Chapter 1: What Self-Interest Leaves Out
Prologue: The Choice That Made Sense at the Time
The application on the screen
Imagine an office judged by the applications it closes. It is Friday, and one file is missing an attachment. The reviewer can reject it as incomplete or ask the applicant to send the document again. Rejection takes the file out of the backlog. Asking leaves work in the queue when the weekly report is due.
Perhaps the applicant sent the document already. To find out, the reviewer would have to open a separate record or contact another team. Neither task produces a completed case for the weekly report. The office set its target to become more responsive; now it can improve the count by leaving the applicant's problem unresolved.
In this hypothetical office, the application concerns assistance with a training fee. The applicant has reserved a place and needs to know whether the fee will be covered. From that side of the encounter, a rejection is the beginning of more work: finding the notice, learning what went wrong, and deciding whether another application can be completed before the course begins. A file can leave the office's queue and become a more urgent task in someone else's life.
The reviewer may care about fair treatment. So may the supervisor who must explain another missed target. The senior official who promised a shorter backlog may have had a good reason: applicants were waiting too long. They can all want applicants to get an answer. What divides them is the measure: each can satisfy the office's demand while leaving the person waiting.
Keeping the file open means asking for time and explaining why it would be premature to close it. The reviewer can defend that choice, but has to defend it. Rejection passes as routine compliance.
That difference will outlast Friday's report.
Nobody has to tell the staff to behave badly. The count teaches them which cases to favor. Supervisors learn which explanations produce awkward questions. Applicants learn whether persistence brings help or another requirement. The weekly count records the outcome of these adjustments without recording all the costs they impose.
A reviewer may keep the file open anyway. A supervisor may accept a lower count; an applicant may expose what happened. Each can act, but each must absorb a cost the routine avoids. An office that performs well only when conscientious people repeatedly pay that price is relying on a fragile form of quality control.
By the following week, the summary can describe a more productive department and an applicant who failed to supply the required evidence. The office's records may appear to support both statements. The account becomes misleading when it ends there, before asking how the evidence was handled and where the unfinished work went.
The circumstances around conduct
Once we know the outcome, we readily find a character trait to explain it. A successful employee is diligent. A customer who does not cancel is loyal. A citizen who misses a requirement is careless. Character can matter, but these descriptions can become substitutes for examining the conditions under which the behavior occurred.
An employee may have received a manageable assignment while a colleague received the difficult cases. A customer may have remained because departure required an unavailable record. A citizen may have complied promptly with a clear requirement and struggled with a contradictory one. To understand the conduct, follow the encounter far enough to see which of these accounts the evidence supports.
This book uses structural incentives to describe arrangements that connect conduct with expected consequences. Money is part of the account. So are professional standing, access, protection from blame, the possibility of another attempt, and the likelihood that a promise will be honored. An incentive can be written in a contract or learned through repeated experience. What people believe will happen can differ from what the designer intended.
The word needs some limits. A constraint limits what a person can do. Information affects what they know or believe. A capability affects whether they can carry out an intended action. A preference concerns what they value. An incentive connects an available action with an expected consequence. These conditions interact, but calling all of them incentives would make the word explain everything by describing very little.
A person who cannot read an instruction has a different problem from one who understands it and expects compliance to be punished. A person who refuses a payment because it violates a commitment has supplied evidence about values as well as rewards. A system can attract people already inclined to behave in a particular way without having transformed their preferences.
Teaching someone to read will not repair a penalty for doing the right thing. A better reward will not supply a missing skill. The remedy depends on which problem we have.
Nor should an account of pressure turn into an account of inevitability. A powerful arrangement can make a course of action easier, cheaper, or more defensible without making it unavoidable. People refuse advantages, protect strangers, maintain difficult standards, and organize to change the terms. Understanding the costs of these actions can make their significance clearer. It should not dissolve them into a claim that a different reward must secretly explain every act.
A designer is also a participant
Whoever set the office's target has a report to submit too, perhaps to someone defending a budget or fulfilling an election promise. A rule designed to influence people can be the designer's own adaptation to pressure elsewhere.
There can still be concentrated power. Some people choose the measures, control the evidence, and retain the right to revise the rules. Others bear consequences without an effective voice. Describing the arrangement as a system should not blur that difference. A network of obligations does not distribute authority equally merely because everyone participates in it.
Power can also be divided. A manager controls staffing but cannot alter the reporting requirement. A regulator can set a standard but depends on information supplied by the industry. A customer can leave one supplier but needs compatibility with a network they cannot replace. Repair requires finding the relevant authority and understanding what can make its commitment credible.
