Optimal Contracting With Endogenous Social Norms

Research in sociology and ethics suggests that individuals adhere to social norms of behavior established by their peers. Within an agency framework, we model endogenous social norms by assuming that each agent’s cost of implementing an action depends on the social norm for that action, defined to be the average level of that action chosen by the agent’s peer group. We show how endogenous social norms alter the effectiveness of monetary incentives, determine whether it is optimal to group agents in a single or two separate organizations, and may give rise to a costly adverse selection problem when agents' sensitivities to social norms are unobservable. Disciplines Accounting | Economics | Social and Behavioral Sciences This journal article is available at ScholarlyCommons: http://repository.upenn.edu/accounting_papers/25 1459 American Economic Review 2008, 98:4, 1459–1475 http://www.aeaweb.org/articles.php?doi=10.1257/aer.98.4.1459 Research in social psychology suggests that social norms arise and influence individual behavior because individuals have innate preferences to conform to the behavior of their peers. When one’s peers are other members of the same organization, such as a firm or a profession, this research suggests that the behavior of individuals in an organization determines the social norms for the organization which, in turn, influence individual behavior. It follows from this interplay between individual behavior and social norms that social norms are endogenous and ultimately determined by exogenous design choices, such as the organization’s incentive contracts, boundaries, and membership. To assess how social norms and organizational design interact, we analyze a principal-agent model in which a principal contracts with a continuum of agents. In our main model, each agent influences the performance measure that determines his compensation through two action choices, called the desirable action choice and the undesirable action choice. The desirable action yields some benefit to the principal, while the undesirable action imposes a cost on the principal. For example, the desirable action may represent the agent’s productive effort and the undesirable action may represent detrimental earnings management activities undertaken by the agent. We In the social psychology literature, Lawrence Kohlberg ( 984) develops a framework of moral development in which individuals prefer to conform to norms established by their peers. Subsequent studies have developed psychometric instruments that measure this preference and conclude that the majority of individuals within professional organizations of adults prefer to conform, although the strength of this preference varies from individual to individual. A related construct in psychology is the locus of control developed by Julian B. Rotter ( 966). An extensive accounting literature considers accounting (e.g., bookkeeping) manipulation. See, for example, Ronald A. Dye ( 988), Paul E. Fischer and Robert E. Verrecchia ( 000), or the survey of empirical evidence discussed in Thomas D. Fields, Thomas Z. Lys, and Linda Vincent ( 00 ). In addition, there is extensive evidence that managers make real decisions (e.g., channel stuffing or foregoing research and development expenditures) to manipulate nearterm performance measures. See, for example, Jeremy C. Stein ( 989) or David Hirshleifer ( 993) or, for some empirical evidence, Paul E. Oyer ( 998) and Merle Erickson, Michelle Hanlon, and Edward L. Maydew ( 004). Optimal Contracting with Endogenous Social Norms By Paul Fischer and Steven Huddart* Research in sociology and ethics suggests that individuals adhere to social norms of behavior established by their peers. Within an agency framework, we model endogenous social norms by assuming that each agent’s cost of implementing an action depends on the social norm for that action, defined to be the average level of that action chosen by the agent’s peer group. We show how endogenous social norms alter the effectiveness of monetary incentives, determine whether it is optimal to group agents in a single or two separate organizations, and may give rise to a costly adverse selection problem when agents’ sensitivities to social norms are unobservable. 