Incentives as Information - The Moral Economy: Why Good Incentives Are No Substitute for Good Citizens - Samuel Bowles

The Moral Economy: Why Good Incentives Are No Substitute for Good Citizens - Samuel Bowles (2016)

IV. Incentives as Information

Machiavelli sought to design policies that would induce the self-interested to act as if they were “good.” Hume wanted to harness the “insatiable avarice” of the citizen-knave in the interests of the public good. This remains an excellent idea.

But a constitution for knaves may produce knaves, and may cause the good to act as if they were “wicked.” This is not news to successful businessmen such as David Packard:

In the late 1930s, when I was working for General Electric … , the company was making a big thing of plant security … guarding its tool and parts bins to make sure employees didn’t steal… . Many employees set out to prove this obvious display of distrust justified, walking off with tools and parts whenever they could.

When he founded Hewlett-Packard, he later wrote, he was:

determined that our parts bins and storerooms should always be open … which was advantageous to HP in two important ways… . The easy access to parts and tools helped product designers and others who wanted to work out new ideas at home or on weekends… . And open bins and storerooms were a symbol of trust, a trust that is central to the way HP does business.1

Like Packard, Aristotle’s Legislator knows that goodwill is important for a well-ordered firm or nation. He also knows that goodwill may be eroded by policies that seem well conceived for a world in which feelings of trust and mutual concern do not exist or do not matter. Perhaps his greatest challenge is to develop policies under which social preferences will be synergistic with incentives that appeal to economic self-interest, each enhancing rather than diminishing the positive effects of the other. Even if the Legislator wants to adopt a more modest, “do no harm” creed and seeks to design incentives that are simply additive to social preferences, he will have to learn why crowding out occurs.

Learning about Preferences from Experiments

In 2011, Sandra Polanía-Reyes and I set out to collect all the evidence from experimental economics bearing on the assumption that social preferences and incentives are separable. We found fifty-one studies, which used over one hundred subject pools, more than twenty-six thousand subjects in all, and which were conducted in thirty-six countries.2 The data set includes subjects playing Dictator, Trust, Ultimatum, Public Goods, Third-Party Punishment, Common Pool Resource, Gift Exchange, and other principal-agent games. These are all settings in which one’s actions affect the payoffs to others, so social preferences may affect a subject’s experimental behavior. In all but one of these games, the relation between the subjects is strategic, meaning that the payoff of each player depends on what the other player or players do, and each player knows this. The exception is the Dictator game, in which one player simply allocates a sum to a passive receiving player. (Recall that these games are described briefly in table 3.1 and in greater detail in appendix 2).

With few exceptions, however, the experiments were not designed to test the effects of incentives on preferences or to determine why these effects occurred, but instead to assess the nature and extent of nonselfish preferences. But as our analysis of categorical and marginal crowding out in the previous chapter shows, we found a method to test hypotheses about the effects of incentives on social preferences.

Because we will use the same method throughout, it is worth pausing to review its logic. The problem that Polanía-Reyes and I faced is that the experiments do not directly measure preferences; instead, what they measure is actions that subjects take under a varying set of constraints and expectations of material rewards. The task then is to use a subject’s actions to reverse engineer what must have been her evaluations of the possible outcomes of the game—that is, the preferences that induced her to act in the particular way she acted in the experiment. This method is called “revealed preferences” in economics.

James Andreoni and John Miller used this method to reverse engineer an altruistic utility function from experimental data in a study titled “Giving according to GARP” (GARP is an acronym for “generalized axiom of revealed preference”).3 An important message from their paper is that altruistic preferences and their influence on social behavior can be studied by using the same analytical tools that have been developed for the study of, say, the shopping habits of an individual and her taste for ice cream. This is the reason why in explaining what people do (in the experiments or in natural settings), I interpret their actions as an attempt to bring about some desired end, given the options open to them. The method is conventional in economics, but the content is not: its application to social preferences takes the method in novel directions. Not the least of these, as we have already seen (in figure 3.1), is that a person’s desired ends are not something fixed, but may be affected by the use of incentives.

A difficulty in inferring social preferences from experimental data arises because in a strategic interaction, it is generally the case that a subject’s actions will depend not only on her preferences, but also on her beliefs about what others will do. Preferences, to be clear, are an individual’s evaluations of the outcomes that her actions may bring about; beliefs are cause-and-effect understandings that translate her possible actions into their expected outcomes. Beliefs come in to play because the payoff that she receives in the above experiments (except the Dictator game) depends not only on what she does, but also on what others do.

To see how this complicates the problem of inferring preferences from the actions taken, recall the Trust game described in the previous chapter. When the investor transfers a significant portion of his endowment to the trustee (to be tripled by the experimenter, resulting in the trustee getting three times what the investor sent), we cannot jump to the conclusion the investor is generous. The reason is that his large transfer may have been motivated by self-interest combined with the belief that the trustee would back-transfer not less than half of what she receives. In this case, he would receive not less than $1.50 on every $1 sent, or a 50 percent rate of return on his “investment”; altruism need not necessarily be at work.

Perhaps we can, at a minimum, infer that the investor trusted the trustee. But even this is not the case, because his considerable transfer might have been based entirely on the belief that she was very poor and would keep every penny, which, combined with his generous preferences, would make this (for him) the best outcome of the game.

Despite difficulties in untangling beliefs and preferences as reasons for experimental behavior, well-designed experiments allow us to narrow the class of motives that may have been at work. In a one-shot (not repeated) Trust game, when the trustee returns a substantial amount to the investor, for example, that action clearly rules out self-interest. When, as in this case, a parsimonious explanation of the experimental behavior would be to attribute social preferences to the individual, I do so, in a phrase like “the subject apparently had motives other than self-interest,” the “apparently” being a reminder that experiments do not directly test for preferences. But given the plausibility of the idea that one can sometimes recover preferences from observed behavior, I sometimes drop the “apparently” where the meaning is not likely to be misunderstood. In addition, we can often predict the actions that would be taken by an entirely selfish person, or one for whom social preferences and incentives were separable (that is, simply additive), as we did in figure 3.4.

