5.1 Pay and selection on dedication

To set the stage, we begin with an extremely simple model in which increasing pay induces adverse selection along an important dimension—dedication to the job. Our case considers a group of workers who are qualified for a particular occupation that for many is a “calling” or “vocation.” There are many examples of this sort of dedication: religious ministry, policy advocacy, nursing, early childhood education, public-interest law, etc. The “calling” in this case is a potentially important form of intrinsic motivation. In this section we take a vocational inclination to be an unobserved feature of preferences. In subsequent sections, however, we allow for the possibility that the intensity of an agent’s “calling” might be reinforced or eroded by the behavior of the principal or peers.

Specifically, we consider the model set up by Heyes (2005) and analyzed further by one of us (Taylor, 2007). The analysis, which uses the market for nurses as the focus of discussion, begins with simple behavioral assumptions. There are qualified nurses, each of whom falls into two categories: (i) A proportion of these nurses, , view nursing as simply a job. These individuals receive utility equal to their wage, , and they produce value on the job. (ii) The remaining proportion, , is comprised of nurses who view work as a “vocation.” They provide higher-quality nursing along an unobservable dimension, . They also find their work fulfilling, and thus earn money metric utility  beyond the earned wage .

Each individual has a reservation wage which is drawn from a log concave p.d.f., , that has a corresponding c.d.f., . The function is assumed to be the same for both types of worker. Thus, at wage , the quantity of nursing labor supplied is

     (35)

and the average quality is of nursing care is

     (36)

where is the proportion of employed nurses for whom nursing is a vocation, i.e.,

     (37)

Heyes’ key insight is that this latter proportion is declining in the wage.⁵² Thus, the higher the wage, the lower will be the quality of services provided.

Consider an employer acting in isolation, e.g., a monopsonistic National Health Service (NHS), that hires nurses. (In different markets one could think of the Roman Catholic church setting wages for priests or nuns, or Habitat for Humanity setting wages for professional builders who take part-time positions constructing affordable housing.) Heyes shows that an employer who understands the adverse selection properties of high wages will set the wage to be lower than would otherwise be chosen. Thus, an NHS that maximizes surplus generated by nurses will operate with an apparent “shortage” of nursing, in the sense that the expected net value of product will be positive for the marginal nurse.

It is possible indeed that the principal will be driven to a corner solution, with pay set to zero. Thus, Habitat for Humanity has the well-known policy of using unpaid volunteers for many key tasks. Organization that seek to remedy injustice in the legal system often rely on pro bono attorney services. Historically, many religious workers take a “vow of poverty,” accepting compensation at near-subsistence levels. The idea, of course, is that the lower the pay, the higher will be the dedication level of individuals willing to adopt the vocation.

Taylor (2007) extends Heyes’ analysis to show that a monopsonist that seeks to maximize surplus generated by workers will always set the wage lower than the socially efficient level. The reason is that the monopsonist fails to take account of the surplus generated to those individuals who view their work as a vocation. Because wages are too low, too few vocationally-oriented workers end up in an occupation in which they create the greatest social value.

On the other hand, a parallel analysis in Taylor (2007) shows that if the labor market is perfectly competitive, the equilibrium wage will be inefficiently high. To see how this happens, notice that under the assumption that all workers must be paid the same wage, a social planner would want to maximize

     (38)

where is taken to be the value of the worker in some other capacity (with being the highest value in the distribution). Maximization of (38) leads to the wage being set to be the weighted sum,

     (39)

Next notice that in a competitive market the wage instead will equal average productivity,

     (40)

Using the property of log concavity given in Footnote 52, we can compare wages in (39) and (40), finding that . The problem with the competitive market is that each firm makes hiring decisions on the basis of average market productivity. A social planner would instead make decisions on the basis of the productivity of the marginal worker, i.e., would take account of the fact that as the wage increases in the market, the productivity of the marginal worker declines.