1JEL D 3, D8 , D86, Z 32 * Fischer: Smeal College of Business, Pennsylvania State University, University Park, PA 680 –3603 (e-mail: [email protected]); Huddart: Smeal College of Business, Pennsylvania State University, University Park, PA 680 3603 (e-mail: [email protected]). We thank Kalyan Chatterjee, Michel Clement, Shane Dikolli, Mark Dirsmith, Don Hambrick, Michael Kirschenheiter, Richard Sansing, Steven Schwartz, Shiva Sivaramakrishnan, Linda Treviño, Amir Ziv, two anonymous reviewers, and seminar participants at the 004 meetings of the American Accounting Association, Carnegie Mellon University, Columbia University, Duke University, the Ohio State University, SUNY-Binghamton, the University of Alberta, and the University of Houston for helpful discussions. We gratefully acknowledge financial support from the Smeal Competitive Research Fund. SEptEmBER 2008 1460 tHE AmERICAN ECONOmIC REVIEW assume a separate norm for each action, where the norm for an action affects the cost of that action to every agent (e.g., by inducing feelings of satisfaction or guilt) and thereby influences every agent’s action choice. To reflect behaviors described in the sociology literature, the social norm ultimately causes each agent to choose a higher action level when his peers choose a higher action level. We incorporate both desirable and undesirable actions, and a corresponding norm for each, in our analysis because both types of actions and norms appear to be economically relevant. In the business realm, for example, shortcomings in corporate culture have been attributed not simply to the demise of social norms encouraging desirable actions, such as diligent effort, but also to the erosion of social norms that thwart undesirable actions, especially the manipulation of financial reports. Furthermore, outside the business realm, recent empirical studies provide evidence that norms for both types of actions affect individual choices. For example, Raymond Fisman and Edward Miguel ( 006) study the differing propensities of Nigerian and Norwegian diplomats posted to New York City to accumulate unpaid parking tickets—an undesirable action—and conclude that social norms related to corruption are significant and persistent because diplomats behave like others in their home countries. As another example, E. Han Kim, Adair Morse, and Luigi Zingales ( 006) report that academics’ research productivity—a desirable action—is influenced by the cultural norm of the department that houses them. Our analysis suggests that norms for desirable and undesirable actions have different implications for organization design. For example, the presence of a norm for desirable action multiplies the benefits of financial incentives, while the presence of a norm for undesirable action reduces the benefits. In addition, agents’ relative sensitivities to each type of norm determines whether it is beneficial to split an organization apart to foster different norms in each part. As a final example, unobservable individual differences in sensitivity to social norms create a costly adverse selection problem when the differences pertain to social norms for undesirable actions, but not when the differences pertain to social norms for desirable actions. Other papers have considered how nonmonetary incentives, such as social norms, have consequences for the returns to financial incentives. In a setting with a single agent, Roland Bénabou and Jean Tirole ( 003) formalize the notion that monetary incentives can crowd out incentives provided by intrinsic factors. Relatedly, experimental and field study work in a nonbusiness setting by Uri Gneezy and Aldo Rustichini ( 000a, b) supports the notion that intrinsic factors can undermine the effectiveness of financial incentives. They find that introducing financial incentives in a nonbusiness setting can lead to less of a desirable action (correct answers to test questions) and more of an undesirable action (the tardy collection of children from a day care center), respectively. Their evidence of important interactions between extrinsic (i.e., monetary) and intrinsic (e.g., a social norm) incentives suggests that norms, in addition to standard compensation mechanisms, merit study. In an agency study of team production incorporating a norm, Steffen Huck, Dorothea Kübler, and Jörgen Weibull ( 006) show that strong monetary incentives tied to aggregate team production can rule out Pareto-preferred equilibria that are attainable if incentives are weak and norms are strong. In another agency study, Dirk Sliwka ( 007) shows that firms may forego highpowered incentive contracts because they attract agents who respond solely to the incentives 3 Models of reciprocation have interdependent preferences, which is a feature analogous to our social norm. Ernst Fehr and Armin Falk ( 00 ) distinguish between two human social desires that interact with economic incentives: the desire for social approval and the desire to reciprocate—see, e.g., Gary Bolton and Axel Ockenfels ( 000) and Gary Charness and Matthew Rabin ( 00 ), who examine how notions of equity and fairness influence behavior. These models do not address social norms with respect to unobservable acts, which is the focus of our study. 