To recover preferences from experimental results, we followed the same strategy used to study marginal and categorical crowding out in the Irlenbusch and Ruchala experiment. We observed the total effect of incentives on the actions taken by the subjects and then noted whether that result differed from the predicted direct effect (the top arrows in both panels of figure 3.1). If the total effect differed from the direct (material costs and benefits) effect, then we could infer that an incentive had somehow altered the subject’s (unobserved) experienced values and had thereby affected her actions (via the bottom arrows in figure 3.1).4

This is what Polanía-Reyes and I found; and in most cases, the effect of incentives on experienced values was negative. The experiments show that policies premised on the belief that citizens or employees are entirely self-interested often induce people to act exactly that way. The challenge is to understand why.

The Meaning of Incentives

To help decipher the causes of crowding out, I will take a page from Friedrich Hayek, who taught economists to consider prices as messages.5 When, say, the price of bread goes up because of a drought in the U.S. Midwest, the following sort of message is conveyed: “Bread is now scarcer and should be economized; you should put potatoes or rice on the table tonight.” The genius of the market as a system of organizing an economy, Hayek pointed out, is that the message comes with its own motivation to pay attention; eating potatoes rather than bread will save money.

Incentives are a kind of price. We will see that had the authorities in Haifa consulted the Legislator, the message sent by a fine for lateness in picking up your child at the day care center would have been: “Your tardiness is inflicting costs on our staff, so you should try a little harder to come on time.” But while the price of bread often conveys the right information at least approximately, the lateness price for the Haifa parents must have sent a quite different message from the one that would have prompted the parents to “economize” on lateness.

In addition to the distinction between categorical and marginal crowding introduced in the previous chapter, we can distinguish two causal mechanisms by which crowding (out or in) might take place. First, incentives may affect preferences because they provide cues to the nature of the situation in which a person finds herself and hence may act as a guide to appropriate behavior, resulting in her applying a different set of preferences. (“If you are shopping, it is okay to be entirely self-interested; if you are with your family, it is not.”) In this case, we say that preferences are situation-dependent and that the presence and nature of incentives are part of the situation.6

The second type of crowding out arises because incentives may alter the process by which people come to acquire preferences over their lifetime. In this case, we say that preferences are endogenous. Evidence from experiments may shed light on the effects of incentives on this process, but most experiments are far too brief (lasting a few hours at most) to capture the kind of social learning or socialization, typically occurring during childhood or adolescence, that makes preferences endogenous. I consider endogenous preferences in the next chapter. Here I use experimental evidence to consider cases in which preferences are situation-dependent.

Situation-dependence arises because actions are motivated by a heterogeneous repertoire of preferences—spiteful, payoff maximizing, generous—the salience of which depends on the nature of the decision situation. Our preferences are different when we interact with a domineering supervisor, shop, or relate to our neighbors. The boss may bring out spite, while the neighbors evoke generosity.

To see how this works, think about gifts.7 Economists know that money is the perfect gift—it replaces the giver’s less well-informed choice of a present with the recipient’s own choice when she takes the money and buys the perfect gift for herself. But at holiday time, few economists give money to their friends, family, or colleagues. We know that money cannot convey thoughtfulness, romantic interest, concern, whimsy, or any of the other messages that gifts express. A gift is more than a transfer of resources; it is a signal about the giver’s relationship to the recipient. Money changes the signal.

Can the same be said of incentives? It is commonplace in psychology to think that it can. Mark Lepper and his coauthors explain why: “The multiple social meanings of the use of tangible rewards are reflected in our everyday distinction among bribes and bonuses, incentives and salaries… . They carry different connotations concerning, for example, [i] the likely conditions under which the reward was offered, [ii] the presumed motives of the person administering the reward, and [iii] the relationship between the agent and the recipient of the reward.”8 All three pieces of information conveyed by incentives—“the likely conditions,” “the presumed motives,” and the “relationship”—may affect the social preferences of the incentives’ target. Sometimes the news is not good.

Bad News

Incentives have a purpose, and because the purpose is often evident to the target of the incentives, she may infer information about the person who designed the incentive, about his beliefs concerning her (the target), and about the nature of the task to be done.9 Incentives may affect preferences for reasons that are familiar to economists, as Mark Lepper and his coauthors say, because they indicate “the presumed motives of the person administering the reward.” By implementing an incentive, one reveals information about one’s intentions (payoff-maximizing versus fair-minded, for example) as well as beliefs about the target (hardworking or not, for example) and the targeted behavior (how onerous it is, for example.) This information may then affect the target’s motivation to undertake the task.

The Boston fire commissioner’s threat to dock the pay of firemen accumulating more than fifteen sick days conveyed the information that he did not trust that the firemen were doing their very best to come to work, especially on Mondays and Fridays. For the firemen, the new situation—working for a boss who did not trust them—seems to have altered their motivation. Of course, we cannot know just what caused the spike in sick call-ins. It could have been a very bad flu outbreak. That is why we use experimental information in addition to natural observation to try to understand why crowding out occurs.

This “bad news” effect commonly occurs in relationships between a principal, who designs incentives (a wage rate, a schedule of penalties for late delivery of a promised service, and so forth), and an agent, who is being induced to behave more in the principal’s interest than the agent otherwise would. To do this, the principal must know (or guess) how the agent will respond to each of the possible incentives he could deploy. The agent knows this, of course, and hence can ordinarily figure out what the principal was thinking when he chose one particular incentive over other possible ways of affecting the agent’s behavior.

