RESEARCH DOSSIER

Social Preferences Altruism and Reciprocity

Abstract. Social preferences are economically consequential because choices can depend on the welfare of others, the distribution of material outcomes, perceived intentions, social norms, and the meaning of interaction. This research dossier reconstructs the analytical foundations of altruism and reciprocity, compares distributional and intentions-based models, derives predictions for canonical experimental environments, and evaluates the identification and transportability of the resulting evidence. It examines unilateral allocation, ultimatum bargaining, trust, public goods, punishment, labor contracts, and charitable giving; it distinguishes individual preference from equilibrium strategy and private incentive from social welfare. Comparative evidence from small-scale societies and an international preference survey situates behavioral heterogeneity in institutional context. The central analytical conclusion is that the interpretation of a prosocial choice requires knowledge of the feasible alternatives, information, strategic horizon, and institutional meaning of the decision. The work is an original interpretive synthesis with transparent illustrative calculations, rather than a new named theory or original empirical estimation.

Keywords: social preferences; altruism; reciprocity; inequity aversion; behavioral microeconomics; experimental economics; conditional cooperation; social norms; trust; welfare

The economic object of social preferences

Economic action distributes resources, creates expectations, and communicates interpretations of relationships. Social preferences are preferences in which another person's material outcome, the distribution of outcomes, or the social meaning of an action enters the decision maker's evaluation. Altruism concerns another person's benefit. Reciprocity concerns a response to the behavior or perceived intention of another person. Fairness concerns the evaluation of allocations and procedures in relation to a relevant standard. Each concept can be translated into restrictions on choice, but each restriction has different empirical consequences.

This dossier asks when a costly transfer, cooperative contribution, or refusal of an advantageous transaction reveals a social preference, and when an observationally equivalent explanation is more plausible. The principal answer is that the observable choice must be interpreted jointly with its feasible alternative, information set, strategic consequences, and institutional setting. A sacrifice in a truly anonymous, terminal decision is strong evidence against a narrowly specified own-money-only benchmark. The same sacrifice in a repeated market can be an investment in a relationship. A refusal to share may express material self-interest, an entitlement judgment, or opposition to the recipient's prior behavior.

Fehr and Charness's (2025) review of social preferences locates distributional, reciprocal, guilt-related and image-related motives within a common empirical research program. Its broad conclusion about the prevalence of social preferences should be read as an assessment of accumulated evidence, rather than a claim that a single scalar describes every person or context. A scientific account must preserve the difference between a preference, an elicited measure, a strategic action and a welfare judgment.

The unit of analysis throughout this work is the economic choice in a specified environment. The following chapters use established formal models and deliberately constructed numerical examples, then assess the identification strategies behind prominent experiments. They do not estimate a new population parameter. The world scope refers to the available comparative evidence, not to an exhaustive census of societies.

Historical development and the material-payoff benchmark

The economic concept of self-interest is often confused with an assertion about human nature. Maximization specifies consistency of choice under assumptions; the content of the objective function is a separate matter. Economics can represent concern for a child, a business partner, or an unrelated stranger within a utility function while retaining constrained optimization. A model that treats own money as the sole relevant outcome is analytically valuable because it produces sharp predictions in controlled environments. Its empirical domain must be established rather than assumed.

Becker's (1974) theory of social interactions represents an important stage in the incorporation of interpersonal dependence into economic analysis. It provides a historical bridge between demand theory and the economic treatment of relationships. The subsequent experimental literature sought to distinguish preferences about other people from the indirect monetary advantages obtained through relationships.

Güth, Schmittberger, and Schwarze's (1982) ultimatum bargaining experiment helped establish a setting in which a responder can reject a proposed division and thereby eliminate both monetary awards. The design makes the responder's power economically consequential. In the game with a known indivisible monetary unit and strictly positive own-money payoff, a selfish responder accepts any positive offer. If the responder rejects a zero offer, the proposing player anticipates acceptance and offers the smallest positive amount; acceptance at zero instead permits a zero offer. A systematic departure invites an explanation of the preferences, beliefs or decision processes that render the departure intelligible.

Rabin's (1993) fairness approach to game theory moved attention toward the intentions inferred from feasible alternatives. The development is significant because distributions alone may be inadequate to distinguish a deliberately exploitative choice from an unavoidable outcome. The literature that followed proceeded along both distributional and intention-sensitive lines. The empirical challenge changed from demonstrating deviations from the simplest monetary benchmark to identifying which extension of the benchmark accounts for observed behavior.

There is no requirement that a useful model reproduce moral language literally. Its scientific value lies in whether its variables correspond to plausible mechanisms and whether choice predictions can be tested against rival descriptions.

Choice, payoff and utility

Let a feasible action a determine a vector of material payoffs x(a) for individuals i and j. Under the narrow monetary benchmark, i chooses a to maximize x_i(a), assuming certainty and linear valuation in the relevant range. In a deterministic problem a strictly increasing transformation of own payoff leaves the ranking unchanged. With uncertainty, expected utility requires additional assumptions concerning risk and the treatment of random outcomes.

A familiar altruistic specification is U_i = x_i + θ_i x_j with θ_i ≥ 0. It represents marginal concern for the other's money relative to one's own money under linear utility. It has a transparent prediction: if a transfer costs c and creates b for the recipient, the person gives when θ_i b ≥ c. A transfer costing 3 and producing 12 is acceptable when θ_i ≥ 0.25. A transfer costing 3 and producing 2 requires θ_i ≥ 1.5. If a subject accepts the first opportunity and rejects the second, the observation supplies an interval for θ_i under the maintained assumptions, not a point estimate.

The formulation becomes inadequate if willingness to help depends on baseline wealth, recipient need, the origin of the funds or the identity of the beneficiary. A nonlinear specification can incorporate decreasing marginal welfare from consumption; a distributional model can incorporate relative positions; an intentional account can condition on whether the beneficiary chose the preceding action. These remain different explanations even when they predict the same transfer.

The econometric implication is elementary but important. A person's utility coefficients cannot be recovered independently of the scales and normalizations used in the model. Cardinal welfare comparisons across people require additional ethical or measurement assumptions. Choice data provide revealed rankings over feasible outcomes. They do not by themselves establish interpersonal units of happiness, moral worth or social contribution.