This simple model serves as a first illustration of an important point that reappears throughout this section of our essay: Pay policies can affect intrinsic motivation, often in surprising ways. Here, high pay reduces intrinsic motivation in a workforce in a particularly transparent way. Low-pay environments attract workers for whom the job is a vocation—workers who have an intrinsic inclination to provide high-quality service. The higher the pay, the greater will be the proportion of workers for whom the job is simply a job, i.e., workers who will provide lower-quality service.

A particularly interesting feature of this simple behavioral model is that markets can lead to wages being either too high or too low relative to an efficient benchmark, depending on the market’s structure. To see the logic of this point, consider this question: “If you were a falsely convicted death-row inmate, would you rather be in a State in which you must rely on an organization that reviews cases using pro bono attorney services, or in a State that purchases legal services on the competitive market?” In the State that relies on pro bono services, attorneys who work on death-penalty cases will be highly dedicated to justice, and will provide excellent legal aid, but that aid will be in short supply. In contrast, in a State that purchases legal services for death-row inmates, access to attorneys may be more extensive, but those attorneys will have a lower expected level of dedication. Our model shows that in the State that relies on pro bono attorneys, the wage is too low and the quality level too high relative to the efficient benchmark. But in the State that uses the competitive market, the wage is too high and the quality too low. Theory alone does not identify the socially preferred second-best outcome.

Recently, a number of papers have examined models in which agents differ in their level of intrinsic motivation. Delfgaauw and Dur (2007), for example, have a wage posting model in which a monopsonist faces the same tension discussed above: the higher the posted wage, the higher the probability of filling a vacancy, but the lower the expected motivation level of workers who apply. In their model, workers have private information about their utility—information which they may wish to signal to or conceal from an employer. Besley and Ghatak (2005) and Delfgaauw and Dur (2008) study public sector employment under the assumption that some agents have a “public service motivation” that takes the form of intrinsic value derived from making a contribution to one’s organization.⁵³

The model we have examined in this section omits, obviously, several relevant issues that merit further consideration. For instance, the set-up abstracts from the core problem of agency; workers are simply assumed to supply effort on the basis of their internal intrinsic values. Second, workers are assumed to be steadfast; motivation is not affected by the actions of those around them. Thus, a worker who is inclined to provide high quality service is not de-motivated when she is surrounded by others who provide low quality service. In short, the analysis abstracts from “social preferences” of the sort discussed in Section 4. Third, the model does not take account of the possibility that a worker’s motivation can be affected by attributions the agent might place on the intentions of the principal. It is easy to see that such attributions might be germane, though, in the case of the “service motivations” we have been discussing.⁵⁴ Finally, the set-up does not allow for the possibility that a worker’s intrinsic motivation ( in the model above) can be reinforced or undermined by pay policies. We turn next to models that take up such issues.

5.2 Social preferences, conformism, and the principal’s use of extrinsic rewards

Recent work by Sliwka (2007) considers the question of agency in a model that draws on a “social preference framework,” i.e., allows for agents’ motivation to be shaped in part on the behavior of those around them. As in the model set out in the last section, there is heterogeneity in worker motivation, and an agent’s motivational inclinations are initially hidden to the principal and to other agents. There are two types of “steadfast” agents. One type is “strictly selfish;” these are agents who care only about their own payoff. Agents of the second type are “fair” in the sense that they care about the wellbeing of others; specifically, for these agents, utility is increasing in the principal’s payoff. The key innovation is to assume that there is yet a third group, “conformists,” whose inclination toward fairness depends the values of those around them. To keep matters simple, Sliwka assumes that when a conformist learns what agent type is in the majority, the conformist behaves like the majority-type agent.

In this set-up, a principal who understands that most of his steadfast workers are fair, might be able to use compensation policies as a credible signal to “conformists.” In turn, if conformists believe that others around them are “fair” they behave like fair agents.

To see how this works, we set up a simple example similar to that developed by Sliwka. In our example, the principal first posts a policy that specifies wage as a function of effort (which is assumed to be observable ex post), . An agent’s best response to the announced policy depends, of course, on his preferences over effort and money, and those preferences in turn vary by type. In particular,

     (41)

where reflects a fair agent’s level of identification with the principal’s objective (with ).

