You are on page 1of 23

Mind Association

Errors about Errors: Virtue Theory and Trait Attribution Author(s): Gopal Sreenivasan Source: Mind, New Series, Vol. 111, No. 441 (Jan., 2002), pp. 47-68 Published by: Oxford University Press on behalf of the Mind Association Stable URL: Accessed: 20/08/2010 11:44
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact

Oxford University Press and Mind Association are collaborating with JSTOR to digitize, preserve and extend access to Mind.

Errorsabout Errors: Virtue Theory and TraitAttribution
GOPAL SREENIVASAN This paper examinesthe implicationsof certainsocial psychologicalexperimentsfor moral theory-specifically, for virtue theory. GilbertHarmanand John Doris have recently argued that the empirical evidence offeredby 'situationism'demonstrates trait. I dispute this conclusion. My discusthat there is no such thing as a character sion focuses on the properinterpretationof the experimentaldata-the data themselves I grant for the sake of argument. I develop three criticisms of the anti-trait position. Of these, the centralcriticismconcernsthree respectsin which the experimental situations employed to test someone's charactertrait are inadequateto the task. First,they do not take account of the subject'sown construalof the situation. Second, they include behaviourthat is only marginallyrelevantto the trait in question. Third,they disregardthe normativecharacterof the responsesin which virtue theory is interested. Giventhese inadequaciesin situationism'soperationalizedconception of a 'charactertrait',I argue that situationism does not really address the traits',properlyunderstood. A fortiori, the proposition that people have 'character social psychologicalevidence does not refute that proposition. I also adduce some limited experimentalevidence in favourof charactertraits and distil two lessons we can neverthelesslearn from situationism.

Philosophical attention has recently been drawn again to a body of empirical researchin social psychology showing, among other things, that people are remarkably liable to be overhastyin their everydayattributions of character traits.' This particular liability is commonly referred to as the 'fundamental attribution error' (Ross and Nisbett 1991,p. 4); and the empirical researchwhich exposes it belongs to the 'situationist' tradition in social psychology. What some philosophers now suggest,in effect,is that situationism'sempiricalfindingscast serious doubt on the idea that there is even such a thing as a character trait,2 everyday attributions notwithstanding. Since philosophical accounts of the virtues-especially those inspiredby Aristotle-often
'I have Harman (1999)and Doris (1998,in press) particularly mind, but comparealso Blackin burn (1998,pp. 36-7) and Railton (1995,pp. 93-6). Essentiallythe same researchis also discussed by Flanagan(1991),whose conclusions I find more congenial.
2 Harman writes, for example, that it 'seems that ordinary attributions of charactertraits to people areoften deeplymisguidedand it may even be the case that thereis no such thing as charactraitsof the sort people think there are, none of the usual moralvirtues ter,no ordinarycharacter

Mind, Vol. 111 . 441 . January2002

? Oxford University Press 2002

pp.4What actuallyfollows standpoint from these experiments?For the most part. 327).standardswhich would be violated if next to no one actuallyturned out to have a charactertrait in the relevantsense.See. p. If the empiricalevidencemarshalledby situationismis well-founded. on a free-standingidentification of the nature of the empirical constraint that social psychology places on moral philosophy. pp.barepage references be to this textbook. as one moves from psychological debates between situationists and 'personologists' to philosophicaltheories of virtue. In so doing. Comparethe title of Harman (2000). a corrective lesson in situationism is equally held out for ordinary morality (Doris 1998. this is becausethe relevantsense of 'character what with the context. the further implication on offer is that many philosophical accounts of the virtues are beset with false empirical presuppositions (Harman 1999. 5 to a trait'shifts someUp point. Henceforth. 513-15).but less so: 'Toput things crudely. pp. But to that extent.Doris is also guarded. I conclude that the empirical constraint faced by a theory of virtue is instructive.people typicallylack character'(1998.Ross and Nisbett (1991. On the whole. To the extent that philosophy simply reflects common sense here. but Harman (1999) and Doris (1998) seem inwill debted to it as well.that is.p. and if its philosophical advocatesare correct.48 GopalSreenivasan analyse a virtue as a species of charactertrait. It even provides some positive evidence in favourof this assumption.situationists pointedly note that more carefulresearchhas neverthelessfailed to overturntheir basic findings. the situationists' empirical data have been received with considerable scepticism by traditional personality theorists. 506). however. However. 316). My enquirywill thereforefocus on the of properinterpretation the situationists'experimentalresultsfrom the of a philosophical theory of virtue. 4 My discussion of these experimentsis largelybased on the presentationin Ross and Nisbett (1991). But are they reallywarranted? That is the question I wish to pursue.p. but not particularly and vices'and that 'we must concludethat thereis no empiricalbasisfor the existenceof character traits'(1999. . for example. 505-9).Not only is theirs an accessible textbook. Doris 1998. as we shall see. I shall take the relevantempiricalfindings wholly for I granted. p.3 shall also assume that a theory of virtue should conform to certainstandardsof empiricalpsychologicaladequacy.the psychologicalevidence does not reallyengage the assumption one way or the other. 319.5at least not to date.the revisions requiredin our theory of virtue are striking indeed-as they are doubtless meant to be. I shall approachthis issue I constructively: shall concentrate. 105-109).and Mischeland Peake(1982). I shall argue that the social psychologicalevidence does very little to underminethe virtue-theoreticassumptionthat certainpeople actually have charactertraits in the relevantsense. these ordinary moral views rest on error' (Harman 1999. 3Within psychology. 'Character based virtue eth- ics may offer a reasonableaccount of ordinarymoral views.