The chapters move through these positions. They begin with cooperation, moral judgment, and usable rules. They then examine rewards, institutional promises, and default paths before turning to companies, public policy, and markets. The final part considers the arrangements that endure across generations, the support needed for discovery, and the changing cost of adaptive persuasion. A decision at one desk eventually reaches a budget, a market, or a rule that outlives its author. Following those connections takes us beyond the desk without making the person sitting there disappear.
Historical records and research findings enter where they establish something that an imagined case cannot. Their periods, comparisons, and limits remain attached. Hypothetical cases are identified and used to expose a sequence of decisions. They do not stand in for evidence about how often a mechanism operates. A plausible story can reveal a question worth asking without answering that question empirically.
There is no promise of a final design that reconciles every interest. A dependable rule can burden an unusual case. A useful reward can displace unmeasured work. A safeguard can become an obstacle after the problem changes. The purpose of structural inquiry is to make these consequences available for examination and revision, including by the people who carry their costs.
Return to the application. The immediate decision concerns an attachment. The larger question concerns an institution's ability to notice what its own measure leaves out. A decent reviewer can keep the file open today. A more dependable arrangement would make fair examination part of the ordinary job, with the time, information, and authority it requires. That is where this book begins.
Chapter 1: What Self-Interest Leaves Out
The person who goes first
In an experiment reported by Urs Fischbacher, Simon Gächter, and Ernst Fehr, participants faced a revealing question. They could specify a contribution to a group project for each possible average contribution by the other members. Some were prepared to give more when others gave more; others were not.[1-1] A refusal to contribute could therefore mean two different things: wanting the benefit without paying, or declining to pay while others took the benefit. The visible action alone did not distinguish them.
A colleague who stops sharing information presents much the same puzzle. Perhaps the colleague wants to protect an advantage. Perhaps others have spent months claiming the shared work as their own. We would need evidence to decide between these accounts, and the evidence would affect what to do next. Asking for greater generosity leaves the first colleague with the advantage worth protecting; it asks the second to accept more of the treatment that prompted withdrawal.
Self-interest sounds like a sufficient explanation until it is made to explain everything. If helping is selfish because the helper values helping, the concept covers both taking and giving without identifying when either should occur. There is no observation left that could count against it. A useful claim must take a risk: specify the benefit being sought, the alternatives available, and what a person would be expected to do if the circumstances changed.
To follow the decision, we need to separate what someone values from what they can do. A preference concerns the former; a capability concerns the latter. Constraints limit available actions, while incentives change their consequences. People act on beliefs about all these things, sometimes mistaken beliefs. A person may want to contribute and have the means, yet expect the contribution to be wasted. Calling that withdrawal a preference for selfishness passes over the expectation on which the decision may have turned.
Consider a hypothetical group maintaining a shared access road. Everyone benefits from a passable road. One household cannot supply labor because its members are caring for someone at home. Another could help but expects others to do enough without it. A third is willing to work if the group agrees on dates and repairs the section everyone uses. Their absences from the first workday look identical. They have different causes and would respond to different changes. A more severe penalty might move the second household while imposing an additional burden on the first and leaving the third's coordination problem unresolved.
Calling all three selfish leaves the road group no closer to knowing how to get the repairs done. Calling all three victims of structure would be equally unhelpful. The household waiting to benefit without contributing is taking advantage of its neighbors. The other households' difficulties give us no reason to soften that judgment. What structural analysis can add is a more exact account of who is doing what, under which conditions, and what change might bring a household to the next workday.
The group also has interests that cannot be summed without argument. A household using the road daily gets a different benefit from one using it occasionally. Heavy vehicles impose different wear. A repair schedule may be convenient for people with flexible work and costly for people paid by the hour. Agreement that the road is valuable does not settle who should pay, what a fair share means, or whether the chosen repair is worth its cost. Cooperation can begin with a real conflict rather than a misunderstanding waiting to be cleared away.
Cooperation with a condition attached
The contribution schedules in the experiment give conditional cooperation a precise meaning. Willingness varied with what other participants contributed. That is evidence against treating every participant as a person who invariably maximizes an individual monetary return. It is not a permanent classification of people: a decision in a short experiment cannot establish how its maker will behave as a parent, colleague, or citizen.
For the people organizing a shared undertaking, conditional willingness creates a practical obligation. They need to let contributors know what others have done and give them some defense against exploitation. Making defection expensive may induce cooperation among people concerned only with their own monetary payoff. It may also reassure those already willing to help, but reassurance requires attention to what they can see and what they reasonably fear. One monitoring arrangement can serve both groups. Its failure may leave some calculating that defection pays and others concluding that further generosity would be foolish.
