CHAPTER 1

Rationality, Social Preferences, and StrategicDecision-making from a Behavioral Economics Perspective

SIMON GÄCHTER

Introduction

The central assumption of the rational choice approach is that decision-makershave logically consistent goals (whatever they are), and, given thesegoals, choose the best available option. This model, in particular its extension tointeractive decision-making (game theory), has had a tremendous impact in thesocial sciences, in particular economics, and has allowed for great theoreticalprogress in the latter. For instance, the model and its formalization have led toimportant insights into the functioning of economic systems (see Mas-Colell,Whinston, and Green [1995] for a leading textbook account). They have alsoshed new light on numerous nonmarket processes, such as crime, addiction,family behavior, political decision-making, and organizational behavior(Coleman 1990; Becker 1993; Hechter and Kanazawa 1997; Gintis, Bowles,Boyd, and Fehr 2005). Rational choice theory in the form of game theory isnow the core theoretical tool of economics. It is therefore certainly justified tospeak of the rational choice model as a "success story."

Rational choice models often assume that agents are "unboundedly rational"and always know what is best for them. This assumption has long found manycritics (Simon 1982). A further assumption that is not part of the canonicalrational choice model but is frequently invoked in applications is that agentsare primarily self-regarding. This assumption has been challenged in particularin the last decade through the accumulation of empirical, in particularexperimental, evidence that has shown substantial and robust deviations fromwhat selfishness predicts.

In the past the discussions about the fruitfulness of the rational choiceapproach were based largely on philosophical convictions and less on facts. Inthis chapter I will argue on the basis of insights from behavioral economicsthat the usefulness of the rational choice approach is also an empirical questionand not just a philosophical one. My approach is related to that of Hechter andKanazawa (1997), who argue for the fruitfulness of rational choice explanationsby discussing its empirical successes across a large variety of interestingsociological domains. My arguments are based on data from many laboratoryexperiments which all share the feature that the theoretical predictions arederived from rational choice models (typically game-theoretic models) and thatdecisions in the experiments have financial consequences for the participants,which give them an incentive to take decisions seriously. Specifically, I willuse experimental results to argue that one can acknowledge that humansare boundedly rational and nevertheless appreciate the predictions made byrational choice models that at times rest on unrealistically strong assumptions.Moreover, I argue that a rational choice approach does not imply assuming thateveryone is selfish.

I start by giving a short characterization of the canonical rational choicemodel, including the selfishness assumption. With the help of one famousresearch program on the functioning of experimental markets I will illustrateone main point that will recur several times. I argue that one can fruitfullyuse rational choice theory to predict social outcomes even if the assumptionsentering the model are empirically invalid. I will first discuss the success andlimits of the rational choice approach. Then, I focus attention on the empiricalinvestigation of the selfishness assumption. I will present evidence from severaleconomic experiments that have been used as tools to uncover the structureof people's "social preferences." Numerous experiments have shown that theselfishness assumption is invalid as a description of typical behavior, although inall experiments we do find selfish people. The results on social preferences raisethe challenge of how to model them, and I will sketch some attempts.

Rational Choice and Homo Economicus

THE RATIONAL CHOICE MODEL

My main goal here is to set the stage for the subsequent analysis. For fulleraccounts of various aspects and criticisms of the rational choice approach Irefer to Coleman (1990); Hedström (2005); Elster (2007); Lindenberg (2008);and Gintis (2009).

The rational choice model aims to explain the decisions of individuals andthe individual and, in particular, the social consequences of those decisions.Figure 1.1 illustrates the core elements of the rational choice model. It is usefulto distinguish two conceptual buildings blocks: decision theory and equilibriumtheory. Decision theory describes how decision-makers should make decisions(normative approach), or actually do make decisions (positive approach). Incase individuals interact, a so-called solution concept describes the predictedsocial outcome. The usual assumption is that the resulting outcome will be anequilibrium.

Decision theory typically makes a conceptual distinction among preferences,beliefs, and constraints. Preferences describe how individuals rank the availablealternatives according to their subjective tastes. Beliefs are the second conceptualbuilding block behind the rational choice model. For instance, in which stocksyou invest will probably depend on your expectations about the future earningspotential of that stock. However, choice depends not only on one's subjectivetaste and beliefs but also on constraints. Constraints are the set of alternativesthat are available to an individual. For instance, you might prefer a Ferrari toa Ford, but your available income might constrain you from buying a Ferrari.After specifying preferences, beliefs, and constraints, the analysis proceedsby assuming that individuals choose a feasible option that best satisfies theindividual's preferences ("the utility-maximizing choice").