4 See, for example, the comments of Arthur Levitt ( 998), who is a former Securities and Exchange Commission Chairman. VOL. 98 NO. 4 1461 fISCHER ANd HuddARt: CONtRACtINg ANd SOCIAL NORmS which, in turn, undermines the behavior of agents employed by that firm who conform to the behavior of others. These agency studies do not consider how norms determine organizational boundaries or membership. Because we use a two-action framework, this study relates to the multitask agency literature initiated by Bengt Holmström and Paul Milgrom ( 99 ). The model considered here differs from the models in that literature because (a) social norms induce interdependence in the agents’ utility functions and (b) the undesirable action is detrimental to the principal even though it favorably influences the performance measure. Because of this latter feature, our study relates to work by Canice Prendergast ( 999), who analyzes how counterproductive influence activities can improve subjective evaluation measures. Finally, some tax and welfare policy research also considers the role of norms. The tax compliance studies of Jon S. Davis, Gary Hecht, and Jon D. Perkins ( 003) and Joel Slemrod ( 004) suggest that social norms are important to tax compliance. In a related vein, N. Soren Blomquist ( 993) and Assar Lindbeck, Sten Nyberg, and Weibull ( 999) consider the effects of tax and welfare policy, respectively, on labor supply when norms influence behavior. The paper proceeds as follows. Section I presents a single (desirable) action model with a single social norm, which introduces and motivates the intuition for the two-action two-norm model presented in Section II. Section III characterizes the cost of implementing a particular set of desirable actions, which is necessary for the analyses in Sections IV and V. Section IV considers how changing organization boundaries, which alters the social norms faced by agents, affects the cost of implementing a set of actions. Section V identifies the behavioral traits that are preferred within particular organizations and considers whether contracts exist that attract only the agents with the desired traits. Section VI concludes. I. Single Action Model Consider an agency model in which a principal employs a continuum of risk-neutral agents, each of whom must undertake a single task. The agents are indexed by a compact set I having strictly positive measure. The agents have a common reservation level of expected utility, v. After contracts are signed, each agent i [ I chooses a level of some desirable action, ai $ 0, which the principal cannot observe. Contracts are linear, so that agent i’s compensation is wi 1 bi ri , where wi [ R and bi . 0 are the contract parameters, and ri is a stochastic report that determines agent i’s compensation. The mean of ri is h 1ai 2 , where h9 . 0, h0 # 0, h9 1x 2 approaches infinity as x approaches 0; and h9 1x 2 approaches 0 as x approaches infinity. Attention is restricted to linear contracts because these contracts make intuitive the interplay between the power of incentives, represented by the bi’s, and social norms. Furthermore, linear contracts are efficient in this setting. Given contract 5wi, bi6, agent i chooses ai to maximize ( ) z 1ai 2 K wi 1 bi h 1ai 2 2 f 1ai 2 Nai 2 , where f is a cost function with continuous derivatives defined over the real line, and f 9 . 0 and f 0 . 0. Finally, agent i’s cost, f 1ai 2 Nai 2 , is influenced by a norm parameter, Nai , which affects Restricting attention to cases where bi is strictly positive, which is necessary and sufficient for agent i to choose an interior level of action, simplifies the proofs without affecting the formal results. 6 If the density function for ri as a function of ai can be written in the form g 1ri, ai 2 , then restricting attention to linear contracts is without loss of generality because linear contracts are as efficient as any other contract that induces any interior set of desirable actions by the agents of an organization. SEptEmBER 2008 1462 tHE AmERICAN ECONOmIC REVIEW i’s total and marginal costs of ai. Formally, a higher value for the norm reduces the marginal cost of ai because d f 1ai 2 Nai 2 /dai dNai 5 2f 0 , 0. Including the norm Nai in the cost function for the action captures the idea that, in addition to any physical or cognitive cost, action choices have an emotional cost that is determined in part by norms of behavior. For example, an individual may have a personal ethic that supports hard work, which we term a personal norm. In addition, an individual may be part of an organization with a cultural ethic that supports hard work with feelings of satisfaction, which we term a social norm. A higher norm of either type inclines an individual to work harder for a given level of financial incentives. Accordingly, we assume that individual i’s norm, Nai, is a weighted average of agent i’s personal norm, Ai [ R, and a common social norm for the action, Sa, which is the per-capita average level of the actions of other agents in i’s organization. Formally, the norm for agent i in an organization I is ( ) Nai K 1 2 ai 2 Ai 1 ai Sa, where eI ai di (3) Sa K . eI di Parameter ai [ 30, a– 4 , where 0 # a– , , represents the extent to which agent i is influenced by the behavior of others in his organization through the social norm, and ai . 