Return to the road. Suppose the work is done on scattered afternoons, and no record shows who supplied material or labor. Participants see their own contributions directly and encounter others' contributions irregularly. A household that provided gravel may appear absent to the household that cleared drainage. Each can feel that it is carrying the group. Lower effort can follow without anyone beginning with a desire to exploit anyone else. Without a shared account of the work, the neighbors turn what they have failed to see into a judgment about one another.
A record could help, although a record would introduce judgments of its own. Labor hours and purchased material require a conversion if they are to count toward one obligation. Some work is more demanding or needs scarce equipment. A public list may improve recognition while embarrassing people whose circumstances should remain private. A useful system would need a way to acknowledge legitimate differences without allowing every participant to declare its own exemption. Someone would have to keep the record, settle disputed entries, and explain how different contributions count. Asking neighbors to pay more attention would leave that work undone.
Conditional cooperation can also become unstable through repeated disappointment. A contributor observes less than expected and reduces effort. Someone else reads that reduction as evidence that the group is no longer serious. A sequence begins in which each person considers the next withdrawal reasonable. The resulting low contribution is consistent with several underlying motives. It cannot, by itself, tell us whether the group lacked cooperative preferences or lacked a credible way to coordinate them.
A determined free rider may respond to a contribution record by learning exactly how little is necessary to avoid attention. Someone who resents another household may refuse even under an excellent procedure. Some disagreements concern the project itself: a participant may prefer a cheaper road, a different route, or no collective work. The presence of conditional cooperation does not dissolve these conflicts. It expands the set of mechanisms worth considering beyond the assumption that everyone must first be made worse off for doing the wrong thing.
Suppose the road group responds to every hesitation with an intrusive inspection. People may spend the time defending themselves that they could have spent coordinating repairs. The neighbors cannot simply accept every claim of hardship without examination either; the willing households would then have to cover whatever others chose to withhold. They need a way to establish how much each household can reasonably undertake, without making an inquiry into capacity an occasion for humiliation. Evidence, proportionate requests, and a chance to correct an unfair assessment make it easier to accept scrutiny.
What an anonymous transfer can establish
Joyce Berg, John Dickhaut, and Kevin McCabe examined trust and reciprocity in an experimental investment setting. Their design studied trust extended to an anonymous counterpart while controlling for repeat-game reputation, contractual precommitment, and punishment threats. They observed transfers and returns inconsistent with a simple prediction of universal pecuniary selfishness.[1-2] The experiment provides evidence of behavior in that designed setting, not a warrant for trusting an unknown borrower with a livelihood.
The practical lesson concerns what counts as an explanation. It is possible for a person to value reciprocity even when repayment cannot buy a future favor. It is also possible for the original transfer to reflect several motives: hope of a return, concern for the other participant, curiosity, or an interpretation of what the occasion calls for. An observed action rarely arrives with its motive attached. Experimental design can exclude some explanations, but it cannot automatically grant the remaining one exclusive ownership of the result.
Outside a laboratory, motives accumulate. A person helps a colleague because the colleague needs help, because they expect to work together again, because the task is interesting, and because being the kind of colleague who helps is part of their self-understanding. These reasons need not cancel one another. An action can have personal rewards without becoming indifferent to another person's welfare. A structural account that insists on finding a single hidden payment becomes as reductive as a character account that insists on finding a single virtue.
Now give that colleague a deadline. Helping makes the helper's own work late, and lateness is punished mechanically. Assistance has acquired a cost that goodwill alone cannot remove. If assistance is recognized only when reported by the recipient, the recipient acquires power over the helper's reputation. If everyone must seek permission before assisting, cooperation may become more equitable but slower. These changes do not tell us what the person values. They affect how much that value can accomplish and how often it can be acted upon without injury.
A buyer may trust a seller's intentions and still insist on agreeing a delivery date before paying. Once recorded, the date protects both parties from different recollections of the promise. Colleagues who respect one another's effort may ask for a handover record because either of them could be absent when the work resumes. Trust in the person does not answer these ordinary problems of memory and continuity. Keeping a record can be a way of sustaining the relationship. If every request for verification is taken as an insult, cooperation becomes difficult to extend beyond people who know one another intimately.
A request to verify every small favor can consume the favor's value. If the demand for evidence is applied selectively, documentation becomes a signal of rank: some people's word is sufficient, while others must prove every claim. A sensible degree of assurance depends on the consequences of error, the availability of repair, and the relationship's history. The same level of formality cannot be justified merely because it is available.