As a counterexample, suppose an allocator chooses (8, 8) rather than (10, 1). A positive weight on the recipient may explain it, and so may disutility from an unequal distribution. Offer both (8, 8) and (7, 20). A sufficiently high altruistic weight selects the latter while a strong inequality-averse preference may select the former. Designs that introduce such divergences are more informative than repeatedly observing equal division.

Distributional fairness and inequity aversion

Fehr and Schmidt's (1999) inequity-aversion model provides a parsimonious way to represent disutility from unfavorable and favorable payoff comparisons. In the two-person case its standard expression is:

Ui(xi,xj)=xi−αimax⁡(xj−xi,0)−βimax⁡(xi−xj,0).U_i(x_i,x_j)=x_i-\alpha_i\max(x_j-x_i,0)-\beta_i\max(x_i-x_j,0).

Here α_i ≥ β_i ≥ 0 and β_i < 1. The parameters describe two different slopes. When i is behind, reducing j's payoff can increase i's utility despite reducing the sum of money. When i is ahead, increasing j's payoff can raise i's utility even at a personal cost. In the many-player formulation the comparison terms are averaged over the other players. These are assumptions of the model, not psychological identities or diagnostic tests.

For an ultimatum endowment of 100, let the responder receive s ≤ 50 and the proposer receive 100 − s. Acceptance gives the responder s − α(100 − 2s), while rejection yields zero. Thus acceptance requires s ≥ 100α/(1 + 2α). With α = 0.5 the threshold is 25; with α = 1 it is 100/3. The derivation assumes rejection eliminates both monetary awards and that comparisons refer to game payoffs. If participants evaluate outside income or future relations, the threshold need not hold.

In a dictator allocation of 100, suppose the allocator transfers t ≤ 50. Utility is 100 − t − β(100 − 2t), or 100(1 − β) + (2β − 1)t. The maximizing transfer is zero when β < 0.5 and 50 when β > 0.5; when β = 0.5 all transfers in that interval are indifferent. The model's simplest form therefore has difficulty generating stable interior allocations without heterogeneity, curvature, noise or additional motives.

The counterexample explains why a formally elegant account of rejection cannot automatically explain the continuous distribution of giving. Empirical research should report the restrictions used to fit both games and the evidence for relaxing them.

Relative shares, social surplus and competing predictions

Bolton and Ockenfels's (2000) ERC model emphasizes absolute own payoff and a person's share of aggregate payoff. For nonnegative payoffs with a positive total, the relative share is x_i divided by the sum of players' payoffs. A reference share of 1/n in an n-person problem makes an appropriate comparison independent of which other participant holds a given portion of the remainder.

The distinction can be tested using three people. Consider outcomes A = (30,30,30) and B = (30,0,60), with i in the first position. Own payoff and relative share both equal 30 and one third. A pure function of those two arguments is indifferent between A and B. Pairwise inequity aversion assigns a comparison cost to B, because one other player is ahead and another behind. With comparisons averaged over two others, the cost is 15(α + β). The example is an analytical deduction, not experimental data.

Charness and Rabin (2002) use simple distributional tests to compare explanations based on reductions in inequality with those based on improvements in social welfare, especially for low-payoff recipients. This direction brings the absolute levels of other people's payoffs and the efficiency of the allocation into the foreground.

The implications are concrete. Suppose a decision maker chooses between (10,10) and (10,20). A positive altruistic weight prefers the latter; inequity aversion penalizes it if α > 0. A social-welfare preference that values the least advantaged might be indifferent about the minimum alone but prefer the higher total when efficiency also enters. In a richer menu the investigator can vary total surplus, the minimum payoff, advantageous disparity and disadvantageous disparity independently.

A useful empirical comparison tests predictions away from equal-split situations. An equal split is consistent with a focal convention, an efficiency concern in some restricted menus, a fairness objective and strategic fear of refusal. The observation cannot carry the full explanatory burden.

Altruism, public goods and warm glow

Pure outcome-oriented altruism values the recipient's welfare or the total public good. Warm-glow giving adds utility from one's own contribution. Andreoni's (1990) warm-glow analysis formalizes the distinction in a public-good setting. A concise representation uses private consumption c_i, total provision G, and own gift g_i: U_i(c_i,G,g_i), where the third argument captures a direct benefit from contributing. The pure public-good specification omits that argument.

Consider a donor with budget 100, a charitable good funded by the donor and others, and no administrative losses. If the donor wishes total provision to equal 40 while others initially contribute 10, a transfer of 30 attains the target. When external funding rises to 20, an unchanged interior optimum of 40 requires the donor to reduce the gift to 20. This full displacement is conditional on the simplified fixed-target example and its budget structure. Once utility depends directly on contributing, the same total provision may no longer compensate the donor for reducing the gift. Equilibrium responses can consequently be partial.

Public and private contributions can interact through beliefs about charity effectiveness, taxes, matching grants, peer responses and the possibility of aid substitution. A design that merely observes a fall in donations after a subsidy cannot identify the donor's objective. If recipients receive more money overall, a decline in private gifts can be consistent with concern for outcomes.

The welfare issue is whether a donor's satisfaction from the act is included alongside recipient welfare and public provision. Counting the same money twice would be an accounting mistake; counting different persons' utilities under explicit ethical weights is a normative choice. The objective of fundraising may differ from that of maximizing social welfare if solicitation expenditures or donor pressure are costly.

Reciprocity, intentions and sequential games

Reciprocity is conditional response to another person's action, perceived intention or relationship. Positive reciprocity can make costly rewards attractive after a beneficent action. Negative reciprocity can make costly sanctions attractive after conduct interpreted as harmful or unfair. The two responses need not have the same strength.

An intentions-sensitive analysis distinguishes a feasible choice from its realized consequences. Imagine two environments in which an agent receives exactly (8,2) from a principal's previous decision. In environment A the principal could instead choose (5,5). In environment B (8,2) is the only feasible allocation. If the agent can pay one unit to reduce the principal's payoff by three, an outcome-only account using the same original reference points predicts the same response in both. An intentional account may predict stronger punishment in A. Holding the available choice set transparent is important: an inferred intention based on incomplete information may be false.

Dufwenberg and Kirchsteiger's (2004) sequential reciprocity theory generalizes reciprocity analysis to games with successive decisions. The agent's evaluation can depend on beliefs about another's beliefs and on the history of choices. Falk and Fischbacher's (2006) theory of reciprocity likewise connects fairness evaluations with responses across strategic settings.