0 implies that i chooses a higher ai in response to an increase in the average action of his peers. The payoff that the principal derives from a given supply of the action is unspecified because our primary focus is on the minimum cost of implementing the desirable action. Nonetheless, the principal cares about the agents’ actions because they generate some valuable noncontractible payoff. The principal uses the report only as a source of information (i.e., the report has no intrinsic value). Further, analogous to models of atomistic agents where each agent is a price taker, observe that agents are “norm takers” because each agent’s action choice has no measurable effect on the social norm, Sa. Finally, each agent correctly anticipates the social norm, Sa, that actually prevails in his organization, which is analogous to a rational expectations assumption. A. post-Contracting Equilibrium A post-contracting equilibrium for contract parameters 5wi, bi6 i [ I is a set of actions, 5ai6 i [ I, such that for each i, ai maximizes objective ( ) given 5wi, bi6 and Sa , and Sa satisfies equation (3) given 5ai6 i [ I. To prove the existence and uniqueness of a post-contracting equilibrium, we first derive each agent’s optimal action choice given the agent’s contract and the social norm. The mathematical properties of agent i ’s objective imply that the first-order condition completely characterizes i ’s action choice: (4) z9 1ai 2 5 bi h9 1ai 2 2 f 9 1ai 2 1 2 ai 2 Ai 2 ai Sa 2 5 0. By applying the implicit function rule to this first-order condition, ai can be written as an implicit function of Sa, ai 1Sa 2 , where: 0ai 1Sa 2 ai f 0 1ai – 1 2 ai 2 Ai 2 ai Sa 2 2 ( ) 5 [ 30, ai 2 . 0Sa 2bih0 1ai 2 1 f 0 1ai 2 1 2 ai 2 Ai 2 ai Sa 2 7 Note that norms influence behavior by altering each agent’s cost of the desirable action. Norms also may influence agent behavior in ways we do not model. For example, an agent may bear a cost when affiliated with an organization with low social norms because he is treated poorly by his colleagues. We thank a referee for suggesting this point. VOL. 98 NO. 4 1463 fISCHER ANd HuddARt: CONtRACtINg ANd SOCIAL NORmS To complete the proof, we establish that there is a unique Sa such that eI ai 1Sa 2di (6) Sa 2 5 0. eI di The left-hand side of equation (6) evaluated at Sa 5 0 is strictly negative and increases monotonically and continuously in Sa at a rate greater than 2 a – . 0. It follows that there is a unique value of Sa, which is positive and finite, that satisfies equation (6). Hence, we have the following lemma. LEMMA : for any set of contracts, 5wi, bi6 i [ I for organization I, there exists a unique postcontracting equilibrium, 5ai6 i [ I. B. multiplier Effects To gain some insight into the role played by the social norm, we present comparative statics in a setting where all agents have identical contracts and preferences so that bi 5 b, Ai 5 A, and ai 5 a for all i [ I. As a result, ai 5 a for i [ I. Analysis of the first-order condition (4) leads to da 0z9 1a 2/0y dSa a f 0 1a 2 Na 2 (7) 5 1 , dy 2bh0 1a 2 1 1 2 a 2 f 0 1a 2 Na 2 dy 2bh0 1a 2 1 f 0 1a 2 Na 2 where y is b, A, or a and dSa/dy is the change in the post contracting equilibrium social norm implied by (6). The first term on the right-hand side of equation (7), which is proportional to 0z9 1a 2/0y, is the direct effect of the parameter change on agents’ actions. The second term, which is proportional to dSa/dy, captures the impact of the change in the norm attributable to the change in the exogenous parameter. Furthermore, because a 5 Sa, this indirect effect is of the same sign as the direct effect. Assuming agents are sensitive to the social norm 1 i.e., a . 02 , the indirect effect multiplies the direct effect. For example, if the direct effect causes each agent to work a bit harder, the additional hard work favorably affects the social norm which, in turn, causes each agent to work harder still. Formally, we have the following corollary. COROLLARY : Assume identical agents with identical contracts. the agents’ actions are increasing in the power of incentives, da/db . 0, and increasing in their personal norm, da/dA . 0. Whether the agents’ actions are increasing or decreasing in the agents’ sensitivities to the social norm depends on whether the social norm is higher or lower than the personal norm: da/da . 