Here is an example of how this sometimes does not work out well in practice. In this experiment, as in the trust game played with Costa Rican CEOs and students, German students in the role of “investor,” the principal, were given the opportunity to transfer some amount to the agent, called the “trustee.” As usual in the Trust game the experimenter then tripled this amount. Recall that the trustee, knowing the investor’s choice, could in turn back-transfer some (or all or none) of this tripled amount, returning a benefit to the investor.10

But this version of the standard Trust game came with a twist. When the investor transferred money to the trustee, he or she was also asked to specify a desired level of back-transfer. In addition, the experimenters implemented an incentive condition: in some of the experimental sessions, the investor had the option of declaring that he would impose a fine if the trustee’s back-transfer were less than the desired amount. In this “fine treatment,” the investor had a further option, namely, to decline to impose the fine, and this choice (forgoing the opportunity to fine a nonperforming trustee) was known to the trustee and taken before the trustee’s decision about the amount to back-transfer. There was also the standard “trust” condition, in which no such incentives were available to the investor. Figure 4.1 summarizes the results.


Figure 4.1. Forgoing the use of incentives apparently crowded in reciprocity in the Trust game (Data from Fehr and Rockenbach 2003.) Reciprocation was greatest when the fine was available to the investor but its use was renounced.

In the trust condition (black bars, no fines), trustees reciprocated generous initial transfers by investors with greater back-transfers. But stating the intent to fine a noncompliant trustee (grey bars) actually reduced return transfers for a given level of the investor’s transfer. The use of the fine appears to have diminished the trustee’s feelings of reciprocity toward the investor. Even more interesting is that renouncing use of the fine when it was available (white bars) increased back-transfers (again, for a given amount transferred by the investor).

Only one-third of the investors renounced the fine when it was available; their payoffs were 50 percent greater than those of investors who used the fines. The bad-news interpretation suggested by Fehr and Rockenbach is that both in the trust condition and when the investor renounced the fine, a large initial transfer signaled that the investor trusted the trustee. The threat of the fine, however, conveyed a different message and diminished the trustee’s reciprocity.11 The fine option and the opportunity publicly to renounce its use provided the investor with an opportunity to send a trusting message to the trustee.

There are lessons here for the design of institutions and organizations. Crowding out as a result of the bad-news mechanism may be prevalent in principal-agent settings but can be averted where the principal has a means of signaling fairness or trust in the agent.

Polanía-Reyes and I wanted to know which subjects responded adversely to the threat of the fine. Our survey of the experimental evidence found that crowding out affects individuals who are intrinsically motivated or fair-minded. For payoff maximizers, it appears, there is nothing to crowd out. This unsurprising fact, like the positive effect of an investor renouncing use of the fines, has lessons for Aristotle’s Legislator.

Moral Disengagement

Incentives may cause crowding out for another reason, one less familiar to economists. In most situations, people look for cues for appropriate behavior, and incentives provide these cues. A plausible explanation of some of the framing effects of incentives is that it occurs because market-like incentives trigger what psychologists term “moral disengagement,” a process that occurs because “people can switch their ethicality on and off.”12

In this case, the adverse indirect effect of the incentive does not work by conveying information about the principal, and it may be at work even in nonstrategic settings. Incentives provide information about (as Lepper and his coauthors put it) “the likely conditions under which the reward was offered” and hence about how the individual ought to behave. In the experimental evidence, moral disengagement can be distinguished from the bad-news cause of crowding out: in the former, incentives are implemented by a principal who is a player in the game, but in the latter, the targets of the incentive are not playing against the incentive designer. Instead, the incentives are introduced by the experimenter (as in Cardenas’s experiments in rural Colombia) or possibly by peers in the game.

Situational cues may be very subtle, and our responses to them unwitting. When experimental subjects had the opportunity to cheat on a test and thus gain higher monetary rewards, less than quarter did so when the room was brightly lit, but more than half cheated when the room was slightly less well lit (the variations in lighting had no effect on whether one’s cheating could be observed). In another experiment, subjects who wore dark glasses were much less generous to their partners in a Dictator game than were those outfitted with clear glasses.13 Dark glasses and a darkened room, the researchers reasoned, gave the subjects a sense of anonymity. But it was entirely illusory: it is difficult to imagine that a subject could really think that wearing dark glasses would make him less observable, especially since the experiment was conducted at computer terminals in closed cubicles. Dogs, too, steal more when its dark, but the reason is probably not moral disengagement.14

The dark glasses were just a cue to anonymity and a sign that acting as if one were really anonymous would be okay. To see why this matters, imagine a survey question about personal pleasures asked in two different ways. First: “Please tell us about some experience that you enjoyed very much.” Second: “Please tell us about some experience that you enjoyed very much. In answering this question, remember that there is absolutely no way that anyone will ever be able to associate your name with what you reply.” The dark glasses, like the second version of the question could invite some kind of transgression.

Our degree of real anonymity changes dramatically as we move among family, the workplace, the marketplace, and other domains of social interaction. Alan Page Fiske provides a taxonomy of four psychological models corresponding to distinct kinds of social relationships: authoritarian, communal, egalitarian, and market, each with culturally prescribed patterns of appropriate behavior.15 Depending on the information they convey, incentives may signal that the situation corresponds to one of these four types, and in part because the degree of anonymity differs, they may evoke distinctive responses. Because incentives are common in markets and markets are associated with anonymity, or at least with impersonal, arm’s-length dealings with others, incentives may be a cue to imagined anonymity.