We assume that the principal sets compensation to have a fixed component and a “bonus” that is a linear function of effort, . Given that types are unobserved, the principal’s posted wage-bonus policy applies to all agents. It is easy to see that for an announced compensation policy, best responses are

     (42)

With this in mind, consider a profit-maximizing principal who earns surplus

     (43)

for a given agent. Given the best responses in (42), it is clear that effort is increasing in the incentive intensity; for both types of steadfast agent. Sufficiently high effort-contingent bonuses would seem to be in order.⁵⁵ Remarkably, it might nonetheless be in the principal’s best interest instead to set equal to 0 and increase the baseline wage, i.e., to rely solely on low-powered incentives.

The key is the emergence of a separating equilibrium in which conformists become convinced that most steadfast agents are fair. It is assumed that the firm has private information about the proportion of steadfast agents who belong to each type, which for simplicity is taken to have a low value or a high value. It is a matter of simple algebra to confirm that there are parameter values for which the following holds: If the firm has a low number of fair agents, it pays a high bonus, , and a low wage. If the firm has a high number of fair agents, it pays no bonus, , and a relatively higher wage, i.e., it uses low-powered incentives. Here low-powered incentives serve as a credible signal, so conformists follow suit and behave like fair agents. This makes sense, because a principal who has a high fraction of steadfast fair agents will incur a smaller loss than a principal with a low fraction of fair agents when it sets the bonus to 0. A willingness by the principal to raise the fixed wage further strengthens the signal. Conformists, in response to this credible signal, behave like fair agents. Profit for this firm is higher than if it used higher-powered incentives.

The operating logic of the model is like the Taylor-Ritter model discussed in Section 3.1, in which the firm uses compensation policy to signal hidden information about itself (i.e., the firm’s financial fitness). In that model, low-powered incentives signal relatively good financial fitness, which allows the firm increased profitability.⁵⁶ Here, the principal’s hidden information is the mix of worker type. There is a cost to low-powered incentives, of course, as effort is set by all agents to be less than first-best. But the low-powered incentive persuades some workers—the “conformists”—to behave in an altruistic fashion, when they would otherwise have not.

Sliwka interprets his model as generating “trust as a signal of a social norm.” In Sliwka’s setting, the principal observes effort and can, if he chooses, condition rewards on effort. By setting no explicit incentives, the principal expresses trust in his workers. This trust directs a social norm. Some workers are more generous in their efforts than they would have been if they perceived a different norm.

The most intriguing possibility in the Sliwka’s model is that monetary incentives crowd out intrinsic motivation. If a firm moves from a “high trust” low-powered incentive scheme to a “low trust” high-powered incentive scheme, the firm shifts the norm and undermines the intrinsic portion of worker’s motivation (the “social component” in the utility of a worker who would otherwise behave as a “fair agent”). Sliwka develops his theory further by looking at employee self-selection into firms. Here again, low-powered incentives can serve to attract workers with high intrinsic motives (fair agents), which serves to reinforce the positive work norms that influence those who conform to others.

The key behavioral underpinning of the Sliwka model is the observation that many people seem to want to conform to those around them. As we have noted, there is considerable evidence about the key component of this story—that many people are influenced by norms. For example, some individuals feel bad about a particular action only in situations in which they think others would experience remorse for that same action.⁵⁷ In previous sections of this paper, we cited empirical work supportive of the social forces that create norms, e.g., studies by Ichino and Maggi (2000), in which worker absenteeism in an Italian bank was affected by the absenteeism of those around them, and Mas and Moretti (2009), in which effort by supermarket checkout workers was affected by other similar workers in their sightline. Yet another study, by Bandiera et al. (2009), shows that the productivity of farm workers is affected by the productivity of friends on the job. Jackson and Bruegmann (2009) document peer learning for teachers, which might be read as providing additional evidence on conformity to norms. Of course, considerably more empirical work will be required to know if conformism plays a sufficiently strong role to generate in real-world organizations the crowding out of intrinsic motivation predicted by the Sliwka model.