0 if Sa . A, and da/da , 0 if Sa , A. Consider, first, changes in the incentive parameter b. The direct effect is positive—greater incentives induce more of the desirable action—and this effect is reinforced by the concomitant increase in the social norm. An increase in the agents’ respective personal norms, A, also increases the desirable action. Again, because the increased personal norm translates into more desirable action, the social norm also increases, which induces yet higher effort. Finally, consider changes in the sensitivity to the social norm. If the social norm, Sa, is higher (lower) than the personal norm for each agent, A, an increase in agents’ sensitivity to the social norm increases (decreases) the total norm, Na, which provides a direct incentive to take a higher action, a. This direct effect is multiplied because the social norm also increases (decreases), which induces even more (less) desirable action. In summary, in the single-action setting, the social norm multiplies the effect on a of any change in the exogenous parameters. SEptEmBER 2008 1464 tHE AmERICAN ECONOmIC REVIEW II. Desirable and Undesirable Action Choices with Two Norms Norms influence desirable actions, but they may also influence undesirable actions. For example, as discussed in the introduction, employees can engage in costly earnings management activities that improve reported performance but ultimately reduce firm value. To introduce norms that influence undesirable actions, we extend the base model to include a second action choice with its own norm. In the main model, each agent i chooses two unobservable actions that affect the principal’s welfare, a desirable action, ai $ 0, and an undesirable action, ui $ 0. The undesirable action, ui, imposes a cost on the principal of ki ui where ki . 0. Agent i’s compensation is again restricted to be linear in a report, ri; however, the mean of ri is now h 1ai 1 ui 2 . As a consequence, the report does not allow the principal to distinguish ai from ui. In addition, the agent’s objective includes a cost associated with the undesirable action: (8) z 1ai, ui 2 K wi 1 bih 1ai 1 ui 2 2 f 1ai 2 Nai 2 2 f 1ui 1 Nui 2 . Consistent with the notion that a higher level of the norm for the undesirable action should discourage the agent from taking that action, a higher value for the norm increases the marginal cost of ui 1 i.e., 0 f 1ui 1 Nui 2/0ui0Nui 5 f 0 . 02 . The norm associated with the undesirable action is defined in a manner analogous to the norm for the desirable action: (9) Nui K 1 2 mi 2 ui 1 mi Su, where eI ui di ( 0) Su K 2 eI di is the social norm for the undesirable action, ui is agent i ’s personal norm for the undesirable action, and mi [ 30, m– 4 , where 0 # m– , , captures the extent to which agent i is influenced by the undesirable behavior of others in the organization. In contrast to the desirable action social norm specification in (3), Su is defined so that the undesirable action norm for an agent is reduced when other agents engage in more of that action. This assumption, coupled with the assumption that a higher norm for undesirable action increases an agent’s personal cost of that action, implies that an agent engages in more of the undesirable action when others in the organization engage in more of that action. A. post-Contracting Equilibrium A post-contracting equilibrium for an organization of agents I, who have contracts 5wi, bi6 i [ I, is defined as a set of actions, 5ai, ui6 i [ I, such that: (a) each agent’s action choices maximize objective (8) given 5wi, bi6 i [ I; and (b) Sa and Su satisfy equations (3) and ( 0) given 5ai, ui6 i [ I. As in the single-action model, the proof that there exists a unique post-contracting equilibrium begins by characterizing each agent’s action choices as a function of the contract parameters and the social norms. The proof is completed by demonstrating that there exists a unique pair of social norms, Sa and Su, that satisfy (3) and ( 0). This and all subsequent proofs are in a Web Appendix (available at http://www.aeaweb.og/articles.php?doi= 0. 7/aer.98.4. 4 9). LEMMA : for any set of contracts, 5wi, bi6 i [ I for organization I, there exists a unique postcontracting equilibrium, 5ai, ui6 i [ I. If all agent choices are interior, then ai 2 Nai 5 ui 1 Nui for all i. The observation that ai 2 Nai 5 ui 1 Nui when the agents’ choices are interior implies that the action mix selected by agent i shifts toward the desirable action as either his norm for the VOL. 98 NO. 4 1465 fISCHER ANd HuddARt: CONtRACtINg ANd SOCIAL NORmS desirable action, Nai , increases or his norm for the undesirable action, Nui , increases. A higher norm for the desirable action reduces the marginal cost of that action to the agent. A higher norm for the undesirable action increases the marginal cost of that action, which induces the agent to supply more of the desirable action. B. multiplier Effects In the single action setting, we demonstrated that the social norm for the desirable action multiplies the impact of changes in exogenous parameters on the desirable action choice. With two actions and a social norm for each, there are two multiplier effects, which can have offsetting effects on an agent’s desirable action choice. To illustrate, we again employ a setting with identical contracts and agents: bi 5 b, Ai 5 A, ui 5 u, ai 5 a, and mi 5 m for i [ I. As a result, ai 5 a and ui 5 u for i [ I. Throughout, interior choices for both actions are assumed. Consider the impact of an increase in the power of incentives, b. Holding constant the social norms, the direct effect of the increase in b on the desirable and undesirable action choice leads to an increase in both actions. The increase in the