Here is an example, one that suggests that the moral-disengagement effects of market interactions are not simply a matter of anonymity taken literally. Armin Falk and Nora Szech entrusted University of Bonn students with the care of “their” healthy young mouse, showing each a picture of one of the very cute mice.16 They then offered them a payment if they would allow the mouse to be gassed to death. Before making their decision, the subjects were shown a very disturbing video of a mouse being gassed.

In addition to this “individual treatment,” they also implemented a “market treatment” in which the subject could sell her mouse to another student, who would then allow the mouse to be killed. They hypothesized that a student who was reluctant to surrender her mouse as simply an individual choice might be more willing to let the mouse die in the market treatment because the sale to another student distanced her from the deed.

Forty-six percent of the subjects in the individual (nonmarket) treatment were willing to surrender their mice for ten euros or less. When the mouse trustee could sell the mouse to a buyer, however, 72 percent were willing to let their mice die for that price or less.

Falk and Szech asked the subjects the minimum payment for which they would be willing to give up their mice. Using this information, they were able to calculate how much they would have had to offer the subjects in the individual treatment to get 72 percent of them to give up their mice—the same percentage of the subjects in the market treatment who had been willing to do so.

What they found was astounding: to get 72 percent of the subjects in the individual treatment to let their mice die, the subjects would have to have been offered 47.50 euros. Recall that in the market treatment, compensation of only 10 euros was sufficient for this number to let their mice go. This almost fivefold difference between the market and individual treatments in the experiment may be considered a measure of the moral disengagement created by the market setting.

Notice that it was not the incentives per se that reduced the price at which the market-treatment students would let the mice die. There were monetary incentives in both the individual and the market settings. What differed was that the subjects in the market setting could disengage morally.

It would be interesting to know how much saving the mouse would have been worth to subjects in a treatment in which an individual’s mouse had been commandeered by another subject and destined for a certain death, but could be saved by purchasing it back. If the results were similar to those obtained with the other market treatment, then we could conclude that it was the arm’s-length nature of the market setup, not the fact that it was a market per se, that accounted for the radical drop in willingness to pay to save the mouse in the market treatment.

A similar finding comes from Elizabeth Hoffman and her coauthors, who illustrated the framing power of names in an Ultimatum game.17 In this game—another workhorse of behavioral economics—one player, called the proposer, is given a sum of money by the experimenter and asked to allocate it to the other player, called the responder. The responder, knowing how much the proposer was initially given, may either accept or reject the offer (which is the ultimatum giving the game its title). If the proposer accepts the ultimatum, the game ends and the two parties take home the sums of money determined by the proposer’s split. But if the proposer rejects the ultimatum, the game still ends, but both go home empty-handed.

This game has been played throughout the world by hundreds of subject pools. In the next chapter, I describe some cross-cultural experiments that I conducted as part of a team of economists and anthropologists. Though results differ from culture to culture, responders commonly reject low offers. In postgame debriefings, some express anger at the unfairness of the proposer as a reason to prefer getting nothing rather than to allow the proposer to benefit from an unjust division of the pie.

The motivation of proposers who offer a large share of the pie to responders is more complicated. Offering a fifty-fifty split, for example, might be motivated by fair-mindedness or generosity, but it could also result when an entirely self-interested proposer believes that a respondent is fair-minded and will therefore reject a low offer.

Hoffman and her coauthors found that proposers’ offers and responders’ rejections of low offers were both diminished by simply relabeling the game the “Exchange” game and relabeling proposers and responders “sellers” and “buyers.” The renaming did not in any way alter the incentives at work in the game. Instead, it affected the subjects’ sense of appropriate behavior independently of the incentives. From the results, the new name appeared to diminish the respondents’ standard of what constitutes a fair offer, or to reduce the salience of fairness in the respondents’ minds. The proposers appeared to become either less generous or less concerned that responders might reject low offers.

The power of names has been confirmed in many (but not all) experiments since then. As may have been the case in the experiment by Hoffman and her coauthors, in some cases the framing effect appears to have altered subjects’ beliefs about the actions of others rather than their preferences.18

Naming the game is not necessary for framing effects to occur. Incentives alone can create powerful frames. A year before the first U.S. reality-TV show, Andrew Schotter and his coauthors found, in a clever modification of the Ultimatum game, that market-like competition for “survival” among subjects reduced their concern for fairness. The subjects played the standard game no differently from how it is conducted in other experiments: proposers’ substantial offers were accepted, and low offers were rejected. But when, in a subsequent treatment, the experimenters told the subjects that those with lower earnings would be excluded from a second round of the game, proposers offered less generous amounts, and responders accepted lower offers. The authors’ interpretation was that “the competition inherent in markets … offers justifications for actions that, in isolation, would be unjustifiable.”19

While this explanation is plausible, the experiment, like the others surveyed thus far, could not provide direct evidence for moral disengagement. Sometimes, however, we can directly measure how incentives cause ethical reasoning to recede in people’s minds.

A large team of anthropologists and economists implemented both Dictator and Third-Party Punishment games in fifteen societies, including Amazonian, Arctic, and African hunter-gatherers; manufacturing workers in Accra, Ghana; and U.S. undergraduates.20 Recall that in the Dictator game, an experimental subject is assigned a sum of money and asked to allocate some, all, or none of it to a passive recipient. The Third-Party Punishment game is a Dictator game with an active onlooker (the third party).

After observing the dictator’s allocation to the second party, the third party can then pay to impose a fine on the dictator. Most third parties tend to punish stingy dictators. And dictators anticipate this. One would expect that in the presence of a third party, and therefore with the prospect of being fined, dictators would adjust their allocations upward compared to their allocations in the simple Dictator game, in which there is no possibility that stingy offers will be punished.

But this was not what happened.

