis denotes whether actors adhere to a different levels of normative commitment, implying a different extent of how much the social norm ought to restrict the agents in pursuing their self-interest. If actors adhere to the same normative content with the same level of normative commitment, we speak of normative consensus. An inter-norm conflict emerges if actors adhere to alternative norms with different contents. One group of norm targets may consider norm A as relevant, whereas another group of norm targets considers the norm B. The distinct behavioral expectations create conflict. An intra-norm conflict emerges if actors rely on the same normative content, but on a different level of normative commitment. One group of norm targets may consider to implement the content without any compromise, while another group may believe that one ought to balance normative expectation with the self-interest. The combination of inter- and intra-norm conflict is referred to as multi-dimensional normative conflict. 9

10 2 Bargaining norms as an exemplification of normative conflict Bargaining norms can serve as an illustration of both, intra- and inter-norm conflicts. The distribution of work in the household (Lundberg and Pollak, 1996), relative wage differences (Fehr and Gächter, 2000), or the criteria for the allocation of organ donations (Elster, 1992; Gross and Kriwy, 2008) are embedded in norm-relevant situations and thus prone to conflict. Conflict can arise, when a good is scarce and demand exceeds supply. In these situations, social norms shape the expectations of the distribution of these goods. Most norms of distributional justice are based on the principle of allocating resources according to the criteria of need, status, effort or equality (see Cook and Hegtvedt (1983) or Deutsch (2000) for an overview). Those who contribute more, who need more, or who have a higher status shall receive more (Homans, 1961; Blau, 1964; Adams, 1965; Cook and Emerson, 1978). These norms can be distinguished with respect to the content of the norm. Equity norms assert that the individual input is the only criterion which determines the individual output to a certain extend. The second part of the norm is determined by the level of normative commitment, stating how much the individual input should be weighted with respect to the individual self-interest. We can speak of a pure equity norm, if there is a full normative commitment, so that selfishness does not justify any deviance from the principle of effort. On the other hand, equality norms do not rely on individual effort, need or status as an element of the normative content. For equality norms, the material equality of outcomes is the only criterion that ought to be satisfied. If the norm prescribes full normative commitment, it does not allow deviation from the equal split. Figure 3 gives an exemplary illustration of both norms with respect to the relative effort of two actors. 2.1 Previous empirical results The empirical literature on bargaining norms supports that social groups are heterogenous in their adherence to either equity or equality norms (Lissowski et al., 1991; Jasso and Wegener, 1999; Liebig and Verwiebe, 2000). Equality norms are frequently observed in bargaining experiments. For instance the equal split is usually the modal outcome in the ultimatum game (Roth, 1995), and is also commonly observed in real world settings, such as the family or 10

11 relative outcome 100% 75% 50% 25% inter-norm conflict intra-norm conflicts (a) (b) (c) 100% 100% nc = equality κ = 1 conflict nc = equity κ = 1 conflict relative outcome 75% 50% 25% nc = equality κ = 1 conflict nc = equality κ = 0.5 relative outcome 75% 50% 25% nc = equity κ = 1 conflict nc = equity κ = 0.5 0% 0% 50% 100% relative effort 0% 0% 50% 100% relative effort 0% 0% 50% 100% relative effort Figure 3. Illustration of inter-norm conflict (left panel) and intra-norm conflict (middle and right panel) for equality and equity norms. NC denotes the normative content and κ the level of normative commitment. The inter-norm conflict between a holder of an equity norm (nc = equity, (κ= 1) and a holder of an equality norm (nc = equality, κ = 1) emerges as a result of the adherence to norms with differing contents. The inter-norm conflict increases with the differences in relative effort. Both norms demand an equal split for equal effort. However, conflict can arise if the actors hold different norms and contributed different relative efforts. The intra-norm conflict in the equality case (panel (b)) between a holder of a pure equality norm (κ = 1) and a holder of an equality norm with a proportion κ = 0.5 of selfishness emerges as a result of different levels of normative commitments. The same holds for the intra-norm conflict between a holder of pure equity norm (κ = 1) and the holder of an equity norm with a proportion of selfishness of κ = 0.5 (panel (c)). In both cases of intra-norm conflict, the conflict is a result of different levels of normative commitment. 11