A reciprocity equilibrium cannot generally be computed by appending an unexplained kindness score to a normal-form payoff matrix. Expected kindness depends on a reference comparison and on the alternatives perceived by the player. In dynamic games beliefs must be compatible with the timing and information structure; otherwise an apparently reciprocal preference is really an unspecified forecast.

Identification benefits from randomized changes in whether an action was chosen voluntarily, determined randomly, or constrained by an external rule. A social norm about entitlements may also react to the same manipulation. A clean comparison reduces the set of explanations but does not necessarily isolate one completely.

Trust, risk and the investment game

Trust describes willingness to expose oneself to another agent's discretion in anticipation of conduct that justifies the exposure. Trustworthiness is the behavior of the agent receiving that discretion. A person who trusts may be optimistic about repayment, risk tolerant, concerned for the other person's outcome, or responsive to what the interaction symbolizes.

Berg, Dickhaut, and McCabe (1995) develop the investment-game experiment in which a first decision maker sends a monetary amount that is multiplied before a second decision maker chooses how much to return. To illustrate the strategic material benchmark, suppose a sender possesses 10 and chooses t between 0 and 10; the receiver receives 3t and sends back r between 0 and 3t. Their incremental payoffs are (10 − t + r, 3t − r), ignoring any separate receiver endowment. In a terminal game with known own-money maximizers, the receiver returns zero and the sender sends zero by backward induction.

If the sender sends 4 and receives 5 back, the realized allocation is (11,7). The sender gains 1 relative to retaining 10 and the receiver gains 7. The transfer alone does not reveal the expected return, and the return alone does not establish a reciprocal preference. A receiver may return money from fairness or altruism even if an identical receipt resulted from a random mechanism.

Cox's (2004) trust and reciprocity identification emphasizes comparisons that separate interpersonal motives from strategic behavior. A useful extension gives one group a choice over distributions without a preceding voluntary transfer and another an objectively identical distribution resulting from a sender's action. A difference in returning behavior estimates the effect of that procedural treatment, subject to the maintained experimental conditions.

In field lending, a decision to provide trade credit can be rational under expected repayment and legal enforcement, even if bilateral social preferences are absent. Measuring default expectations independently and describing enforcement are necessary steps toward making trust more than a label for successful exchange.

Dictator games and experimental design dependence

The dictator game removes the recipient's immediate rejection power. It therefore helps distinguish an allocation under unilateral discretion from an offer in bargaining. Its baseline interpretation depends on what the allocator believes about ownership, anonymity, the experimenter's expectations and whether participants regard the experimental funds as earned property.

List's (2007) dictator-game design comparison varies taking options and the origin of the endowment. The underlying methodological lesson is that the choice menu helps define the meaning of transfers. When only nonnegative gifts are permitted, the smallest observed gift is zero even if the person would prefer to take resources under another institution. When taking is allowed, the previously censored region becomes observable, but the change in menu may also alter perceived rights.

Engel's (2011) dictator-game meta-study compares diverse experimental specifications, including approaches distinguishing positive giving from amount conditional on giving. The original journal issue is dated 2011; the later Cambridge platform posting is not the original research date.

Suppose treatment A produces a mean gift of 4 from ten participants: five give zero and five give 8. Treatment B has ten participants each giving 4. Both means are 4, but participation, dispersion and responsiveness can differ radically. A policy requiring at least a small gift affects all participants in the first group differently from the second. A report should show the extensive margin, positive amounts, distribution of negative transfers if permitted, and the choice menu. Averaging alone conceals mechanisms.

An experiment with a minimum positive transfer does not measure the propensity to initiate giving. A design with a preselected default may measure inertia alongside preferences. A context with visible participant decisions may measure social signaling along with other-regarding evaluation. These are estimands to be specified explicitly.

Ultimatum bargaining and the interpretation of refusals

An ultimatum responder can reject an offer and reduce both players' monetary payoffs to the disagreement outcome. This creates a credible costly veto. A proposer must consider both the responder's evaluation of offers and the potential loss from rejection. In a heterogeneous population, even a strictly self-interested proposer may choose a substantial offer because the acceptance probability rises with the offered amount.

Write P(s) for the acceptance probability of offering s out of a total T. The material expected return is (T − s)P(s). For an interior optimum under a differentiable response, (T − s)P'(s) = P(s). A proposer who disregards the recipient's welfare can nevertheless offer a large share when acceptance is sensitive to the offer. The observed distribution of offers thus combines beliefs, preferences and strategic risk.

A numerical example uses a stake of 100 and only two allowable offers. An offer of 20 is accepted with probability 0.6, giving the proposer expected payoff 48. An offer of 40 is accepted with probability 0.9, giving expected payoff 54. The higher offer is optimal under risk neutrality and monetary self-interest. These invented probabilities are illustrative and do not describe a particular experimental population.

The interpretation of rejections also needs care. A responder's negative utility from an unfair allocation can account for rejection; retaliation against intentional mistreatment can also account for it. A public rejection may communicate commitment or resolve to future partners. A terminal anonymous setting eliminates some strategic reputational benefits if its anonymity and terminal nature are credible to participants.

A richer experimental comparison manipulates the disagreement payoff, the proposer's feasible allocations and the information about the decision process. It also elicits a response schedule before observing an offer, if the strategy method is appropriate. The resulting behavior can then be compared with decisions made in real time, rather than assuming emotional immediacy is irrelevant.

Public goods and conditional cooperation

For n participants endowed with w, let each contribute g_i in the interval from zero to w. The classic linear voluntary-contribution payoff is:

xi=w−gi+m∑j=1ngj,0<m<1<nm.x_i=w-g_i+m\sum_{j=1}^{n}g_j, \qquad 0<m<1<nm.

The private return on another unit contributed is m − 1, which is negative. The group's aggregate material return is nm − 1, which is positive. Individual monetary optimization predicts zero contribution in the one-shot game, whereas maximizing the aggregate monetary surplus calls for full contribution.

With n = 4, w = 20 and m = 0.4, zero contributions yield total payoff 80. Full contributions yield each participant 32 and a total of 128. A unilateral additional contribution of one unit reduces the contributor's payoff by 0.6 and increases the other three payoffs by 1.2 in total. In the simplified linear altruism specification, a uniform weight θ on each other person's payoff can support such a contribution when θ ≥ 0.5. This result assumes additive weights and ignores inequality, intentions and strategic effects.