Surprisingly, in only two of the fifteen populations studied were the dictators’ offers significantly higher in the Third-Party Punishment game than in the Dictator game, and in four populations they were significantly lower. In Accra, where 41 percent of the dictator’s allocations resulted in fines by the third party, the allocations in the Third-Party Punishment game were 30 percent lower than in the Dictator game. The incentives provided by the fine did not induce higher allocations; they had the opposite effect.

Were ethical motives crowded out? There is some evidence that this is what happened. In the standard game, dictators who adhered to one of the world’s major religions (meaning, for this subject pool, Islam or Christianity, including Russian Orthodoxy) made allocations in the Dictator game that were 23 percent higher than those of dictators who did not follow one of these world religions. But in the Third-Party Punishment game, this “religion effect” virtually disappeared. Adherents of world religions behaved no differently from the unaffiliated. The presence of a monetary incentive appeared to define the setting as one in which the moral teachings of one’s religion, including generosity as a value, were not relevant.

Another piece of evidence suggests that the incentive provided by the prospect of a fine from a third party increased the salience of economic concerns. In the standard Dictator game, the dictator’s offer was uncorrelated with the dictator’s economic need (in the real world). But in the Third-Party Punishment Game, the dictator’s economic situation strongly (and significantly) predicted his or her offers. The incentives implicit in the Third-Party Punishment Game appeared to substitute economic motivations for religious or moral concerns. The results of these experiments were consistent with the expectation that crowding out operates through the effect of incentives on those with preexisting social preferences. It has no effect on those with little morality to crowd out.

The evidence on moral disengagement gives Aristotle’s Legislator plenty to work with, for he knows that tangible rewards may be framed as “bribes and bonuses, incentives and salaries,” as Lepper and his coauthors say, and one might add “and as prizes, fines, and punishments.” He also knows that moral frames for social interactions are not difficult either to construct or to suppress.

Control: Incentives Compromise Autonomy

The third reason that incentives can crowd out social preferences is that people may respond adversely to the political nature of the incentives, which are often transparently an attempt to control the target.21 Psychologists have explored how incentives (or constraints) can compromise a subject’s sense of autonomy, explaining how this harm reduces an intrinsic motivation to undertake the targeted task.22 The psychological mechanism at work appears to be a desire for the “feelings of competence and self-determination” that come from intrinsically motivated behavior.23

It is easy to see how incentives might convey the message that a principal (an employer, for example) wishes to control an agent (his employee). But most of the experimental evidence for this mechanism comes from nonstrategic settings (meaning that the experimenters, not a principal, implement the incentive). This self-determination mechanism thus differs from the previous two mechanisms—bad news about a principal and moral disengagement—because it arises from the target’s desire for autonomy per se. It does not depend on the target’s inferring the principal’s desire to control the agent, on other negative information about the principal, or on clues about appropriate behavior.

Lepper and his coauthors in a different paper explained why “a person induced to undertake an inherently desirable activity as a means to some ulterior end should cease to see the activity as an end in itself.” An incentive, they explained, may affect a person’s perceptions of his own motivation: “To the extent that the external reinforcement contingencies controlling his behavior are salient, unambiguous, and sufficient to explain it, the person attributes his behavior to these controlling circumstances. But if external contingencies are not perceived … the person attributes his behavior to his own dispositions, interests, and desires.”24 In this perspective, the person is constantly constructing or affirming an identity and is acting so as to signal to herself that she is autonomous. The presence of the incentive makes the signal less convincing because it provides a competing explanation for her behavior.

In cases where people derive pleasure from an action per se in the absence of other rewards, psychologists say that the introduction of incentives may “overjustify” the activity and reduce the individual’s sense of autonomy. An example is provided by the experiment mentioned in chapter I in which toddlers who were rewarded with a toy for helping an adult retrieve an out-of-reach object became less helpful than those who weren’t given a toy. The authors of that study concluded: “Perhaps when rewards are offered children simply come to perceive a formerly self-sufficient activity as merely a means to some more valuable end.”25 The effect may be to reduce child’s intrinsic motivation to help, and if the reward alone is insufficient, the child may simply stop helping.

In an iconic overjustification experiment, children were promised a reward for engaging in an activity that they had previously pursued enthusiastically for no reward. Children were selected for the experiment on the basis of their prior interest in painting and then asked whether they would like to draw with felt-tipped markers, under three conditions.26 Under the “unexpected reward condition,” after drawing for a period of time, the child was offered a “Good Player Award” card with a red ribbon, a gold star, and the child’s name on it. The second, “expected reward condition” differed in that the experimenter showed the child the award card and asked: “Would you like to win one of these Good Player Awards?” All assented and were given the award following the drawing session. In the third treatment there was no reward (expected or unexpected).

A week or two later, the experimenters observed the subjects’ play choices in their ordinary school setting. Those who had anticipated a reward for choosing drawing took up painting only half as frequently as those who had not anticipated a reward. Moreover, those who had been promised a reward also painted less than they had done before the experiment. And during the experiment itself, the drawings produced by children who received rewards were judged to be of substantially lower quality than those done by the students in the control group. (The judges of artistic quality did not know which treatment groups the children were in.)

The researchers designed the experiment to isolate whether the negative impact occurred because an activity was rewarded or because the subject chose the activity knowing that it would be rewarded. The fact that it was the anticipation of the reward, not the reward itself, that affected the children’s subsequent behavior suggests that the adverse effect was associated with compromised autonomy, not the receipt of rewards per se.