12 among friends (Morgan and Sawyer, 1967). Equity norms are often found in survey and vignette studies in which the outcome of joint work had to be allocated (Konow, 1996). According to these studies, about two thirds of the respondents regarded it as fair if income was allocated according to the working time. In experimental studies, earned property rights, for instance buying in an auction the right to be the proposer in an ultimatum game (Güth and Tietz, 1986) or winning it in a quiz (Hoffman et al., 1994), dramatically decreased the proportion of equal offers in bargaining games. Moreover, the heterogeneity of norms was confirmed with evidence of multiple distribution norms within one group. In early reward allocation experiments, the participants were often payed for commonly solving an experimental task. Afterwards, the common money could be divided by one of the players. Mikula (1972) and Mikula and Uray (1973) report two-person games, where the over-achieving player applied the equality norm to compensate the under-performing partner. However, the under-performing player usually honored the efficient player s merits and allocated according to equity norms. 4 Thus, two norms were applied in the same situation. Charness and Grosskopf (2001) experimentally tested the preferences of players towards efficiency and equality. In their dictator-like experiment, norms of efficiency were often preferred to norms of equality. Their design let one person choose between two allocations, one dividing equally, the other dividing a larger amount of money unequally (usually in favor of the other person). Between 66 and 88 percent of dictators (N=61) chose the unequal allocation, sometimes even if they would have been better off with the equal but less efficient allocation. Finally, Gantner et al. (2001) experimentally identified different patterns of allocation norms, e.g. equality, equity and efficiency, but, as the previous studies, they did not explicitly investigate the consequences of competing norms and the emergence of normative conflict. 2.2 The model As can be seen in figure 3, equity norms have the same implications as equality norms for some situations, but their expectations differ in other cases. As long as the relative entitlements are the same, both norms imply an equal 4 The results might have been influenced by a generosity bias due to deception of the participants and due to the experimental design. See Konow (2003) for a discussion of this argument. 12

13 split if they imply full normative commitment. However, the differences between equity and equality norms increase with increasing differences in entitlements of ego and alter. Thus, we can regard the difference in entitlements as a measure of the strength of the inter-norm conflict. We have defined the equality of outcomes as the normative content of equality norms. As the equal share of a distribution depends on the number of actors n, we normalize the individual share and consider this requirement by a factor 1 n. The second part of a norm is given by the normative commitment κ ν [0, 1], where the index ν denotes the respective norm. For reasons of simplicity, we assume that a norm is given by the product of the normative content and the normative commitment, but one can easily apply other functional dependencies. We restrict κ ν to values between 0 and 1, as this parameter is a measure of egoism, not of altruism. For our case, we assume that the norm target of a norm of distributive justice is the person who has the power to decide how to divide a good between her and the beneficiary of the norm. Thus, we can write the share that an equality norm prescribes to offer to the beneficiary as κ equality 1 n. (1) Equity norms, on the other hand, do not refer to the equality of outcomes. Adams (1965) classical definition of equity refers to the equivalence of the quotient of outcome and investment (O i /I i = O j /I j ) for all involved actors i, j. As Harris (1976) points out, this formulation of equity is rather simple, but it captures the relevant point. We define I i as the individual effort of player i in a norm-relevant situation. For reasons of simplicity, we restrict I i to positive values and use I i = I i if and only if all I i are negative. We can write this assumption regarding the content of equity norms as I i n j=1 I j The formalization of the equality norm is analogous to the previous case, so the share demanded by an equity norm is written as the product of the equity rule and the normative commitment: κ equity I i n j=1 I. (2) j 13

14 The intra-norm conflict is defined as the conflict emerging from disagreement about the normative commitment. The probability of the intra-norm conflict is strictly increasing with the strength of conflict, which for the case of conflicting equality norms is given by ( κ1 C intra-equality = n κ ) 2 2 n (3) An example of this conflict is given in figure 3b. We can specify the strength of intra-norm conflict for the case of two equity norms as C intra-equity = ( κ 1 I i n j=1 I j κ 2 I i ) 2 n I j, (4) for which an example is given in figure 3c. Note that the strength of conflict is strictly increasing with the square of the differences in κ n. We assume, that smaller deviations between the values of κ n are perceived as more acceptable, and that greater deviations dramatically increase the probability of conflicts. We now continue our analysis of normative conflict by investigating the conflict among norms of distributive justice in a pure inter-norm conflict. We assume that the normative commitment imposed by the norm is equal for both norms (κ equity = κ equality ), but there is disagreement about the normative content. The strength of an inter-norm conflict is then given by C inter = ( κ equity I i n j=1 I j j=1 κ equality n ) 2 (5) (see figure 3a). A more general model of normative conflict would consider variable normative contents r, where equity or equality are only two of many possible contents, and a norm would generally be formalized by κ ν r ν. The strength of conflict could then allow to predict the probability of normative conflict based on the general model of normative conflict, which could be given by C = (κ ν,1 r ν,1 κ ν,2 r ν,2 ) 2. (6) The probability of conflict is even higher, if actors weight their selfishness stronger than demanded by the norm. One could, for instance, think of a generally shared norm that those who have the power to divide a good should 14