Fischbacher, Gächter, and Fehr's (2001) conditional cooperation experiment elicits willingness to contribute at different contribution levels of other participants. Conditional cooperation is empirically distinct from unconditional altruism: a subject can choose high contributions when peers contribute and low contributions when they do not.

The dynamic implication is that pessimistic beliefs may induce low contributions even among conditional cooperators. Observing the low contribution can then reinforce pessimism. A temporary intervention that establishes credible expectations of participation may change an equilibrium path without changing the underlying distribution of preferences.

This also limits institutional comparisons. A group with low observed contributions may contain willing cooperators operating under unfavorable expectations, or a large share of persons who prefer to free ride. A policy that simply exhorts generosity addresses different mechanisms from a policy that makes contributions observable or creates enforceable assurances.

Punishment and the costs of social enforcement

A punishment institution permits participants to impose a loss on another person, often while incurring a private cost. In a one-shot monetary benchmark without future effects, costly punishment is unattractive. If the decision maker values norm enforcement or retaliation, it may be chosen.

Fehr and Gächter (2002) study costly punishment in public goods as a mechanism affecting cooperation. The term altruistic punishment describes the willingness to bear a private cost for sanctioning in the experimental framework; it is not a conclusion that every punishment is morally justified or welfare improving.

Consider a sanction costing the punisher 2 and destroying 6 units of the target's payoff. The immediate aggregate loss is 8. If the sanction prevents future free riding that would otherwise destroy a surplus of 15, it can generate a positive material net effect of 7, before discounting and administrative costs. If the target was already cooperating and becomes less willing to contribute after being punished, the same instrument can destroy even more surplus. The calculations are illustrative.

Herrmann, Thöni, and Gächter (2008) identify antisocial punishment across societies as an important limitation of the simple enforcement story. An institution that authorizes punishment does not guarantee that sanctions will target conduct contrary to collective welfare. The reported relationships with civic norms and legal institutions should be regarded as associations unless a causal design supports further inference.

Appeals, monitoring accuracy, predictable penalties and restrictions on retaliation have economic value precisely because privately motivated enforcement can be misdirected. The relevant comparison is not between a group with punishment and a group without any accountability. It is between feasible enforcement arrangements, including centralized adjudication, peer sanctions and combinations of both, each with distinct information and resource costs.

Norms, entitlements and the social meaning of choices

A social norm is a shared expectation concerning conduct in a specified situation. Descriptive expectations concern what others do; normative expectations concern what others regard as appropriate. A person may follow an appropriateness judgment despite believing that most others will violate it. Conversely, a person may follow a common action while considering it unjust.

Krupka and Weber (2013) introduce norm elicitation through coordination as an incentivized method of asking participants to match the relevant social assessment. The logic is that common expectations of appropriateness can help explain why otherwise similar distribution games generate different behavior. Such scores approximate shared judgments under the elicitation design and do not independently establish an ethical standard.

Entitlement judgments can depend on how an endowment arose. Equalizing the outcome of an arbitrary lottery may be perceived differently from equalizing unequal earnings obtained by different effort. To test this distinction, a decision maker's own payoff and the recipient's payoff can be kept identical across environments while the allocation procedure changes. Equal distributions in the subsequent decision need not have identical normative status.

An important issue is selection of the comparison group. Workers may assess a wage by comparing it with a colleague doing comparable tasks rather than with every worker in a country. Donors may regard family responsibilities as obligations with a different priority from those owed to strangers. A change in reference group can change behavior even without a change in the intensity of underlying concern.

Social preferences are embedded in institutions partly because institutions define rights and categories. A contract establishes ownership and promises; a benefit program defines eligibility; a public contribution system defines what counts as shirking. Preference measurement must describe these background rules, not merely the final numbers on a screen.

Social image, self-image and endogenous information

The economic benefit of appearing helpful can arise from future partnerships, approval or internal self-assessment. Visibility and attribution determine whether an action generates a signal. A public donation may provide stronger reputational benefits than an anonymous one, while anonymity can leave intrinsic concern, norms and self-image intact.

Ariely, Bracha, and Meier (2009) examine image motivation and monetary incentives. A reward may change how an audience interprets a prosocial decision, so the monetary effect and signaling effect need not reinforce each other. For a clear inference, the visibility of the action and the visibility of the incentive must be specified separately.

Information avoidance complicates the interpretation of observed selfishness. A person can choose whether to learn the harm an action creates. When acquiring information is optional, subsequent decisions conditional on being informed are selected by the prior information choice. The resulting sample cannot be treated as if its information state were randomly assigned.

Fehr's December 2025 corrigendum on willful ignorance corrects the discussion of Grossman and van der Weele's model in the Fehr–Charness review. It clarifies that deliberately avoiding information is itself adverse evidence about altruism in the observer's inference, while knowingly choosing a selfish outcome produces a more adverse inference. This dossier does not rely on the superseded interpretation and does not claim an independent derivation of that model.

A practical design separates information cost, control over ignorance, observability of the information decision and observability of the final act. Forcing disclosure can reveal the consequences of an action but may also remove a decision from the menu. In contrast, rewarding informed behavior may alter its signaling value. These distinctions are essential when translating an experiment into disclosure policy.

Labor markets, wage fairness and incomplete contracts

In incomplete employment contracts the worker often controls effort dimensions that cannot be written into an enforceable agreement. A firm's offer may therefore shape discretionary effort through perceived fairness, gratitude, relational expectations and the anticipated consequences of evaluation.

Fehr, Kirchsteiger, and Riedl's (1993) experimental market investigation studies fairness within a market structure. A simplified gift-exchange example clarifies the possible economic logic. A firm can pay a contractual wage of 10 or 14. The worker can supply ordinary effort or enhanced effort at an incremental private cost of 3. If enhanced effort generates 8 additional revenue and a higher wage reliably induces it, the firm's incremental return is 8 − 4 = 4. The worker receives the wage increment of 4 and bears effort cost 3, gaining 1. Both sides can gain relative to the low-wage, ordinary-effort outcome.

The example establishes possible complementarities, not the frequency or persistence of gift exchange. If the increased effort lasts only one of four equally productive periods while the wage premium is paid in all four, the added revenue is 8 and additional wage expense is 16. The policy then loses money even if the initial reciprocal response is real. Time horizon matters.

Gneezy and List's (2006) field tests of gift exchange move the inquiry into actual work assignments. The study is especially useful for separating an early response to an unexpected reward from a durable improvement in performance. The field setting also differs from an abstract laboratory game because fatigue, task learning, outside expectations and employment norms can influence the path of effort.