5.3 Extrinsic incentives when agents value the principal’s esteem

Ellingsen and Johannesson (2008) present a model that, like Sliwka’s, relies on social preferences. Also, like Sliwka’s, their model opens up the possibility that extrinsic incentives can undermine valuable intrinsic motivations.

The key innovation in the Ellingsen-Johannesson model is the postulate that human motivation is often rooted in social esteem—the desire to be well regarded by others. In this conception, an agent reasons as follows: “I wish for others to hold a high opinion of me. While I cannot know with certainty what others think of me, I do have beliefs about what others think, and these beliefs about others’ opinions are an important source of pleasure or discomfort.” The identity of the audience that the agent wishes to impress plays a key role in this model, and the agent might well have multiple audiences. For example, a college professor might care about opinions of her students, her dean, other professors in her department, and/or colleagues in the profession more generally. She desires the respect of the intended audience(s), i.e., gains utility if she believes that others think highly of her.

Ellingsen and Johannesson focus on the case in which the relevant audience to an agent is the principal. ⁵⁸ To simplify use of personal pronouns, we let the principal be male and the agent be female. In the model, then, the agent’s utility depends on the “respect” she earns from the principal, which is defined to be her beliefs about his beliefs about her.

A simple example shows how the desire to earn respect can affect an agent’s effort decisions.⁵⁹ Suppose there are two types of agent, “talented” and “untalented,” and type is not initially observable to the principal. An agent hired by a principal is paid an agreed-upon wage , and then chooses any effort level she likes, . We suppose that her utility is the sum of three components: (1) compensation , (2) the cost of effort, which is for a talented worker and for an untalented worker, with , and (3) “respect” of the principal, which has value , where is a positive constant and is the agent’s subjective probabilistic assessment of the principal’s belief that the agent is talented. In sum,

     (44)

Now, given (44), an untalented agent is clearly better off supplying effort 0 than supplying effort greater than , even if the higher effort level would “earn maximum respect” (i.e., would induce the principal to believe with probability 1 that the agent is talented). So we have a separating equilibrium satisfying the Intuitive Criterion if a talented agent supplies effort , which is just high enough so that the untalented agent will decline to mimic.⁶⁰

In this example, a talented agent provides positive effort. She is not motivated by her own material wellbeing, as she would be in a standard principal agent model. Nor is she motivated by an innate desire to see the principal’s wellbeing improve, as with the other-regarding preferences assumed in the Sliwka model (or other such models discussed in Section 4). She provides effort because by so doing she can be confident that the principal holds a high opinion of her. Put another way, she is motivated by social esteem—the desire to earn respect.

With this basic logic in place, we can set out the Ellingsen-Johannesson model of principal-agent interaction. The model is built around three components. First, agents and principals hold social preferences. They care about their own material wellbeing as well as well as the material wellbeing of others. Second, there is unobserved heterogeneity in the extent to which agents and principals value others’ wellbeing. In particular there are two types on people, who vary in the extent to which they are pro-social. Third, and most distinctively, both the agent and the principal are motivated by social esteem, so the agent cares about what the principal thinks about her, and the principal cares about what the agent thinks of him. Both want to be thought of as pro-social by the other. Moreover, the agent’s concern about the principal’s opinion is highest if she thinks highly of him, i.e., believes it is likely that he is highly pro-social. Similarly, the principal places greater weight on the agent’s opinion if he believes that she is highly pro-social.