desirable action choice causes an increase in the desirable action social norm, which yields the same multiplier effect as in the single action setting by making the desirable action less costly to each agent. The increase in the undesirable action choice causes a decrease in the undesirable action social norm, which makes the undesirable action less costly as well. If the two multipliers are unequal, the action mix shifts toward the action choice with the larger multiplier. Hence, if the multiplier on the undesirable action is larger, which occurs when m is large, the impact of the change in b on the desirable action is mitigated by the social norm for the undesirable action. It turns out that this effect is never large enough to dominate the direct effect, although the effect of an increase in b on the desirable action approaches zero as the agents’ sensitivity to the social norm for the undesirable action, m, approaches its upper bound, one. When the personal norm for either the desirable or the undesirable action increases, the direct effect of the change causes the action mix to shift toward the desirable action. The change in the mix causes both social norms to increase, implying that each social norm creates a positive multiplier effect on the desirable action choices. If the social norm is higher than the personal norm 1e.g., Sa . A or Su . u2 , then an increase in the sensitivity to either social norm causes the desirable action to increase. If the social norm is lower than the personal norm, then an increase in the sensitivity to either social norm causes the undesirable action to decrease. The following corollary summarizes these observations. COROLLARY : Assume identical agents with identical contracts in the setting where the agents choose desirable and undesirable actions, and these actions are interior. the desirable and undesirable action choices for each agent are increasing in the power of incentives: da/db . 0 and du/db . 0. the desirable action choice is increasing and the undesirable action choice is decreasing in either personal norm: da/dA . 0, da/du . 0, du/dA , 0, and du/du , 0. Whether the desirable and undesirable actions are increasing or decreasing in the agents’ sensitivities to the social norms depends on whether the social norm is higher or lower than the personal norm: (i) da/da . 0 and du/da , 0 if Sa . A, (ii) da/da , 0 and du/da . 0 if Sa, A, (iii) da/dm . 0 and du/dm , 0 if Su . u, and (iv) da/dm , 0 and du/dm . 0 if Su , u. III. Per-Capita Implementation Costs The post-contracting equilibrium allows us to determine the minimum cost to implement a specific set of desirable actions. The cost functions prove useful in analyzing how organizations SEptEmBER 2008 1466 tHE AmERICAN ECONOmIC REVIEW can be structured to exploit endogenous social norms and how agents with differing behavioral traits (e.g., one’s sensitivity to a social norm) can be allocated to organizations so as to minimize implementation costs. We develop the case of the desirable and undesirable actions. The results in the single-action setting are easily inferred from the more general setting. An optimal contract set for implementing 5ai6 i [ I is a 5wi, bi6 i [ I that minimizes the sum of the per-capita expected compensation and undesirable action costs, eI wi 1 bi h 1ai 1 ui 2 1 ki ui di , eI di subject to each agent attaining his reservation level of expected utility, z 1ai , ui 2 $ v; ai and ui maximize z 1ai , ui 2 for each i; and, the endogenous social norms satisfy conditions (3) and ( 0). The reservation constraint for each agent i is always binding, so the principal’s objective can be rewritten as a function of the desirable actions, 5ai6 i [ I, and the induced set of undesirable actions, 5ui6 i [ I: eI 3v 1 f 1ai 2 Nai 2 1 f 1ui 1 Nui 2 1 ki ui 4 di ( ) . eI di The restated objective function ( ) identifies the components of the cost to the principal of inducing a given level of the desirable action. The desirable action cost to the principal is simply the cost to the agent, f 1ai 2 Nai 2 . In contrast, the undesirable action imposes costs on the principal from two sources. First, there is the cost that arises from putting the agents into a situation where they are induced to take undesirable actions, f 1ui 1 Nui 2 . This cost of undertaking ui is borne directly by agent i and indirectly by the principal, who pays agent i’s wages. Second, there is the direct cost of the undesirable action to the principal, ki ui. If both ai and ui are strictly positive for i [ I, which we assume for the remainder of the paper, then the principal’s problem can be exploited to characterize the minimum cost of implementing 5ai6 i[I as a function of those actions, as well as the exogenous variables. In particular, recall from Lemma that ai 2 Nai 5 ui 1 Nui for each agent. It follows that the cost function ( ) can be written eI v 1 f 1ai 2 Nai 2 1 ki 1ai 2 Nai 2 Nui 2 di ( ) , eI di where Nai 5 1 2 ai 2 Ai 1 ai Sa, Nui 5 1 2 mi 2ui 1 mi Su, Sa 