A wage structure should be evaluated jointly with productivity measurement, worker retention, selection effects, outside options and employer commitment. A positive laboratory effect does not by itself establish the optimal wage premium for a particular industry.

Incentives, autonomy and the interpretation of contracts

Material incentives modify the budget and may also communicate meaning. A minimum effort requirement changes the feasible action set; its introduction by a discretionary manager may signal skepticism. The incentive therefore has a mechanical component and a possible motivational component.

Falk and Kosfeld (2006) find in an experimental principal–agent game that the hidden costs of control can include reduced voluntary performance. The publisher abstract reports a nonmonotonic relationship between control and principal payoff, and participants' perceptions include distrust and reduced autonomy. The experimental result does not establish that all monitoring is harmful.

Gneezy and Rustichini's (2000) A Fine Is a Price supplies a complementary problem: a sanction attached to a previously normative obligation can change the perceived terms of compliance. If subjects interpret a fine as a payment authorizing the behavior, the institution may shift from a duty to a priced option. The relevant outcome depends on perceived obligations, not solely on the marginal financial penalty.

There are compelling cases for monitoring: verifiable minimum quality protects clients, reduces uncertainty about performance and can prevent exploitation by agents. The economic question is the net effect after taking account of direct deterrence, interpretation, administration and any change in effort above the minimum.

A credible experiment should compare an identical floor introduced by the manager, by an impartial rule and by random assignment where feasible. Under each arrangement one can measure effort above the floor. The treatment contrast provides a way to distinguish the effect of the constraint from the significance of who instituted it. Field implementation should also account for informed consent, contractual rights and data minimization.

Charitable giving and the welfare of donors

An observed donation may arise from concern for the beneficiary, enjoyment of giving, fear of disapproval, or reluctance to refuse a direct request. The social value of a donation is therefore not measured by its gross amount alone. Costs of solicitation, pressure and alternative uses of the donor's resources matter.

DellaVigna, List, and Malmendier (2012) use charitable solicitation and avoidance to investigate altruism and social pressure in door-to-door giving. The possibility of avoiding the interaction provides information about whether people welcome the request or wish to evade its potential cost. The design is informative because a solicitation treated as purely an opportunity to benefit a favored charity would have different avoidance implications from a solicitation that creates an unwelcome obligation.

Suppose a donor transfers 10, the charity incurs 4 to obtain the transfer, and the donor experiences an additional psychological burden estimated at 3 monetary-equivalent units under a specified welfare model. Net charitable receipts equal 6; the donor's money cost is 10; the pressure cost is separate. Whether the donation increases social welfare depends on the utility of giving, the benefit generated by the 6 that reaches the program and the social weights used in aggregation. Subtracting 3 mechanically from the charity's net cash would confuse monetary accounting with utility accounting.

A well-designed solicitation system can preserve transparent information about the beneficiary while making refusal simple. Evaluation can distinguish opt-in participation from gifts elicited under direct personal pressure. Where matching donations or public badges are introduced, investigators can test visibility and financial payoffs separately.

The charitable organization has an institutional objective, but the researcher must distinguish organizational revenue maximization from donor welfare and beneficiary welfare. A high-response fundraising technique may be optimal for the organization's short-term revenue and inferior when all relevant welfare consequences are included.

Reputation and repeated exchange

Repeated interaction creates material incentives for cooperation even when participants care solely about their own long-run payoff. A person who sacrifices today may retain profitable transactions tomorrow. The presence of a reciprocal-looking action therefore does not uniquely establish reciprocal preferences.

Use a repeated prisoner's dilemma with mutual cooperation payoff R = 3, unilateral defection payoff T = 5 and mutual defection payoff P = 1. Assume perfect monitoring, an indefinitely repeated game and a grim-trigger strategy. Cooperation is sustainable against a unilateral deviation when R/(1 − δ) ≥ T + δP/(1 − δ). Rearranging yields δ ≥ (T − R)/(T − P) = 0.5. This calculation does not require social preferences; it requires patience and the specified strategy and information conditions.

If a relationship is known to end immediately, a conditional material incentive to maintain that relationship disappears. Other concerns may remain, such as concern for an observer, spillovers into another market or an intrinsic obligation. A laboratory experiment should describe both the stated terminal condition and the credibility of anonymity.

In a market with incomplete monitoring, an adverse outcome may stem from a partner's mistake or external shock. Rigid retaliation can then destroy a profitable relationship. A forgiving response can be rational under uncertainty and also arise from preferences. Consequently, comparing the behavior of two partners after a negative shock requires information about what was observed and how reliable it was.

Reputation systems can facilitate exchange but also propagate biased reports or punish newcomers with limited histories. Institutional welfare depends on the design of information and appeal mechanisms. The same public signal that encourages honesty may amplify social pressure. It is necessary to analyze the equilibrium effects on entry and on who receives the benefits of trust.

Cross-cultural evidence and comparative institutions

Cross-cultural comparisons show why a model calibrated in one university laboratory should not be interpreted as a universal quantitative description. Henrich and colleagues (2005) report experiments in small-scale societies that broaden the range of observed economic organizations. The authors associate aspects of prosocial behavior with market integration and everyday cooperation. The comparisons do not establish that increasing market integration in any particular society will causally increase a specific preference measure.

The Global Preference Survey analyzed by Falk and colleagues (2018) uses the 2012 Gallup World Poll to cover 76 countries and more than 80,000 respondents. The global survey methodology includes probability sampling, weights, translated items and an experimental validation procedure used to select questions. It includes separate elicited measures of altruism, positive reciprocity, negative reciprocity and trust. Most observations are standardized survey responses rather than individual paid games administered in every country.

The two research programs illuminate complementary levels of evidence. Small-scale studies offer detailed institutional context and behavioral choices in selected settings. A standardized survey improves systematic comparative coverage. Neither eliminates all semantic differences in cooperation, entitlement, hypothetical response patterns or respondent selection. A national average is a weighted summary of measured responses in a particular sampling design and period; it is not a permanent cultural essence.

To compare places, investigators need a common unit and a transparent account of local equivalences: the real value of experimental stakes, the meaning of anonymous transfer, the reference group in a fairness question, language and the role of existing obligations. A difference between regions may result from institutions, preferences, measurement differences or combinations of these. A rigorous interpretation leaves each pathway visible until evidence discriminates among them.