The behavior an arrangement retains
When an organization becomes more aggressive, it is tempting to imagine that its members have collectively changed their values. That may happen, but other processes can produce the same appearance. People willing to follow the aggressive practice may enter. People uncomfortable with it may leave. Those who stay may alter their conduct without altering their private judgment. Promotions may give a small subset greater visibility and authority. The institution's apparent personality can change through selection and influence before anyone's convictions change.
Consider a hypothetical professional partnership. It evaluates members almost entirely on individually billed work. Time spent helping another member solve a difficult problem is rarely recognized. Initially, some partners help anyway. Over time, people who protect their own billing targets are more likely to receive resources and influence. New recruits hear advice from those who prospered under the arrangement. The observed reduction in assistance could reflect adaptation, sorting, imitation, or all three. Declaring that success made people selfish compresses several mechanisms into a moral story.
Changing the arrangement after years of selection is harder than changing its instructions. Recognizing assistance might alter behavior among existing members. It would not immediately restore colleagues who left or rebuild expertise that no longer exists. Recruiting cooperative people without changing advancement criteria might only restart the same process with a new cohort. A partnership can sincerely revise its statement of values while continuing to allocate scarce work, status, and opportunity through the old measure. Members will look to the allocation of work and promotion to see what sort of future the partnership actually offers them.
Selection also complicates comparison between organizations. A firm with cooperative behavior may recruit people already inclined to cooperate. Its procedures may help, but good outcomes do not by themselves establish how much the procedures caused. Another firm may face more fragmented work, greater turnover, or customers who cannot readily evaluate quality. Borrowing the first firm's visible rituals may leave the second firm's underlying problem untouched. Institutional imitation becomes more reliable when it identifies the problem a practice solves and the conditions required for the solution to travel.
People have some ability to select their environments, but that ability is unequally distributed. An experienced professional with savings can reject a damaging workplace more easily than someone whose immediate obligations depend on the next paycheck. Freedom to leave exists in degrees. A formally voluntary arrangement can still attach a very high cost to refusal. This does not make it identical to physical compulsion or erase every remaining choice. It does mean that the presence of a signed contract cannot settle every question about power.
Before describing a response as adaptation, we should ask who had to adapt and what refusal would have cost them. A customer can change suppliers; a resident may be unable to relocate. A senior employee can negotiate an exception that a junior employee cannot safely request. The same nominal incentive can operate as an attractive opportunity for one person and a threat to basic security for another. Aggregating their reactions into an average response may be useful for some purposes, but it can conceal the very distribution of pressure that explains how an arrangement works.
Scarcity without a universal psychology
An unpaid bill has a date; a future benefit may have only a probability. Someone who cannot pay next week's bill may reasonably decline a proposal with a larger expected return next year. No change in personality is needed to explain the priority. From the position of someone with reserves, the decision can look short-sighted. For the person making it, the difficulty may be surviving until the larger return arrives. The interval is part of the choice, even when a comparison of final returns leaves it out.
We should first establish which options can actually be financed. Suppose, hypothetically, that a worker can attend unpaid training that might improve future earnings. Attendance requires transport, childcare, and foregone work. A colleague with savings sees an investment; the worker without savings sees an immediate deficit. Both can believe the training is valuable. Advising the second person to develop the first person's patience fails to supply the resources that make patience feasible.
Resources do not explain everything. People with similar means can choose differently because they assess risk differently, have different obligations, or care about different futures. Some make costly mistakes. Some refuse opportunities for reasons they later regret. Respecting material constraints does not require pretending that every constrained choice is wise. It requires specifying the feasible alternatives before assigning responsibility for not choosing one of them.
A poorly designed assistance program can compound the problem if gaining help requires repeated absence from work or uncertain processing time. The formal benefit and the usable benefit are different. A person may decline to apply because an apparently available option cannot be incorporated into a reliable plan. Conversely, a very simple program can still fail if its benefit is too small to address the underlying shortfall. Administrative friction and inadequate resources are separable problems even when both end in low participation.
Low participation leaves the observer with a problem of inference. It may reflect a design failure, but calling every refusal an administrative obstacle would simply reverse the original error. Some eligible people may not want the offered service. Others may have information about its quality that the designer lacks. We have to be willing to discover that the offered service was unwanted or poor, rather than difficult to reach. An explanation that survives every possible finding tells us little about why people declined.
A further difficulty is that assurance has to arrive in time to affect a choice. In the road example, learning after the workday that other households intended to help will not recover the wages someone already gave up. A contribution record may be accurate and still arrive too late. Cooperation depends on the sequence in which promises, costs, and benefits occur. Someone has to commit before the final outcome can be known, and the risk of that interval has to fall somewhere.