This interpretation is consistent with the observation that close supervision or arbitrary deadlines for completion of an otherwise enjoyable activity has almost the same negative effect as financial or other rewards. Lepper and his coauthors point out that the “detrimental effects of unnecessarily close adult supervision or the imposition of unneeded temporal deadlines suggest strongly that the effects … are the result of superfluous constraints on children’s actions, not a specific function of the use of tangible rewards.”27 Thus it may not be the material reward per se that is the cause of crowding out. The Legislator knows that if true, this idea will have important implications for the design of public policy. We will return to it.

In contrast to psychologists’ overjustification experiments, in which incentives are typically implemented by the experimenter, economists have studied strategic interactions in which the incentive is implemented by a player in the game. Here is an example.

Armin Falk and Michael Kosfeld implemented a principal-agent game with adult subjects to explore the idea that control aversion motivated by self-determination could explain why incentives sometimes degrade performance.28Experimental agents in a role similar to that of an employee chose a level of “production” that was costly for them to provide and beneficial to the principal (the employer). The agent’s choice effectively determined the distribution of gains between the two, and the agent’s maximum payoff came if he produced nothing at all.

Before the agent (the employee) made his decision, the principal could elect to leave the level of production completely to the agent’s discretion or instead to impose a lower bound on the agent’s production. (The levels of production required of the employee by these bounds were varied by the experimenter across treatments; the principal’s choice was simply whether to impose it.) The principal could infer that a self-interested agent would perform at the lower bound or, in the absence of the bound, at zero. Thus imposition of the bound would maximize the principal’s payoff.

But in the experiment, agents provided less production when the principal imposed the bound. Apparently anticipating this negative response, fewer than a third of the principals in the moderate- and low-bound treatments opted to impose the bound. This minority of “untrusting” principals earned, on average, half the profits of those who did not seek to control the agents’ choice in the low-bound treatment, and a third less in the intermediate-bound condition. The use even of the upper bound reduced the profits of the principals, although not significantly.

These results are consistent with the adverse reaction to control by others seen in psychological experiments. But it may have resulted either from control aversion per se or from the “bad news” about the principal conveyed by his placing limits on the agent. Gabriel Burdin, Simon Halliday, and Fabio Landini used a clever experimental design to distinguish between these two reactions to control.29 They first confirmed Falk and Kosfeld’s crowding-out results by using an identical treatment. They then introduced a third party who could impose the lower bounds on the “employee” but who would not benefit in any way from the employee’s “hard work,” which, as before, would contribute to the principal’s payoffs. If the adverse reaction to the lower bounds observed in the Falk and Kosfeld experiment was due to control aversion per se (rather than bad news about the principal), their third-party control treatment should have exhibited a similar negative reaction.

But it did not. Their interpretation is that the crowding out in the Falk and Kosfeld experiment was due primarily to the message that it sent about the principal and not to control aversion per se.

Consistent with the idea that incentives sometimes unintentionally are messages, the imposition of the lower bound in the Falk and Kosfeld experiment gave employees remarkably accurate information about employers’ beliefs about them. In postplay interviews, most agents agreed with the statement that the imposition of the lower bound was a signal of distrust, and the principals who imposed the bound in fact had substantially lower expectations of the agents. The untrusting principals’ attempts to control the agents’ choices induced over half the agents in all three treatments to produce at the minimum allowed, thereby affirming the principals’ pessimism.

This is an entirely new twist for Aristotle’s Legislator. Looking up from the details of the control-aversion experiment to think about entire societies, the Legislator has a disturbing thought. In a real economy, he worries, if most employers have low expectations of their workers, they will adopt compensation and supervision policies that, like the imposition of the lower bound in the experiment, will evoke minimal performance from their workers. And the workers’ slacking on the job will affirm the employers’ dismal expectations.

Depending on the distribution of principals’ prior beliefs about agents, he ruminates, a population with preferences similar to these experimental subjects’ could support either a trusting and highly productive outcome or an untrusting one in which production barely exceeds the minimum that could be extracted by fiat. In either case, the prior beliefs are perpetuated, so either outcome could persist indefinitely.

This might not matter much, the Legislator tries to reassure himself, in an economy in which the imposition of limits would be sufficient to secure most of what employers cared about. The first example that comes to mind, unhappily, is cane harvesting and other work done by slaves, whose pace of work is easily observed and maintained by an overseer with recourse to severe punishments. He is relieved to come up with the assembly line next, whose pace is a kind of mechanical regulator of the work effort of employees. Those who fall behind are easily singled out. But he is at pains to think of any examples from a modern economy—that is, one based substantially on services, intensive in knowledge and interpersonal care—in which a high level of production could be enforced by the imposition of such lower bounds on otherwise unwilling producers.

Happily it occurs to him that a onetime intervention that induced employers not to impose lower bounds on work input might reveal to them that workers, under the right conditions, were willing to provide substantial input in return for more trusting treatment. With this new knowledge, employers would adopt new, more trusting treatments of their workers even after withdrawal of the intervention that induced them to give up the lower bound in the first place. Thus a onetime intervention could transform a vicious circle into a virtuous one. In this case, the policy would not work by changing workers’ or employers’ preferences; instead, it would force employers to act in a way that would alter workers’ beliefs about them, in turn leading workers to act in ways that would induce the employers to adopt different beliefs about their work habits.

The same could be said of an intervention that would somehow induce workers to challenge employers’ low expectations of them by providing more input than required, even when the lower bound was imposed. Economists use the term “equilibrium selection” to describe this kind of convention switching. The Legislator adds it to his tool kit.

Emotion, Deliberation, and Crowding Out

Can the Aristotelian Legislator go beyond the insights of the “incentives as messages” approach to explore the proximate neural basis for crowding out? Could he map how an incentive acts as a stimulus to our deliberative and affective neural activity, and use this information to design policies that would activate rather than deactivate the neural pathways associated with cooperative and generous behavior?