Survey measurement and aggregation

Survey-based measurement offers scale and comparability at lower cost than a large set of paid economic decisions. Its scientific strength depends on construct definition, question validation, translation, sampling, response uncertainty and the suitability of the measure for the proposed application.

If x_ic is an elicited altruism score for person i in country c and w_ic is an appropriate survey weight, a weighted country mean is Σ_i w_ic x_ic / Σ_i w_ic. That mean estimates the target population's score under the relevant weighting assumptions. A mean across countries with equal weights describes the typical sampled country; a population-weighted global mean describes the typical sampled person. These are different estimands. Neither can be compared to an unweighted experimental mean without examining the sampling design.

A standardized composite score is not a monetary transfer rate. If an index reports a value of 0.5 standard deviations above a pooled mean, that is a relative position under the chosen normalization. It does not imply that the subject gives 50 percent more money, nor that the person's utility coefficient is 0.5. An individual score also contains measurement error that can attenuate estimates of associations.

Cross-national regressions face additional difficulties: institutional indicators may be measured at the country level, error terms can be correlated geographically, and the effective independent sample for country-level comparisons may be closer to the number of countries than the number of respondents. Multiple country-level covariates can be jointly endogenous to long-run development. A causal interpretation requires more than statistical significance.

For policy analysis, measure outcomes such as actual contributions, delivery, persistence, repayment and welfare consequences independently when possible. Survey measures can assist in characterizing heterogeneity, but predictive validation against the intended outcome is a separate task.

Causal identification and experimental estimands

A treatment is an experimentally assigned change in the decision environment. A causal estimand can be expressed as E[Y(1) − Y(0)] for outcome Y, where the potential outcomes correspond to the two treatments. Random assignment supports estimation of a mean causal effect when assignments and measurements are implemented as specified and relevant assumptions about interference and missing outcomes hold.

For example, randomize the availability of a taking option in a dictator task. The outcome is the net transfer to the recipient, with taking recorded as negative. If the mean is 4 under the restricted menu and 1 under the extended menu, the illustrative estimated effect is −3 units. The effect describes changing the available menu. It does not separately identify shifts in norm perception, entitlement or expression of a previously censored preference.

In a two-by-two design, let A indicate public visibility and B indicate monetary reward. The interaction estimand is (Y_11 − Y_01) − (Y_10 − Y_00), where each Y is the corresponding mean. If the effect of visibility is 3 without a reward and 1 with a reward, the interaction is −2. Such a pattern is consistent with altered signaling, but mediation through signal interpretation requires more evidence, including measurement of what observers infer.

Interference is particularly relevant in public goods and reciprocal relationships: one participant's treatment can affect another participant's outcome. Individual assignment may estimate a mixture of direct and spillover effects. Assigning at a group level, or explicitly randomizing exposure patterns, addresses a different causal estimand. Statistical inference must align with the unit of assignment and the structure of dependence.

An adequate report states preregistered primary outcomes where available, eligibility criteria, treatment compliance, attrition by arm, uncertainty intervals, and procedures for multiple comparisons. Exploratory heterogeneity remains useful when labeled as exploratory. Estimating the behavioral mechanism requires a further identification argument beyond the validity of the average treatment effect.

Structural estimation and model comparison

Structural estimation attaches explicit behavioral assumptions to observed choices. Suppose a decision maker chooses option A over B with probability determined by a logistic function of the utility difference ΔU: P(A) = 1/(1 + exp(−λΔU)), with λ > 0. The parameter λ controls responsiveness to the utility index. Without a normalization, multiplying all utility coefficients by a common factor and dividing λ by that factor leaves predictions unchanged.

The issue is consequential in social preference studies because experimental conditions can alter both preference expression and decision noise. A higher proportion of equal divisions under one treatment could reflect increased inequity aversion, more consistent choice near a focal outcome, or greater salience of equality. A model that fixes the noise parameter across conditions may incorrectly attribute all behavioral differences to preferences.

Consider a design with alternatives whose utility differences depend on both a social coefficient and a monetary cost. Variation in only one price or one endowment can leave the two effects poorly separated. Presenting multiple trade-offs in which the social benefit and personal cost vary independently can identify distinct slopes, conditional on a stable decision model.

Model comparison should emphasize out-of-sample predictions and identifying moments. A fit to the same distribution used to choose the model is weaker evidence than predicting responses in a preregistered counterexample. Mixture models may describe heterogeneous types, but their latent categories are conditional on the chosen number of types, specification and observed choice tasks. They are not immutable classifications of persons.

Sensitivity analysis can compare linear versus curved utility, outcome versus intention dependence and alternative error specifications. A model chosen for one institutional decision may be useful despite being inadequate for another. Parsimony is valuable when it preserves the predictions central to the intended application.

External validity, replication and cumulative evidence

Levitt and List (2007) ask what laboratory social-preference evidence can reveal outside the laboratory. The issue is transportability: laboratories deliberately control dimensions that natural environments may vary. Stakes, scrutiny, selection, history and available alternatives can differ from the setting in which the economic question matters.

Internal validity concerns whether the treatment comparison identifies the stated effect in the study. External validity concerns whether this effect or an estimated mechanism can be applied to another population or institution. A well-identified result in a convenience sample does not automatically become an estimate for workers worldwide. Conversely, a nonrepresentative experiment can test whether a mechanism is possible under transparent conditions.

Camerer and colleagues (2016) undertake replication of economic laboratory studies covering eighteen selected studies. Such a project informs cumulative evidence but does not measure the replication probability of every social-preference result. Investigators should distinguish replication of the original design, reanalysis of the original data, and conceptual replication testing a similar mechanism in a new context.

A meta-analysis benefits from clear study selection criteria and from distinguishing decisions within one study from independent studies. Multiple reported estimates sharing a control group are correlated. Publication selection may preferentially preserve large or statistically significant results. Meta-regression on study characteristics can describe differences across designs but is not in general a randomized experiment about those characteristics.

For applied use, the strongest evidence often comes from convergence: a controlled treatment isolates a response, an independent replication assesses its robustness, a field study addresses implementation, and institutional records characterize the target population. Where such convergence is absent, the result remains a bounded finding rather than a universal parameter.

Welfare evaluation and the limits of revealed preference

A decision can be individually preferred and socially inefficient under a specified material objective, or individually costly and socially beneficial. Social preferences complicate aggregation because a person's utility can depend on another person's outcome and on norms surrounding the allocation.