The rationality assumption enters the picture on the conditions that areplaced on preferences. The typical rationality assumptions are that (i) preferencesare complete, which means an individual can compare all relevant alternativesand rank them, (ii) preferences are transitive—that is, if an individual prefersalternative A over B, and B over C, then that individual should also prefer Aover C, and (iii) preferences are independent of irrelevant alternatives—thatis to say, the relative attractiveness of two options does not depend on otheroptions available to the decision-maker. This rationality assumption ensuresthat preferences are noncyclical and therefore contain a "best element," whichwill be chosen if available. Notice that the rationality assumption is nothingmore than a consistency requirement and is completely mute on the content ofpreferences. This has not gone without criticism. For instance, Amartya Sen, inan influential article, has mocked this conception of economic man as being a"rational fool" (Sen 1977).

Rational choice theory becomes a theory of social interactions if theindividual decision-makers interact with each other. The social sciences aretypically interested in the social outcomes of the interaction of individuals.One particularly important approach, especially in economics, is to analyzesocial outcomes from an equilibrium perspective. An equilibrium is a situationin which, given the decisions of all other decision-makers, no agent has anincentive to change behavior. If there is still an incentive to change behavior forat least one decision-maker, then the resulting outcome of the social interactioncannot be an equilibrium.

Two important classes of social interaction are markets and (small) groupinteractions. In markets decision-makers face prices and have to decidehow much they want to produce or buy at given prices and income. In theprototypical case a single individual cannot affect prices and thus takes themas given. An equilibrium is a situation in which at given prices agents want tochange neither their production plans nor their demands (for a comprehensivetextbook account, see Mas-Colell et al. 1995). In strategic situations anequilibrium is reached if, given the strategic behavior of other players, no onewants to change strategy. Thus, in many social science applications the analysisdoes not stop after looking at individual decisions but proceeds to predictingsocial consequences under the assumption that the resulting outcome will bean equilibrium. Of course, it is an empirical question whether the equilibriumprediction (which is derived for specific assumptions about people's preferences)is consistent with the data.

THE SELFISHNESS ASSUMPTION

The rational choice approach is a flexible framework that can account forany preferences as long as they obey the consistency axioms. This flexibilityis also a weakness, because if almost any preferences are permissible almostanything and therefore nothing can be explained. For that reason economistsvery often made additional preference assumptions to give the rational choiceanalysis more structure or assume that people have the same preferences (see,for example, the influential discussion by Stigler and Becker 1977). The mostcommon and long-standing assumption is that decision-makers are self-regarding.Self-regarding agents will take the welfare of others into accountonly for instrumental reasons to increase their own well-being.

In essence there are two justifications for the selfishness assumption. A firstjustification is tractability in modeling, as selfishness can simplify the analysisconsiderably. The selfishness assumption often allows for exact predictions,which can be confronted with appropriate data that might refute the selfishnessassumption. Moreover, it is often of independent interest to understand whatwould happen if everyone were selfish. Understanding the consequences ofselfishness serves therefore as an important benchmark for understandingnonselfish behavior.

I have already alluded to a second justification of the selfishness assumption,that assuming nonselfish preferences, although logically possible, is tantamountto opening "Pandora's Box"—which in this case means that any bizarre behaviorcan be rationalized. This argument has considerable merit in the absence ofempirical means to assess the structure of people's social preferences. Yet, asI will demonstrate below, the development of experimental tools allows usto observe people's social preferences under controlled conditions. Behavioralscientists have lately even added neuroscientific tools to understand the neuralmechanisms behind people's social preferences (Fehr 2009b).