For example, we will see that on the basis of neuroimaging and other evidence, one might be tempted to conclude that incentives activate cognitive processes that are more deliberative and less affective, and that deliberation tends to result in self-interested action. Could we make incentives synergistic with social preferences by framing subsidies and fines in such a way to stimulate emotional and visceral responses rather than calculating ones?

I have my doubts, as you will see, but the idea is not as farfetched as it sounds. To start, there is some evidence that incentives and social rewards activate different regions of the brain. To identify the proximate causes of the crowding-out behavior in Fehr and Rockenbach’s Trust game, described earlier, Jian Li and his coauthors studied the activation of distinct brain regions that occurs when trustees are faced with an investor who threatens to impose a fine for insufficient back-transfers. They compared these results with the neural activity in trustees who were not threatened with sanctions.30 As in the Fehr and Rockenbach experiment, the threat tended to reduce rather than increase back-transfers made by the trustee for a given level of investor transfer.

The brain scan showed that the threat of sanctions deactivated the ventromedial prefrontal cortex (a brain area whose activation was correlated with higher back-transfers in this experiment) as well as other areas related to the processing of social rewards. The threat activated the parietal cortex, an area thought to be associated with cost-benefit analysis and other self-interested optimizing. Li and his coauthors concluded that the sanctions induced a “perception shift” favoring a more calculating, self-interested response.

If it were confirmed that incentives like the investor’s threat of a fine in the Trust game activate brain regions associated with calculative self-interest, perhaps we could design incentives that do not have this effect. The “perception shift” in the subjects was between two quite different ways of responding to a stimulus: affective (meaning visceral or emotional) and deliberative (or cognitive). The philosopher-neuroscientist Joshua Greene describes these two processes as follows: “The human brain is like a dual-mode camera with both automatic settings and a manual mode.”31 The manual mode requires deliberate choice in picking the right settings; the automatic-setting option circumvents deliberation. In Greene’s terms, deliberative or cognitive processes are the manual mode, and visceral and emotional ways of responding, which make up the affective processes, are the automatic mode. Psychologists call this idea “dual-process theory.”32

In “Moral Tribes: Emotion, Reason, and the Gap between Us and Them,” Greene provides a framework for thinking about incentives and morals. First, deliberative processes are outcome based (in philosophical terms, “consequentialist”) and utilitarian, while affective processes support nonconsequentialist judgments (termed “deontological”) such as duty or the conformity of an action to a set of rules. Second, these ways of behaving are associated with activation in different brain regions, respectively, the (deliberative) prefrontal cortex and the (affective) limbic system.

The neuroscientific evidence then implies that economic incentives induce consequentialist reasoning (activation of the prefrontal cortex) and implicitly reduce the salience of deontological judgments (deactivation of the limbic system). If this view is correct, then the crowding out seen in many behavioral experiments implies that consequentialist reasoning is often (but not always) less prosocial than deontological judgment.

There is some evidence that this dual-process approach can identify the proximate causes of behavior not only in the Trust game but also in the other experiments. Alan Sanfey and his coauthors, interpreted an earlier experiment by Sanfey’s group in this way:

A neuro-imaging study examining the Ultimatum Game found two brain regions that were particularly active when the participant was confronted with an unfair offer, the anterior insula and the dorsolateral prefrontal cortex (dlPFC). Activation in these areas has been shown to correlate with emotional and deliberative processing, respectively, and it was found that if the insular activation was greater than the dlPFC activation, participants tended to reject the offer, whereas if the dlPFC activation was greater, they tended to accept the offer. This offers neural evidence for a two-system account of decision-making in this task.33

Struck by the activation of the insula, a region associated with negative emotions such as fear and disgust, and further associated with rejection of (presumably low) offers in the Ultimatum game, Colin Camerer and his coauthors comment: “It is irresistible to speculate that the insula is a neural locus of the distaste for inequality and unfair treatment.”34

Dual-process theory may also explain why the effects of activation of both the affective and the deliberative processes may be less than additive. Deborah Small, George Loewenstein, and Paul Slovic found that a picture of a needy girl induced more charitable giving than did statistics on need, and that providing both statistics and the picture yielded less giving than the picture alone.35 They concluded: “When thinking deliberatively, people discount sympathy towards identifiable victims but fail to generate sympathy toward statistical victims.” Here the affective system appeared to promote generosity, but was overridden by the deliberative system when that process was stimulated by the presentation of statistics. In this case, the presentation of statistics may have activated the prefrontal cortex and the deliberative process, just as the threat of the fine did in Li’s Trust game experiment, competing with and crowding out activation of more affective neural processing.

But it strikes me as unlikely that human goodness should emanate largely from the reptilian brain, namely, the limbic system and other brain regions in which humans are less distinctive than other animals, compared with our quite special prefrontal cortex. The idea that consequentialist reasoning is more likely to induce self-interest, and that deontological logic is more likely to favor ethical and other-regarding behavior, is far from obvious. If I take pleasure both in helping people in need and in eating ice cream, but cannot at the moment do both, it is not at all clear why activating my deliberative processes would lead me to go for the ice cream when my visceral response would be to help the person in need.

Why would incentives stimulate deliberation rather than emotion, and why would deliberation override positive social emotions such as sympathy? The first part of the question is easy: an incentive invites us to do cost-benefit calculations to determine whether the incentive is sufficient to motivate the targeted activity. These calculations are qualitatively different from our responses to emotions like sympathy, pain avoidance, or fear. If you have a painful burning sensation in your hand, you don’t generally deliberate about whether getting away from the fire is to your benefit. But why might deliberation result in less prosocial behavior? This remains an open question.