Consider two distributions, A = (10,10) and B = (9,20). Under sum of money, B is superior because 29 exceeds 20. Under a criterion maximizing the minimum material payoff, A is superior because 10 exceeds 9. Under a positive weighted sum of individual utilities the comparison depends on weights and utility definitions. The examples illustrate normative alternatives rather than establishing which choice society ought to make.

If a decision maker experiences satisfaction from a recipient's welfare, valuing both recipient utility and donor satisfaction can be legitimate under some ethical commitments. It can also bias aggregation toward beneficiaries whose welfare is socially salient. A policy that values everyone's welfare directly must explain how it treats prejudice, envy and satisfaction from harming others. Descriptive utility is not automatically a suitable argument in a normative social welfare function.

Economic evaluation of punishment must include both the immediate losses and prospective effects on the coordination environment. Economic evaluation of public goods needs to identify beneficiaries beyond contributors. Economic evaluation of charity must distinguish net receipts, recipient outcomes, donor opportunity cost and pressure.

Distributional data and explicit normative assumptions permit meaningful debate without presenting an author's ethical weights as empirically discovered preferences. The scientific task is to explain the feasible trade-offs, identify whose interests change, and state how the social comparison is constructed.

Institutional design and economic development

Social preferences can modify economic outcomes in settings where contracts are incomplete, collective contributions are voluntary, enforcement is costly or information about intentions is imperfect. Institutions have a dual role: they structure the feasible actions and communicate expectations concerning what those actions mean.

A mutual insurance arrangement, for example, can exploit local information about genuine need. It may function through reciprocity, altruism, reputation and social enforcement. It may also exclude outsiders, impose conformity or be vulnerable to correlated shocks. Formal insurance can extend the risk pool and reduce dependence on discretionary favors, while still requiring trust in claims administration. The appropriate comparison concerns coverage, reliability, transaction cost and distribution of burdens rather than the apparent warmth of the institution.

In a market of independent firms, transparent quality standards can reduce reliance on personal networks. In a neighborhood project, visible pledges and conditional commitments may help coordinate contributors who are willing to participate if others do. In a large tax-funded public program, formal rules can make contribution less dependent on private beliefs about other citizens' generosity.

The institutional recommendation follows from mechanism identification. If the binding constraint is coordination among willing cooperators, credible collective commitment can outperform a penalty designed for determined free riders. If deliberate opportunism is prevalent, enforceability may matter more than communication. If the central problem is biased sanctions, oversight and appeal may improve outcomes.

Economic development can alter opportunities, information structures and relationships over time. A change in measured prosocial behavior following institutional development therefore cannot be attributed solely to a change in preferences. Comparative research should examine who enters a market, what options exist, how outcomes are observed, and which behaviors receive public or private reinforcement.

Original analytical synthesis

The preceding research supports a specific interpretive conclusion: a social contribution is simultaneously an allocation and, in many environments, a relational action. The payoff consequences and the meaning of the action can move in different directions. A higher wage can increase material opportunity and signal recognition; a fine can reduce wealth and convey permission; a gift can benefit a recipient and create an obligation. Separating these consequences provides a more informative economic account than attributing all observed cooperative behavior to a single coefficient.

A useful analytical comparison begins with three otherwise identical allocations. In the first, a person privately and voluntarily chooses to give the recipient an extra ten units at a cost of two. In the second, a random process produces the identical allocation privately. In the third, the person voluntarily chooses the same transfer and the recipient can observe the choice. In all three conditions, repayment and future interactions are excluded. The first versus second contrast changes agency; the first versus third changes observability while keeping voluntary agency and material outcomes fixed. An observed difference can identify a treatment effect, but intentions, appropriateness and image concerns must still be distinguished with additional design features.

The conclusion is that experimental evidence is most useful when the relevant rival explanations make different predictions. The equal division of a stake is often an uninformative point at which competing theories agree. Unequal but efficient allocations, constrained intentions, anonymous transfers, and variation in the origin of property create the discriminating cases. The selection of contrasts is therefore a substantive contribution of experimental design, not a secondary statistical matter.

A second conclusion concerns institutional transfer. The economic effect of a social preference is mediated by the technologies and rules through which the preference can be expressed. A person who values public provision may contribute less in a system that wastes donations, without becoming less concerned for beneficiaries. A trusting person may refuse credit in an environment with unreliable enforcement or incomplete information. Policy effectiveness therefore cannot be inferred from the level of a preference measure alone.

These conclusions synthesize existing models and evidence and extend their practical interpretation through the explicit counterexamples and derivations in this dossier. They establish no newly named theory, universal behavioral law or independently estimated causal parameter.

Research agenda and unresolved scientific questions

The field's main empirical challenge is to reconcile stable differences among people with changes in observed choices across decision environments. Repeated measures of the same person can separate between-person variation from within-person variation, provided the experiments do not introduce learning or strategic history that confounds comparison. Longitudinal field evidence can address persistence but faces mobility, attrition and institutional change.

A second challenge is measuring intentions under incomplete information. An employee may regard a manager's higher wage as generosity, a market necessity or an attempt to compensate for unpleasant conditions. A credible design can vary the manager's information and available alternatives while holding the employee's material compensation fixed. Beliefs should be recorded in a way that avoids inducing the very consideration the study seeks to measure.

A third challenge concerns the distribution of social preferences across different economic relationships. A person's allocation toward a household member, colleague and stranger can differ because of obligations, information or group boundaries. A representative survey designed for international comparison cannot independently recover every relationship-specific motive. Linking survey measures to field outcomes with appropriate privacy protection remains a complementary project.

Further research should also assess long-run welfare effects of enforcement and incentive redesign, including adaptation, displacement across tasks, pressure on vulnerable people and changes in market participation. A short-term increase in giving or effort cannot establish that the intervention is superior after such effects are accounted for.

This dossier is a targeted literature synthesis rather than a systematic review with a preregistered inclusion protocol. Source access ranges from publisher full HTML and abstract records to indexed excerpts and metadata. The formal examples are independently constructed and algebraically checkable; no raw-data reanalysis, exhaustive bibliometric coverage or independent experimental replication is represented. These limits define the evidential reach of the conclusions and are material to the final editorial assessment.

Conclusion

Social preferences expand the economic analysis of choice by making the outcomes and conduct of other people relevant to individual decisions. Altruism, distributional fairness, reciprocal response and social image have distinct implications under appropriately differentiated feasible choices. The monetary self-interest benchmark remains essential for recognizing the strategic incentives that can generate similar behavior without those preferences.