15 apply the equality rule with a commitment of κ = 0.5. This would result in a division of 75 % for the norm target and 25 % for the beneficiary in the 2-person case. This division would not foster conflict if both actors hold the same norm. A division of 90 % to 10 %, however, would be seen as a violation of the norm, as the division has a stronger bias towards egoistic preferences than demanded by the norm. The argument is also in line with recent developments in experimental economics, stating that social norms do influence behavior, but that the behavior is usually a tradeoff between the behavioral expectations and selfishness (see Camerer, 2003). 3 Conditions for measuring normative conflict Measuring normative conflict in the laboratory is a challenging task. We identify the following four criteria as necessary conditions for measuring normative conflict. First, the participants have to perceive a situation as norm-relevant. Thus, almost everybody has to believe that almost every other participant has a specific behavioral expectation. We will employ an ultimatum game (Güth et al., 1982), one of the most parsimonious method to measure normrelevant situations. In this game one proposer and one responder bargain over a given amount of money (the cake). The proposer offers a share of the cake to the responder. If the responder accepts the offer, she receives the share and the proposer can keep the rest of the cake. If the responder rejects the offer, the cake is lost and nobody gets anything. The subgame perfect equilibrium for players with standard preferences is rather simple: The responder accepts every offer because a little amount is still better than carrying the consequences of rejection and receiving nothing and because she is indifferent if the proposer keeps everything for herself. The proposer anticipates the responders choice, offers the smallest positive amount and the responder accepts. However, social norms complicate things, as the participants often identify the ultimatum game as a norm-relevant situation. Empirical results indicate that responders actually do have expectations about the proposers behavior so that low offers are frequently rejected. These rejections can be regarded as punishment of norm violations. Proposers anticipate such potential punishment and, therefore, offers below 20 percent of the cake are 15

16 rare and close-to-equal splits are the most frequent outcome (Roth, 1995; Cameron, 1999; Hoffman et al., 1996; Oosterbeek et al., 2004; Güth et al., 2007). Second, the measurement of inter-norm conflict requires that at least two normative alternatives are potentially influential for the specification of the normative behavior. In the usual ultimatum bargaining experiments, endowments are supposed to compensate the efforts of the participants, i.e. the time spent in the lab (for an overview see Güth, 1995). These efforts are usually the same for all participants, so that the straightforward allocation norm is to split the cake equally. Different achievements, however, can introduce effort as a different criterion, triggering the behavioral expectations of an equity norm. Third, the measurement of intra-norm conflict requires heterogeneity in the subjects level of normative commitment. This means that some actors should believe that the norm ought strongly restrict the pursuit of selfinterest, while others should believe that these contents ought only mildly restrict it. Norm targets and beneficiaries have different levels of normative commitment in situations, in which they have opposing interests. Such asymmetric situations are given in the case that the less powerful beneficiaries claim that they ought to be benefitted by a disjoint norm which is addressed to the more powerful target actors. The ultimatum game is an ideal representation of such asymmetric interests, which may provoke the emergence of conflicting disjoint norms with respect to the implied level of normative commitment. In less abstract terms, responders might claim a large compensation from the proposer, while the proposers might only consider a mild normative commitment. 5 5 We can specify in more detail why the ultimatum game implies asymmetric access to power for both egoistic and social preferences. In both cases, the responder represents the beneficiary of the respective norm, as she benefits from her anticipated willingness to punish norm violations. The norm target is the proposer, who has a restricted choice set due to the anticipated punishment of the responder. In the case of egoistic preferences, the position to divide the cake is associated with the maximum power, and the position to reject the offer is associated with the least power: While every accepted offer is a Nash equilibrium, the responder s threat to reject small offers is a blunt sword. In the case of non-egoistic preferences, however, the responder holds normative expectations towards the proposer that she has to offer at least a certain, decent amount. The stronger the normative commitment of the responder, the more credible is her threat to reject low offers and, therefore, the greater her power. It is, however, incredible that the responder will reject hyperfair offers, so that the ultimatum game remains asymmetric, even with the 16