With these assumptions in place, Ellingsen and Johannesson examine the equilibrium of a game in which the principal takes an initial action (e.g., makes a wage offer, or makes a decision about how much discretion to allow the agent in her work), and then the agent takes an action which affects both her material wellbeing and the principal’s wellbeing. As in the simpler example in the preceding paragraphs, there is a set of parameters on preferences and the distribution of types such that a separating equilibrium emerges that satisfies the Intuitive Criterion. In that equilibrium a pro-social principal can take a credible action that signals that he is pro-social, and, if she is sufficiently pro-social, the agent responds with an action that benefits the principal. The key driving behavioral force is that a pro-social agent wishes to be highly regarded by a pro-social principal. Having learned that the principal is pro-social, the agent takes a pro-social action herself as a means of securing the knowledge that the principal believes her to indeed be a pro-social individual.

Figure 2 provides a nice illustration of the behavior predicted in this model. The game presented is the two player sequential “trust game” of McCabe et al. (2003):

Figure 2 A principal agent game of trust.

In Case 1, the first player (the principal) can choose “not trust” (NT), which leads to payoffs (20, 20) for the principal and agent respectively, or “trust” (T), which accords discretion to the agent. If the principal plays T, the agent can reward the trust (R), giving payoffs (25, 25) or not (N), giving payoffs (15, 30). With conventional preferences, the subgame perfect equilibrium is clearly “not trust.” However, in the McCabe-Rigdon-Smith experiments, many principals chose T, and most agents responded by rewarding such trust by playing R.

In Case 2, the principal has no choice but to play T. In contrast to Case 1, here most agents responded by playing N. Thus, the intentionality of the principal’s trust appears to matter for the agent’s response.

The Ellingsen-Johannesson model provides a rationale for these observed outcomes in the trust game. Start with Case 1. There are parameters in the model such that two key conditions are met. First, the principal will play T only if he is sufficiently pro-social, i.e., only if he cares sufficiently about the wellbeing of the agent. Second, a pro-social agent, having received a credible signal that the principal is pro-social, cares sufficiently about the respect of the principal that she in turn takes the action R, confirming that she is pro-social.

In contrast, in Case 2 the principal has no opportunity to signal that he is pro-social. In turn, the agent cares less about his respect, and so she plays N.

Ellingsen and Johannesson (2008) show, using similar logic, that their model predicts behavior consistent with Falk and Kosfeld’s (2006) experimental evidence on the hidden “cost of control” in a principal agent game. In the Falk-Kosfeld game, agents are given an endowment of 120 and can transfer to the principal, who in turn receives payoff , thus resulting in payoffs . The important twist is that in some conditions of the game, there is a first stage in which the principal can play “control” by imposing a minimum transfer (e.g., a transfer of 10) from the agent to the principal, or can choose instead to “trust.” Agents motivated solely by material gain would always play the minimum available , and knowing this, the principal would always “control” to the maximum extent allowed. But, in fact, consistent with the Ellingsen-Johannesson set-up, many principals sent a signal of being pro-social themselves by choosing “trust” when they are allowed to do so, and in such cases many agents responded with larger values of than if the principal had played “control.”

One nice way to see the distinctive contribution of the “esteem model” is to view it in the context of Akerlof’s (1982) concept of gift exchange. Recall that in Akerlof’s model (discussed above in Section 4.2), an agent’s best response to a sufficiently generous “gift” by the principal is to reciprocate by providing high effort. The agent’s motivation to do so is captured in a clear, but stripped down fashion—with a utility function in which the agent experiences disutility only when her effort level exceeds a psychologically determined threshold (the effort “norm”). Ellingsen and Johnnesson take an additional step, positing an explicitly specified behavioral mechanism (“esteem”) that drives this motivation. This approach, the authors show, allows them to predict gift exchange behavior. But there are two advantages to the Ellingsen-Johannesson model:

First, the model gives a clear way of understanding the role of a principal’s intentions in shaping agents’ responses. This is important, given experimental evidence (such as (Charness, 2004)) that intentionality is important to understanding gift exchange.

Second, the model provides a rigorous way of approaching an important and under-appreciated aspect of principal agent problems as they apply in the workplace—the delegation of decision rights. Essentially, the delegation of consequential actions to agents plays the role of a “gift” here, and provides the agent with the opportunity to earn the respect of the principal. In contrast, highly intrusive job design diminishes intrinsic motivation.