FIVE METHODOLOGICAL REMARKS ONTHE RATIONAL CHOICE MODEL

I conclude this discussion with five methodological remarks and refer theinterested reader to Gintis (2009) for a more in-depth discussion of the issuesraised here. The first remark concerns the selfishness assumption, to which wewill return in more detail below. Nothing in the rational choice frameworkdictates that preferences have to be self-regarding; the only necessary assumptionis that preferences obey some consistency axiom such that optimal choicesare logically possible. Thus, a rational decision-maker can have other-regardingpreferences and still obey all the rationality axioms. As we will show below, thereis substantial experimental evidence for both that many people are not purelyself-regarding and that other-regarding preferences often do obey consistencyaxioms (at least in simple situations). We will also show evidence that behavioralpatterns in experiments in which social preferences are important are consistentwith predictions derived from rational choice models.

Second, one may criticize the rational choice approach as being unrealisticin the sense that human decision-makers do not reason and behave like thoseportrayed in the theory. This argument has a lot going for it, as has been pointedout emphatically from different angles by Simon (1982); Gigerenzer and Selten(2001); or Loewenstein (2007).

The limits of the rational choice approach can be discussed with the help ofFigure 1.1. Psychologists and other behavioral scientists have long argued thatpeople are faulty statisticians, as they often do not update information rationallyand therefore hold nonrational beliefs, they overweigh small probabilities andunderweigh large probabilities, and they fall prey to various framing effects (see,for example, Kahneman and Tversky 2000; Lindenberg 2008). Furthermore,people are often swayed by their emotions and evaluate options differentlydepending on whether they are in a "hot" or a "cold" state (see, for example,Vohs, Baumeister, and Loewenstein 2007). People make mistakes in predictingtheir future utility (for example, Loewenstein, O'Donoghue, and Rabin 2003)and are often too impatient and show weaknesses of will (Loewenstein, Read,and Baumeister 2003).

These are all important findings. They have been possible because exact modelpredictions as derived from formalized rational choice theories were confrontedwith appropriate data. These findings have stimulated extensive research in allsocial sciences. In economics they have helped to pave the way for "behavioraleconomics," which is the integration of psychological and sociological insightsinto economics. Behavioral economics is now branching out into the varioussubdisciplines of economics and transforming them by integrating psychologicaland sociological insights into rational choice frameworks (see Camerer,Loewenstein, and Rabin 2004; Diamond and Vartiainen 2007). This developmentwould not have been possible without the rational choice approach, which helpspinpoint the problems in standard theory.

Yet, as I will demonstrate by way of an example below, a rational choiceanalysis often makes surprisingly accurate predictions despite considerableviolations of some assumptions that underlie the theory. This point has alreadybeen made by Becker (1962). Such a viewpoint does not necessarily implydiscarding the importance of the empirical findings mentioned above, or amethodology that cares only for prediction accuracy and not so much for theempirical validity of the underlying assumptions. Quite the contrary, I willdiscuss evidence that points to the violation of the selfishness assumption asbeing a cause of a failure of prediction accuracy in several games of interest tosocial scientists. It is an empirical question how sensitive theoretical predictionsare to particular violations of underlying assumptions. The rational choiceapproach provides theoretical rigor in uncovering which violations matter forprediction accuracy and which not. Moreover, very often we are interested inthe comparative statics prediction of a model and, as we will see throughoutthis chapter, these predictions are often empirically validated.

Third, and somehow relatedly, equilibrium theory does not explain how aparticular equilibrium is actually reached. Yet evolutionary game theorists haveshown that equilibria may be reached through a process of trial and error orother adaptive processes (Gintis 2000a). Moreover, behavioral game theorists,who study actual human decision-making in strategic contexts, have devisedtheoretical and empirical learning models that help us understand and predicthow and when equilibria are approached under a given learning dynamic (seeCamerer 2003 for a comprehensive overview).

Fourth, equilibria need not be socially optimal, despite the fact that they arisefrom all players choosing their individually optimal action. The "tragedy of thecommons" or the famous prisoner's dilemma are prime examples. Moreover,even in games with multiple equilibria ("coordination games"), inefficientequilibria can result from individually optimal interaction, and players can evenbe fully aware that they are playing an inefficient equilibrium. In general, avery important task of modern social science is to understand inefficient socialoutcomes (Bowles 2003).

Fifth, the rational choice approach does not necessarily advocatemethodological individualism, although most defenders of rational choiceprobably believe that social phenomena should be explained solely from theactions of individuals. Gintis (2009) argues convincingly against this doctrine.In his view individuals' common understanding (coordinated beliefs) is crucialto understand many important social phenomena such as social norms, whichcoordinate the interaction of rational individuals.