17 Fourth, the quantification of inter-norm conflicts requires heterogeneity in the subjects beliefs of the validity of different normative alternatives. The trigger of heterogeneous efforts may enable the emergence of heterogeneous behavioral expectations in the population. Thus, the two alternative norms could induce different actors to adhere to different norms, making the internorm conflict visible. 4 Method 4.1 Participants In order to study norm conflicts, we have performed an ultimatum game under anonymous conditions in the lab. The subjects were 92 undergraduate students of University of Leipzig. 47 subjects were male and 45 female, coming from a wide range of academic disciplines. The experiment was conducted in two separate PC-Labs. Proposers and responders were always assigned to separate rooms. Subjects participated in the experiment in groups of two ( one proposer and one responder). Three of our experimental sessions consisted of ten groups, one of nine groups and one of seven groups Procedure Five days before the experiment, the subjects received a seven page long text of a Wikipedia entry on the Westminster Palace via . 7 An accompanying letter informed the subjects that their preparation of the text will influence their possible earnings in the experiment. We chose a rather specific topic to ensure that everybody actually had to learn the text and nobody could benefit from her respective field of studies (such as mathematics or paleontology). We induced a feeling of personal effort and inherent monetary earnings by asking the subjects to invest their time prior to the experiment. 8 consideration of social preferences. 6 In sessions 1 and 4, fewer subjects showed up than expected. 7 Wikipedia contributors, Westminster Palace, Wikipedia: The Free Encyclopedia, http : //de.wikipedia.org/wiki/p alaceof W estminster (accessed May 04, :40) 8 A univariate linear regression shows that the invested time prior to the experiment reflects the effort, as both variables correlate with β = 0.28, t= 2,79. 17

18 We measured this effort by conducting a respective knowledge quiz. 9 The experiment was conducted using the z-tree software developed by Fischbacher (2007). At the beginning of each session, the subjects were randomly assigned to one of the computer terminals. Some general instructions regarding the procedure were given on paper, informing the subjects about the following quiz. It was once again pointed out that their preparation of the text will have a strong influence on their performance in this quiz. The subjects had to answer twenty questions with the opportunity to earn up to 20 Euro (one Euro per question). The average earnings of the participants were Euro. After completing the quiz, the subjects received the instructions for the following experiment, and they had to perform a questionnaire up to three times, which allowed us to see whether everybody understood the rules. The experiment started when there were no further questions to the experimenter. Communication was prohibited from that point. 4.3 Design We applied the strategy vector method (Selten, 1967; Fischbacher et al., 2001; Falk and Fischbacher, 2002) to the standard ultimatum game. In contrast to the simple ultimatum game, the subjects did not have full information about the cake and their respective roles. Although they knew how much they earned in the quiz, they were not told how much their opponent won. The roles were not assigned until everybody made all her decisions, so that each subject had to make her decisions for both roles respectively. If a player earned x Euro in the quiz, she was asked how much she would offer, if she and the opponent together earned x Euro, if they together earned x + 1 Euro,..., x + 20 Euro. On the next screen, the subjects were asked for their acceptance threshold if they were in the role of the responder, given she and the opponent earned x, x + 1,..., x After this, the computer matched two players from different rooms according to their results in the quiz. 10 The roles of the proposer and the responder were assigned at random. The computer compared the proposer s offer for the actual size of the cake 9 Previous experiments (Gächter and Riedl, 2005; Rauhut, 2009) used the idea of a knowledge quiz. However, we extended the design using a real effort task. 10 We implemented two mechanisms: best against best to enhance the variance in the stake size and best against worst to enhance variance between subjects. Matchingrelated results are not discussed in this paper for length restrictions, but available upon request. 18