Indeed it would be a mistake to think that deliberation is the enemy of generosity. A study by Linda Skitka and her coauthors showed that among American liberals (compared with conservatives), the deliberative process is more generous toward those affected with HIV-AIDS, and is less influenced than the affective process by considerations of “responsibility” for contracting the illness.36

To study the balance of affective and deliberative processes, Skitka and her coauthors overloaded the deliberative capacities of experimental subjects by inducing high levels of what psychologists call “cognitive load,” as could be done, for example, by asking them to remember two seven-digit numbers (high load) as opposed to one two-digit number. In dual-process terms, high load is designed to degrade the deliberative process. Here is how Loewenstein and O’Donoghue interpreted the results: “The study found that subjects were less likely to advocate subsidized treatment under conditions of high [cognitive] load, which we would interpret as evidence that deliberative reactions are more concerned than affective reactions to AIDS victims. More interestingly, under conditions of high load, both liberals and conservatives were less likely to provide subsidized treatment to those deemed responsible (relative to those deemed not responsible), whereas under conditions of low load, liberals treated both groups equally whereas conservatives continued to favor groups who were seen as less responsible for contracting the HIV-AIDS.”37 They went on to suggest that the experiment might show that the “affective and deliberative reactions were consistent for conservatives—so cognitive load has no effect—but conflicting for liberals.”

I suspect that the deliberation-self-interest coupling does not really work: deliberative processes can produce moral and other-regarding judgments and behavior. Taking seriously the golden rule is an example.

Correspondingly, the coupling of affective processes with generous and ethical motivations seems questionable. Self-interested behavior must have a solid foundation in our less deliberative processes. As in all animals, visceral and other nondeliberative reactions in humans have evolved under the influence of natural selection. A good case can be made that among humans, natural selection may have resulted in a genetically transmitted visceral or emotional predisposition to help others even at a cost to oneself.38 But it would be surprising if such “automatic” reactions did not also induce self-interested behavior such as pain avoidance, satisfaction of sexual desire, and flight from danger. One could expect, therefore, that self-interested behavior would be aligned with the neural pathways associated with emotion no less than with deliberation.

If this is the case, we would expect that the difference between self-regarding and other-regarding behavior does not map neatly onto either the deliberation-emotion distinction in cognitive processing or onto the prefrontal cortex-limbic system distinction in neuroscience. The neuroscientist Jonathan Cohen says that the deliberative prefrontal cortex “may be a critical substrate for Homo economicus.”39 Cohen is right about the calculative aspect of economic man. But the prefrontal cortex may be no more implicated than the affective limbic system in the self-interest attributed to Homo economicus.

The neuroscience of a dual-process theory of social preferences is a burgeoning field still in its infancy. But if a last word is to be had at this early date, it goes to Loewenstein and O’Donoghue: “[The] deliberative system has a stable concern for others driven by moral and ethical principles for how one ought to behave. The affective system, in contrast, is driven toward anything between pure self-interest and extreme altruism depending on the degree of sympathy that is triggered.”40

It appears that incentives do make the deliberative processes more salient. But whether deliberation results in more generous behavior (as with liberals in the AIDS case) or less generous (as with the statistics about need and the picture of the needy girl) depends on whether the imperative to generosity resulting from deliberation (such as a Benthamite utilitarian calculation) is stronger than the generosity-inducing emotions (such as sympathy) of the affective process. A taxonomy of these cases, showing the lack of a simple mapping between the deliberative-emotional distinction and social preferences, appears in table 4.1.

Table 4.1. Dual-process theory and social preferences.


Note: From the table it appears that there is no simple mapping between cognitive-processing styles and preference types.

A Puzzle

The three reasons why incentives may crowd out social preferences—bad news, moral disengagement, and control aversion—provide some of the information that Aristotle’s Legislator needs in order to design incentives that complement rather than substitute for a desire to uphold social norms and to act generously toward one’s fellow citizens. In each case, policies can be devised to minimize the crowding-out problem and perhaps even induce crowding in. In considering the Legislator’s policy options in the final two chapters, I provide examples of effective incentives that contain clearly conveyed mutually beneficial purposes (avoiding moral disengagement) deployed by peers who have nothing to gain personally (avoiding bad news) and that can complement rather than erode intrinsic motivation (avoiding control aversion).

But we cannot go there quite yet. Evidence that social preferences are common, and that they underwrite mutually beneficial exchanges and other foundations of social life but are often crowded out by explicit economic incentives, presents us with a puzzle, one that if left unresolved might cast a shadow on my reasoning thus far. The adverse effect of incentives on generosity, reciprocity, the work ethic, and other motives essential to well-functioning institutions would seem to portend instability and dysfunction for any society in which explicit economic incentives are widely used.

Have societies somehow avoided the vicious cycle in which markets and other incentive-driven institutions erode the cultural foundations on which they depend, leading to the increasing use of incentives to compensate for the increasing deficiency of ethical and other-regarding preferences?41

Why did the Boston fire commissioner and the firemen not wind up in a kind of economic and cultural arms race to the bottom, in which the commissioner upped the ante by imposing ever more draconian pay deductions, and the firemen responded by acting in increasingly self-interested ways, until the firemen completely abandoned their sense of civic obligation in favor of precisely the opportunism the commissioner attributed to them at the outset?

And why would this dynamic not play out right across any market-based economy? Wouldn’t we end up with exactly the constitution for knaves that Hume advocated, but in contrast to Hume’s account, with a citizenry of knaves as well?

The experiments discussed so far, as well as casual observation, show that this is not our plight today. There are two possible explanations. The first is that I have misunderstood the crowding-out problem and perhaps overstated it. The second is that the corrosive effect of markets and incentives on social preferences indeed exists, but in many societies has been offset by other social processes allowing for the survival, and even flourishing, of a robust civic culture.