The first contributor might reasonably accept part of that risk. Shared undertakings cannot begin only after every uncertainty has vanished. But a group that repeatedly asks the same household to advance labor or money is drawing on a reserve that other participants do not supply. Apparent generosity may become a hidden loan to the undertaking. When the household finally refuses, treating refusal as the beginning of the story erases the earlier advances.
This also explains why a promise to make things fair eventually can fail to restore cooperation. The promise asks an exposed person to bear another interval of uncertainty. A smaller commitment that can be checked soon may be more persuasive than an expansive assurance about the distant future. That does not make the smaller promise morally superior. It gives the participant something on which a next action can depend. Cooperation can recover through evidence accumulated in manageable steps, provided those steps address the original burden rather than merely asking the disappointed person to demonstrate trust again.
Cooperation beyond a comfortable circle
There is no reason to assume that cooperative behavior is always socially beneficial. People can cooperate closely to exclude outsiders, conceal wrongdoing, or preserve an unfair advantage. Trust inside a group can make exploitation outside it easier. A reliable agreement among powerful participants may be precisely what weaker participants need protection from. The existence of cooperation says something about coordination, not everything about justice.
A hypothetical neighborhood association illustrates the difficulty. Members contribute regularly, attend meetings, and enforce decisions consistently. Their agreement reserves a shared facility for long-standing residents while excluding newer tenants who help finance the property through rent. The association may be internally successful and externally unfair. Making the members more cooperative would not correct the exclusion. The analysis has to include the boundary of membership, the source of authority, and the people who bear costs without receiving a voice.
The same question arises in organizations. A team can protect one another by withholding information from reviewers. Loyalty may be experienced as a virtue by participants and as obstruction by customers. To ask for stronger culture without specifying the objects of loyalty is to leave the direction of cooperation unresolved. An ethical commitment must be able to reach beyond the nearest relationship when that relationship becomes a vehicle for harm.
Impersonal rules can help extend cooperation beyond familiar people. A person need not establish a private bond with every official or supplier if a claim can be evaluated through a procedure. But impersonality can also obscure differences that matter. A formally equal requirement may demand very different sacrifices from different applicants. Officials need discretion to consider relevant differences. They also need to give reasons that applicants can question, so that a claim does not succeed merely because its holder has found a sympathetic official.
Such an arrangement needs people capable of judgment. A perfectly specified instruction cannot anticipate every situation in which reciprocity, hardship, or unfair exclusion becomes relevant. Training, discussion, and moral education can improve the quality of interpretation. They cannot guarantee that an interpreter will have time, authority, or protection to act. Conversely, protection and authority cannot ensure that an interpreter recognizes the person whose claim deserves attention. Capability and permission have to meet at the point where a decision is made.
The resulting picture of human behavior is less convenient than a single proposition about selfishness. People defend themselves, reciprocate, misjudge, care, resent, learn, and sometimes accept losses for reasons they cannot reduce to a payment. An arrangement works on this mixed material. It may make cooperation safer, exploitation easier, or sacrifice more visible. Its effects depend partly on the responses it encounters and partly on the responses it retains.
When a shared undertaking fails, I would begin with the person who contributed first. What had others promised, and what could that person check before committing labor or money? Who absorbed the delay when a promise failed? Following those questions through the work gives us something firmer than a verdict on human nature. We can see whether the next willing person will inherit the same exposure and, with it, the same reason to withdraw.
Notes and Sources
[1-1] Urs Fischbacher, Simon Gächter, and Ernst Fehr, “Are People Conditionally Cooperative? Evidence from a Public Goods Experiment,” *Economics Letters* 71, no. 3 (June 2001), 397–404. Experimental design and contribution schedules reproduced in coauthor Ernst Fehr, “Cooperation and Public Goods Provision,” undated lecture, slides 20–24, https://web.mit.edu/14.193/www/Fehr-MIT-Lecture7-8CooperationPublicGoods.pdf. Contribution schedules distinguish conditional willingness from free riding in this experimental setting; the chapter makes no population estimate from the sample.
[1-2] Joyce Berg, John Dickhaut, and Kevin McCabe, “Trust, Reciprocity, and Social History,” *Games and Economic Behavior* 10, no. 1 (July 1995), 122–142. https://econweb.ucsd.edu/~jandreon/Econ264/papers/Berg%20et%20al%20GEB%201995.pdf. The anonymous investment experiment separates reciprocity from several strategic explanations. Its results do not establish that every transfer expresses pure trust or predict ordinary lending reliability.