19 with the corresponding threshold of the responder. The money was paid, if the offer was as high or higher than the responders acceptance threshold, otherwise the money was lost. Subjects were individually paid at their seats at the end of the session. 5 Results 5.1 Heterogeneous normative behavior on the macro level In section 3, we have identified necessary conditions for measuring normative conflict. We begin this section by showing that these prerequisites are met on the macro level (sec. 5.1) as well as on the micro-level (sec. 5.2). Subsequently, we investigate whether our findings confirm our perspective on normative conflict (sec. 5.3). Our statistical estimation results are based on multilevel models. These models quantify the impact of effort on the proposer s offer and on the responder s acceptance threshold, and, more importantly, the variance in the subjects level of normative commitment in their consideration of effort. The multilevel structure takes the clustering of the 21 decisions per subject into account. 11 Table 1 demonstrates the coherence of our data with our four conditions for measuring normative conflict. First of all, the participants do understand the game as a norm-relevant situation. There is significant evidence that most players are guided by social norms rather than by playing the subgame perfect Nash equilibrium of offering and accepting the smallest possible amount: The intercept of the proposer shows that even for the case that the responder contributed nothing, proposers offer 33 % of the cake on average. Furthermore, the responders positive intercept of 31 % in the fixed-effects part of the model reveals that the responders are punishing norm violations at own costs, which supports further the claim that responders perceive the ultimatum game as normrelevant. Second, the requirement that at least two norms with different content are influential for the specification of the normative behavior proved to be 11 Each subject provided a conditional proposal and an acceptance threshold for every possible effort of her opponent. For the exemplification of multilevel analysis see Snijders and Bosker (1999); Raudenbush and Bryk (2002); Rabe-Hesketh and Skrondal (2005). 19

20 Table 1. Linear multilevel model for the impact of the responder s effort on the proposer s offer and on the responder s acceptance threshold. Fixed effects N=1931 decisions Proposer s offer Responder s threshold Intercept (0.021) (0.025) Responder s effort Random effects J= 92 subjects (0.041) (0.037) Standard deviation intercept (0.015) (0.018) Standard deviation responder s effort (0.030) (0.027) Correlation (responder s effort/intercept) (-0.090) (-0.065) Log-Likelihood Observations Standard errors in parentheses p < 0.05, p < 0.01, p < Note: The model reveals that (1) subjects do adhere to norms of distributional justice, as the intercepts are significantly different from zero. (2) The effort has a large average impact on offers and acceptance thresholds. (3) The intercept and the slope differ considerably among subjects, so that it is plausible that our population is heterogeneous and consists of people pursuing different norms. This can be seen from the large standard deviation of the intercept and the responder s effort and from the large negative correlation between the intercept and the slope of the responder s effort. 20

21 successful. In addition to the proposers and responders intercepts, the relative contribution to the common pool significantly affects both, the offer and the acceptance threshold: The slope of the responders efforts shows that the offers of the proposers increase by 0.29 units and the acceptance thresholds of the responders by 0.15 units for each effort unit of the responders. The empirical relevance of the second normative content, therefore, builds a basis for heterogeneity in behavioral expectations. Third, the random effects suggest that our subjects are heterogenous in their level of normative commitments and fourth that they are heterogenous in their beliefs about the validity of alternative norms. The random effects parts suggest that the players can be categorized into distinct behavioral types. This argument is supported by three aspects. We have (i) large standard deviations in the intercepts, (ii) large standard deviations in the slopes and (iii) strong negative correlations between intercepts and slopes. 12 Based on these facts, we hypothesize the existence of three distinct types of normative behavior in our population: Equality players with high intercepts and low slopes, equity players with the inversed pattern of low intercepts and high slopes and egoistic players with low intercepts and low slopes. Moreover, the strong negative correlations between intercepts and slopes suggest that equality and equity players are more common than egoistic players. In the following, we investigate the distinct types of normative behavior in more detail by exploring graphically all data points for the bivariate relation between the effort and the offer, or the effort and the acceptance threshold. Figure 4 illustrates the proposers offers (top) and responders acceptance thresholds (bottom) for given responders effort. In the top figure, it can be seen that most offers cluster around the two lines representing the equity norm and the equality norm, respectively. We can regard the pure equity and equality norm as attractors or focal points in Schelling s sense. But not all proposers adhere to pure equity or equality norms. There are 12 For understanding that standard deviations in the random part are large, we can compute the respective confidence intervals. E.g. the (average) intercept ± the z value of 0.97 times the standard deviation of the intercept returns the interval, in which 3/4 of the population can be found, e.g. 3/4 of offers are in between 10 % and 52 % if the respondent contributed nothing. We refer to Snijders and Bosker (1999) for a detailed discussion of interpreting the random part of a multilevel model. Multilevel models usually assume homogeneity of errors. However, neither our theory nor our data justify this assumption. However, Verbeke and Lesaffre (1996) can show that estimators are still reliable if this assumption is relaxed. 21