These ideas are potentially valuable for understanding otherwise-inexplicable practices within organizations. For example, charitable organizations like Habitat for Humanity often rely on volunteers who receive little or no pay, and then delegate key decisions to these same individuals. By providing low pay, we have suggested (in Section 5.1 above), the organization is less likely to attract opportunists. Then the “esteem model” shows the important advantages to relinquishing bureaucratic control. A second example is the widespread use of “psychological contracts” (Rousseau, 1995) in which contracting parties find it advantageous to leave many elements unspecified, relying instead on mutual goodwill.

Conversely, the esteem model indicates why high pay and clearly delineated direction might be required in other circumstances. This can happen when the desire for esteem leads agents to have intrinsic motivation that works at cross purposes to the principal. For example, very strong financial incentives might be required to induce a physician to cut costs if the physician values the esteem of her patients more than the esteem of her boss (e.g., the managed care organization she works for).⁶¹

5.4 Extrinsic rewards and reputation

The Ellingsen-Johannesson model we have just discussed is one of a number of recent papers that focus on the interaction of pro-social motivation and reputation or esteem. This literature starts with the observation that people undertake altruistic and reciprocal actions (in the workplace and elsewhere), but recognizes that such behavior is often difficult to rationalize solely by other-regrading preferences. The degree to which people undertake pro-social behavior often depends on the social context and economic environment, and is driven in part by the desire to be highly regarded by others.

An important contribution to this literature is the recent work of Bénabou and Tirole (2006). Among the remarkable insights of this paper is a clear demonstration that extrinsic rewards can undermine intrinsic motivation when people care about reputation.

Consider the following anecdote: One of us has a particularly personable colleague who was asked by the dean to accept a somewhat onerous task—advising masters students—that would have high value to his colleagues. In exchange, the dean offered a $2000 bonus. The professor replied that it was certainly not worth taking on the task for $2000, but that he would be willing to do the job for free!

To demonstrate how the Bénabou and Tirole approach explains the behavior of this public-spirited professor, we set up a simplified case of their more general model. We suppose that a principal asks agents to undertake a pro-social activity by providing effort . The agent can decline, instead providing . The agent’s effort choice is observable by all, including members of an audience whose opinion matters to the agent.⁶²

The agent’s effort decision is assumed to affect his utility via four channels, which for simplicity are taken to be additive: (1) The agent is other-regarding, and so earns direct utility, , from providing effort ; (2) he stands to earn a material reward in the form of a bonus of , which provides utility (where is the marginal utility of money); (3) he faces an effort cost of ; and, most distinctively, (4) he stands to gain from the reputation-enhancing effect of his effort choice.

Central to the Bénabou-Tirole model are the following two assumptions, which drive development of “reputation.” First, people differ in the extent to which they have other-regarding inclinations, and in the extent to which they value money. So each individual’s set of preference parameters, and , is drawn from a known distribution.⁶³ An agent’s preference type can be thought of as his or her “identity.” Second, reputation is taken to be other’s views of one’s own identity. This reputation is increasing in the degree to which one is seen as having concern for others (having a high value of ) and decreasing in the degree to which he is seen as materialistic (having a high value of ).⁶⁴ Thus, reputation is taken to be , where and are weights that reflect the degree of image-consciousness (and are taken to be common knowledge constants here).

To summarize,

     (45)

So our agent provides effort if

     (46)

The three terms on the left-hand side of (46) are, respectively, the agent’s intrinsic, extrinsic, and reputational motivations. Effort is provided when the sum of these motivations exceeds the cost of providing effort.

What makes matters interesting here is that the reputation the agent earns depends on the level of the bonus chosen by the principal. To see this point, consider first the effort decision if the principal chooses , so that motivation is strictly intrinsic and reputational. Then the agent provides effort only if his “concern for others” exceeds the cut-off , where

     (47)

Such highly pro-social identity types lie to the right of the vertical “No Bonus” line in Fig. 3. Notice that in this situation the agent’s audience can draw an informative inference about the agent’s (i.e., can infer if the agent is in an other-regarding pool to the right of or narcissistic pool to the left of ) simply by observing the effort level, but can learn nothing about his materialism .