The evidence supports attention to heterogeneity and context rather than a universal numerical coefficient. The proper interpretation of experiments distinguishes treatment effects, underlying mechanisms and the scope of generalization. Economic institutions alter incentives, feasible actions, available information and social meaning; the effectiveness of their design follows from the mechanism they address. Welfare analysis requires explicit treatment of distribution, resource costs, private satisfaction and the interests of affected persons.

The independently developed comparisons and derivations in this work give a coherent account of why identical outcomes can reveal different motivations and why similar motivations can yield different outcomes across institutions. They are transparent analytical illustrations, not newly estimated facts. The research agenda is to test discriminating predictions, assess the persistence of responses and connect credible field and comparative evidence to well-defined institutional decisions.

References

Andreoni, James (1990). Impure Altruism and Donations to Public Goods: A Theory of Warm-Glow Giving. The Economic Journal, 100(401), 464–477.

Ariely, Dan, Anat Bracha, and Stephan Meier (2009). Doing Good or Doing Well? Image Motivation and Monetary Incentives in Behaving Prosocially. American Economic Review, 99(1), 544–555.

Becker, Gary S. (1974). A Theory of Social Interactions. Journal of Political Economy, 82(6), 1063–1093.

Berg, Joyce, John Dickhaut, and Kevin McCabe (1995). Trust, Reciprocity, and Social History. Games and Economic Behavior, 10(1), 122–142.

Bolton, Gary E., and Axel Ockenfels (2000). ERC: A Theory of Equity, Reciprocity, and Competition. American Economic Review, 90(1), 166–193.

Camerer, Colin F., Anna Dreber, Eskil Forsell, Teck-Hua Ho, Jürgen Huber, Magnus Johannesson, Michael Kirchler, Johan Almenberg, Adam Altmejd, Taizan Chan, Emma Heikensten, Felix Holzmeister, Taisuke Imai, Siri Isaksson, Gideon Nave, Thomas Pfeiffer, Michael Razen, and Hang Wu (2016). Evaluating Replicability of Laboratory Experiments in Economics. Science, 351(6280), 1433–1436.

Charness, Gary, and Matthew Rabin (2002). Understanding Social Preferences with Simple Tests. The Quarterly Journal of Economics, 117(3), 817–869.

Cox, James C. (2004). How to Identify Trust and Reciprocity. Games and Economic Behavior, 46(2), 260–281.

DellaVigna, Stefano, John A. List, and Ulrike Malmendier (2012). Testing for Altruism and Social Pressure in Charitable Giving. The Quarterly Journal of Economics, 127(1), 1–56.

Dufwenberg, Martin, and Georg Kirchsteiger (2004). A Theory of Sequential Reciprocity. Games and Economic Behavior, 47(2), 268–298.

Engel, Christoph (2011). Dictator Games: A Meta Study. Experimental Economics, 14(4), 583–610.

Falk, Armin, Anke Becker, Thomas Dohmen, Benjamin Enke, David Huffman, and Uwe Sunde (2018). Global Evidence on Economic Preferences. The Quarterly Journal of Economics, 133(4), 1645–1692.

Falk, Armin, and Urs Fischbacher (2006). A Theory of Reciprocity. Games and Economic Behavior, 54(2), 293–315.

Falk, Armin, and Michael Kosfeld (2006). The Hidden Costs of Control. American Economic Review, 96(5), 1611–1630.

Fehr, Ernst (2025). Corrigendum to Fehr and Charness (2025), Social Preferences: Fundamental Characteristics and Economic Consequences. December 17. American Economic Association.

Fehr, Ernst, and Gary Charness (2025). Social Preferences: Fundamental Characteristics and Economic Consequences. Journal of Economic Literature, 63(2), 440–514.

Fehr, Ernst, and Simon Gächter (2002). Altruistic Punishment in Humans. Nature, 415, 137–140.

Fehr, Ernst, Georg Kirchsteiger, and Arno Riedl (1993). Does Fairness Prevent Market Clearing? An Experimental Investigation. The Quarterly Journal of Economics, 108(2), 437–459.

Fehr, Ernst, and Klaus M. Schmidt (1999). A Theory of Fairness, Competition, and Cooperation. The Quarterly Journal of Economics, 114(3), 817–868.

Fischbacher, Urs, Simon Gächter, and Ernst Fehr (2001). Are People Conditionally Cooperative? Evidence from a Public Goods Experiment. Economics Letters, 71(3), 397–404.

Gneezy, Uri, and John A. List (2006). Putting Behavioral Economics to Work: Testing for Gift Exchange in Labor Markets Using Field Experiments. Econometrica, 74(5), 1365–1384.

Gneezy, Uri, and Aldo Rustichini (2000). A Fine Is a Price. The Journal of Legal Studies, 29(1), 1–17.

Güth, Werner, Rolf Schmittberger, and Bernd Schwarze (1982). An Experimental Analysis of Ultimatum Bargaining. Journal of Economic Behavior & Organization, 3(4), 367–388.

Henrich, Joseph, Robert Boyd, Samuel Bowles, Colin Camerer, Ernst Fehr, Herbert Gintis, Richard McElreath, Michael Alvard, Abigail Barr, Jean Ensminger, Natalie Smith Henrich, Kim Hill, Francisco Gil-White, Michael Gurven, Frank W. Marlowe, John Q. Patton, and David Tracer (2005). 'Economic Man' in Cross-Cultural Perspective: Behavioral Experiments in 15 Small-Scale Societies. Behavioral and Brain Sciences, 28(6), 795–855.

Herrmann, Benedikt, Christian Thöni, and Simon Gächter (2008). Antisocial Punishment Across Societies. Science, 319(5868), 1362–1367.

Krupka, Erin L., and Roberto A. Weber (2013). Identifying Social Norms Using Coordination Games: Why Does Dictator Game Sharing Vary? Journal of the European Economic Association, 11(3), 495–524.

Levitt, Steven D., and John A. List (2007). What Do Laboratory Experiments Measuring Social Preferences Reveal About the Real World? Journal of Economic Perspectives, 21(2), 153–174.

List, John A. (2007). On the Interpretation of Giving in Dictator Games. Journal of Political Economy, 115(3), 482–493.

Rabin, Matthew (1993). Incorporating Fairness into Game Theory and Economics. American Economic Review, 83(5), 1281–1302.