22 also a few hyper-fair offers (in the upper left corner). A large fraction of offers is located between the equity line and the equality line (lower left and upper right corner), or even below both lines (lower right corner). Thus, some proposers are biased by self-interest. Although some proposers enlarge their prospective share by giving moderately low offers, only a few play the subgame-perfect equilibrium solution and offer the minimal amount; but still, egoistic players do exist. Among the responders, there are surprisingly many players who adhere to the equality norm and who are willing to punish offers below 50 % with rejection. This is an unusual finding, as many previous studies report that offers of 40 % and above are commonly accepted. 13 In contrast to the results for the proposers, the attraction of the pure equity and equality norm is not as strong. As a result, the variance in the responders decisions is much higher. In addition, we find a second local maximum: From the histogram on the left it can be seen that 20 % of the acceptance thresholds are below 20 %. The respective players are willing to accept very low offers, sometimes even if they contributed much more than their proposer. We also find these patterns in the responders scatter plot of figure 4, showing a third cluster on the very bottom. 5.2 Distinct normative types on the micro-level How can we understand the structures on the macro-level by micro-level behavior? We continue our evaluation of the necessary conditions of normative conflict by examining the individual decisions and find surprisingly clear patterns. Figure 5 depicts four characteristic proposer profiles and figure 6 four characteristic responder profiles. Player 2 (upper left) represents a pure equality player: In the role of the proposer (figure 5) and in the role of the responder (figure 6), player 2 offers and claims always 50 % of the pool, regardless of the differences in effort between proposer and responder. In the upper right part of figure 5 and 6, player 3 shows pure equity behavior. Player 3 offers always exactly as much as the responder s effort and always demands at least her effort as a responder. As a third type, we find players who play the subgame perfect Nash equilibrium. Player 13 (lower left) constantly offers and accepts the smallest possible positive unit of This result might also be influenced by the strategy method. There are mixed results for the comparability of the strategy method and the direct response or game method (see Charness and Levine (2007) for an overview). 22

23 proposer s offer equity equality fraction responder s relative effort responder s threshold equality equity fraction responder s relative effort Figure 4. Proposer s offers as a function of the responder s effort (top) and responder s acceptance thresholds as a function of the responder s effort (bottom). The histograms on the left show the distribution of the relative height of offers and acceptance thresholds. The axis labels depict percentages. The distinct normative types are illustrated by additional lines for equality and equity norms. Most offers cluster around these lines, while there is considerable noise, too. Responders cluster less around the equity and equality lines. Players show more risk-averse behavior if they are in the role of a proposer, forming a third cluster with very low acceptance thresholds. 23

24 proposer s offer Player 2 equality equity Player egoistic population population Player Player 20 cherry picker responder s relative effort population population Figure 5. Offers of selected players. The larger figures depict individually observed strategies. We find 54% (n=50) equality proposers (upper left), 39% (n=36) equity proposers (upper right), 3% (n=3) egoistic proposers (lower left) and 3% (n=3) cherry picker proposers (lower right). The insets depict the superposition of all individual decisions in the population classified as belonging to the corresponding type. Euro. Consequently, we call this player type the egoist. We found even another type, showing an interesting hybrid behavior between self-interest and norm-compliance. This fourth type plays according to the equity norm as long as she is a relatively high achiever, and switches to the equality norm if she is a relatively low achiever. See player 20 as a proposer and player 36 as a responder of this type on the lower right of figure 5 and 6. We call this players cherry pickers, as they seem to adhere to norms, but pick the particular norm, which serves their self-interest best. Most of the players in our experiment followed consistently the pure characteristics of equity, equality, egoist or cherry picker, illustrated by the individual cases in figure 5 and 6. The player types have been categorized as 24