Figure 3 The effects of an extrinsic reward on the pool of agents providing effort.

Now suppose that the principal provides a bonus . For the moment, ignore any impact on reputation. The bonus has no effect on an agent with , of course, but for all other agents (those with ) the bonus is motivating. Agents now provide effort if they have an identity that lies to the right of the negatively sloped line marked “Positive Bonus , Unadjusted for Reputation.” Absent reputational effects, an extrinsic reward expands the pool of agents providing effort.

However, this is not the end of the story. Inspection of Fig. 3 makes it is clear that the average level of (concern for others) has declined in the pool of agents providing effort and also in the pool of agents not providing effort. It is similarly clear that the average level of (greed) has risen for the pool of those providing effort and declined for the pool not providing effort. So the overall effect on reputation is ambiguous. The most interesting possibility is that the effect of the bonus is to drag down overall reputation for those providing effort. This is demotivating. In Fig. 3 this is illustrated by the parallel shift to the right in the sloped line dividing those who provide effort and those who do not, with now giving the value of that separates the two pools for individuals along the horizontal axis (i.e., for individuals with ). On net, effects of the bonus are two-fold: Some identity types—those in Area A of Fig. 3—switch behavior to providing effort. Others—those in Area B—are induced to switch from effort provision to not providing effort. Overall, an extrinsic reward can increase or decrease effort, depending on the distribution of identity types.

The public-spirited professor in our anecdote is apparently an individual with an identity of the sort represented by Area B. Such a person has a relatively high concern for others and a relatively low level of greed. By offering a bonus for the task, the dean deprived the professor of the opportunity to demonstrate his public spiritedness.

In short, in the Bénabou-Tirole model, extrinsic rewards can spoil the reputational value pro-social action, thereby crowding out intrinsic motivation. The logic of the model is one of “signal extraction.” People take pro-social actions in part to signal one’s own identity to others. Extrinsic rewards, even very small extrinsic rewards, can serve to increase the noise-to-signal ratio of such actions.⁶⁵

One of the most interesting ideas in the Bénabou and Tirole model appears when the authors reinterpret the “reputational” terms in (46) to instead be the reinforcement of one’s own self image. The idea is described as follows:

Thus, the public-spirited professor might have found the offer of a $2000 bonus to be demotivating even if his colleagues were unaware of the bonus. Accepting the task without pay, in this conception, served to reinforce his identity; the professor can look at himself in the mirror and argue convincingly, “I must be a pro-social person. Otherwise I wouldn’t have taken on this task with no pay.”⁶⁶

The striking prediction that extrinsic rewards can crowd out desired behaviors does have empirical support. One widely cited example in economics is work by Gneezy and Rustichini (2000b) showing that the imposition of a monetary penalty for late child pick-up at a daycare center increased the likelihood of late pick-up. Bénabou and Tirole’s reputation/self-respect model strikes us as applicable here. When there was no explicit monetary penalty for on-time pick-up, parents were presumably motivated by genuine concern for daycare center workers and by a desire to project to others (or to oneself) character traits of responsibility or concern for others. The imposition of a monetary penalty of course increased the inclination for on-time pick-up among those parents with a materialistic orientation and a low level of concern for others. This very fact led parents more generally to no longer view on-time pick-up as a reliable signal of kind and responsible identity, and so reduced the strength of those motivating forces. Similar arguments apply to Gneezy and Rustichini’s (2000a) demonstration that extrinsic incentives reduced effort by school children collecting donations for a charitable organization.⁶⁷