5 eI ai di / eI di, and eI 1 2 ai 2 1Ai 2 Sa 2 1 1 2 mi 2 ui di ( 3) Su 5 . eI 1 2 mi 2 di Consider the relation between the set of desirable actions induced and each of the social norms. When financial incentives are increased, there is an interdependent set of effects on action choices and norms. First, the level of desirable action chosen by the agents increases. Second, the social norm for the desirable action increases, which follows directly from the definition of the social norm for the desirable action. Third, it follows from ( 3) that the social norm for the undesirable action decreases when the social norm for the desirable action increases. This VOL. 98 NO. 4 1467 fISCHER ANd HuddARt: CONtRACtINg ANd SOCIAL NORmS relation arises because more undesirable action is induced by the increased financial incentives. The increase in undesirable action, in turn, erodes the social norm that discourages the undesirable action. Hence, motivating the agents to increase the level of the desirable action necessarily leads them to supply more of the undesirable action and also erodes the social norm thwarting the undesirable action. IV. Organizational Boundaries Economic entities often include individuals who specialize in different productive activities (e.g., auditing and consulting, or sales and service). If the returns to desirable actions across productive actions differ, it is preferable to induce higher levels of desirable action from some individuals than from others. Section III established that the cost of implementing a level of desirable action from an individual is a function of the social norm faced by that individual which is, in turn, a function of the behavior of others in the same organization. By managing the boundaries of each agent’s organization, then, it is possible to alter the implementation cost of a given set of desirable actions. In this section, a simple example highlights factors that may induce organizations to combine or subdivide so as to alter social norms and thereby lower implementation costs. Assume a set of identical agents indexed on 30, 4 are to be employed. All agents have the same behavioral traits: Ai 5 A, ai 5 a, ui 5 u, and mi 5 m for i [ 30, 4 . A high level of the desirable action, ah, is required for a proportion d [ 10, 2 of the agents, and a lower level, al , ah, is required from the remaining agents. The former opportunity might be thought of as a new line of business with growth prospects and the latter as a mature line of business. Throughout our analysis, we assume that these actions are independent of organizational structure. Hence, our analysis is concerned with identifying the least costly structure for a given set of desirable actions. To demonstrate how organizational boundaries matter, consider two possible cases. In the first case, all agents work in the same organization, so the social norm is computed over all agents. In the second case, there are two organizations, each with its own social norm. One organization implements al from each agent and employs proportion 2 d of the agents. The other organization implements ah from each agent and employs the remaining agents. Two sources of cost must be considered when comparing the costs associated with single and separate organizations: the costs of compensating the agents, and the costs of undesirable actions incurred by the principal. We assume that these latter costs increase linearly in the per capita level of the undesirable action. Parameters kh and kl denote the unit costs to the principal of the undesirable actions by the agents producing ah and al, respectively. We assume these costs, too, are independent of organizational structure. Given the specific additional assumptions in this section, the single-organization implementation cost is ( 4) C 5 v 1 d 3 f 1ah 2 Nâ2 1 f 1û h 1 Nû2 1 kh û h 4 1 1 2 d 2 3 f 1al 2 Nâ2 1 f 1û l 1 Nû2 1 kl û l 4 , where û j denotes the undesirable action of the agents who implement aj when there is a single organization, and Nâ 5 1 2 a 2 A 1 a 3dah 1 1 2 d 2 al 4 and Nû 5 1 2 m 2 u 2 m 3dû h 1 1 2 d 2 û l 4 8 One approach to creating organization boundaries is to create separate firms. A more subtle approach that may introduce boundaries within a firm involves putting physical or social distance between employees engaging in different tasks. SEptEmBER 2008 1468 tHE AmERICAN ECONOmIC REVIEW are, respectively, the norms associated with the desirable and undesirable actions in the single organization. The two-organization implementation cost is ( ) C 5 v 1 d 3 f 1ah 2 Nah 2 1 f 1uh 1 Nuh 2 1 kh u h 4 1 1 2 d 2 3 f 1al 2 Nal 2 1 f 1ul 1 Nul 2 1 kl u l 4 , where uj denotes the undesirable action of the agents who implement aj when there are two organizations, and Nah 5 1 2 a 2 A 1 aah, Nuh 5 1 2 m 2 u2 muh, Nal 5 1 2 a 2 A 1 aal, and Nul 5 1 2 m 2u 2 mul are the norms on desirable and undesirable actions associated with the implementing ah and al, respectively, in separate organizations.