Evidence of a potentially important form of crowding out is also found also in Frey and Oberholzer-Gee’s (1997) analysis of public reaction to the siting of a nuclear waste facility in one’s community. Their paper indicates that the provision of substantial compensation to residents of a host community reduces willingness to accept such a facility. The reputation/self-respect model might speak to these results, but it is quite possible that mechanisms described in Bénabou and Tirole (2003) are more germane. In that paper the authors set up a problem in which a principal seeks to motivate an agent to take a desired action in an environment in which the principal has better information than does the agent about some crucial aspect of the task—for example, the cost the agent will incur if she undertakes the task, the personal satisfaction she will experience if the task is completed successfully, or the likelihood that the agent will indeed be successful at the task. In this setting, the offer of a substantial monetary reward for some action can signal “bad news” to the agent about one of the elements of the action. Thus, if residents of a potential host community are asked to site a nuclear waste facility and are offered substantial compensation for doing so, that compensation might be seen as “bad news” about the eventual consequences of the facility.

The two papers we highlight in this section of our paper, Bénabou and Tirole (2003, 2006), are but two of a number of recent contributions in behavioral economics that might form solid building blocks for a new generation of behavioral principal agency models.⁶⁸ The challenge going forward is to place the psychological subtleties introduced in these new economic models into workable (and testable) theories of firm organization and labor markets. It is important to have carefully constructed, psychologically correct models in behavioral economics, but important also to work forward to understand the implications of these models for the allocation of resources in markets and in society broadly.⁶⁹

5.5 A concluding puzzle

The economic approach to agency places a primary emphasis on the use of material incentives (pay, promotion, etc.) as devices to resolve principal agent problems. The economic literature offers a rich and varied set of evidence in support of the critical efficacy and importance of material incentives. The theoretical literature reinforces these empirical findings. There are many situations in which firms eschew high-powered incentives, but for the most part this is the result of incentives having a powerful effect on behaviors. It is possible, as we have seen, to construct models where extrinsic rewards undermine intrinsic motives, but these models appear largely as elaborations and qualifications of the fundamental message: well designed extrinsic rewards are crucial to the resolution of fundamental and ubiquitous agency problems.

Things are quite different in the field of psychology. Here there has accumulated a vast amount of evidence that extrinsic rewards actually undermine intrinsic motives.⁷⁰ What explains the difference?

One important part of the explanation is a cross-disciplinary difference in the definition of intrinsic motives. Psychologists typically view as extrinsic any sort of action undertaken for instrumental reasons. Thus many of the pro-social and other regarding preferences we discuss in Sections 4 and 5 would be regarded as part of an extrinsic reward system in the psychology literature.⁷¹

Economics is concerned with the efficient use of society’s material resources. In societies characterized by specialization and a sophisticated division of labor, almost all economic activity involves some degree of instrumental motives. Thus by defining the notion of intrinsic rewards so narrowly, psychologist’s have restricted their attention to a very small subset of economically relevant behaviors.

The focus, as peculiar is it might appear from an economist’s perspective, makes perfectly good sense from the perspective of psychology. After all, psychology is concerned with understanding the reward structures that drive human behavior. Why then should psychology privilege economically relevant motives?

There is another important difference in the ways that psychology and economics analyze extrinsic and intrinsic motivation: the handling of autonomy. Deci, Ryan, and other psychologists argue that feelings of autonomy and competence are fundamental to human happiness. To the extent that they cause people to become accustomed to responding to rewards rather than their own intrinsic drive for self-realization, extrinsic rewards undermine a fundamental determinant of psychological wellbeing.⁷²

Positive economics, in contrast, conceives of autonomy simply as a means for achieving some productive end. For example, in standard principal agent models, high levels of autonomy are warranted when an agent has better information than does the principal about the consequences of actions, and can be rewarded on the basis of the value created by selecting the best action from a choice set. Even in Ellingsen and Johannesson’s esteem model, autonomy awarded by the principal serves the instrumental purpose of allowing the agent to signal valuable information to the principal. Economists have only begun to explore the interesting and provocative possibility that autonomy has value in and of itself and that the use of targeted extrinsic rewards (in the psychological sense) undermines an individual’s feeling of autonomy and competence.⁷³