and the cheating situation.the correlationbetween cheatingon a classroomtest last week and cheatingon one this week. concerns By contrast. one is correctingone's own test sheet in class and there is an opportunityto cheat. 317). after a fashion. Consider the following situations in which one might plausiblyassess someone's honesty: some change has been left on a table in an empty classroomand there is an opportunityto steal it. with respectto our cheatingsituation (in the case of honesty).Errors aboutErrors 49 difficult to satisfy. 6These situationsdescribesome of the . 509). The most careful of Doris's formulations is cast in terms of 'virtue' instead of 'character trait': 'if a person possesses a virtue. The question of what follows from all of this for common-sense moralityI shall have to leaveto one side. Let us use honesty as our organizing example of a character trait. occupy our attentionthroughout. 1. and not to cheat. the lying situation. We naturally imagine that an honest person would decline the opportunities presented in these situations.We suppose that honesty is or includesa reliable disposition not to steal. But we can take Harmanand Doris as our initial points of reference. p.6Call these respectivelythe stealingsituation.with respectboth to our cheatingsituation and to our stealingsituation (in the case of honesty). It will be useful to begin by explainingsome stock terms. seeminglywith impunity. Psychologists distinguish two dimensions along which someone's stadisposition to honesty may be reliable. not to lie. she will exhibitvirtue-relevant behaviorin a given virtue-relevantelicitingcondition with some markedly above chance probabilityp' (1998.the behaviouralmeasuresactuallyemployed in a landmark study of honesty in school children (Hartshorneand May. character traits 'are relatively long-term stable disposition[s] to act in distinctive ways' (1999. the dimension of cross-situational consistency the reliabilityof a character trait acrossdifferentspecificsituationsrelevant to the assessmentof that trait-for example.The dimension of temporal concerns the reliability of a character trait with respect to a bility specificsituation relevantto the assessmentof that trait-for example. As Harman has it. anotherchild is going to get in trouble and there is an opportunity to avertthis by lying. Cross-situational consistencymeasuresthe correlationbetween a set of behaviouralresponsesto one relevantsituationand anotherset of behaviouralresponsesto a second relevantsituation-for example. p. How character trait itself is to be understood will.1928). Temporal stability measures the correlation between one set of behavioural responses to the situation in question and another set of the same responsesto the same situation-for example.

'stealing'-are fixed or coded by the observer.Situationists do not dispute that people'sbehaviour exhibits considerabletemporal stability. was information on one occasion test key copyfroman answer on a general test froman answer on a similar sixmonths correlated withcopying . Consider. Finally. whether a temporal stability or a cross-situational consistency.Each is based upon observations of someone else's overt behaviour in a carefully controlled situation. higher. at the expense of variousforms of subjectiveassessmentfavouredby traditionaltheorists of personality-for example. self-reports.but also that he can be reliedupon not to steal or to lie either.for not simply tied to cheating situations. in principle. It is tied to cheating on a classroom test.we should acknowledge the importance.50 GopalSreenivasan correlationbetween cheating on a classroom test and stealing change left on a table. 1oo-log). The next point is that the debate about character traits primarily concerns the dimension of cross-situationalconsistency. 1ol) Temporal stability. as opposed to stealing or lying situations. of a distinction that I shall henceforth elide. peer evaluations. and personality assessment scales (pp. between measures thesamebetwo of coefficients-thecorrelation Stability much occasions-often exceed. Before introducing the fundamental attribution error. But the measuresare also 'objective'in the furthersense that the operative descriptions of the responses and the situations being studied-for example. But we ordinarily suppose not only that an honest person can be reliedupon not to cheat. as we shall soon see.the claim that Jezebelhas some temporallystable charactertrait.That is where the opposition from situationismwill begin.can alwaysbe distinguishedfrom a correspondingcharacter trait. for example.79 key later.To begin with. So the reliability of honesty as we ordinarily conceive it evidently extends into the dimension of cross-situational consistency. we should registerthree points about these measures of trait reliability. however. is narrowly tied to fairly specific situations: Hartshorneand May'scorrelationof.(p. on classroomtests or elsewhere. notice that both temporal stabilityand cross-situationalconsistency are objectivebehaviouralmeasures.79. This goes beyond the more basic claim that . It is one of the hallmarksof situationism to privilege objectivebehavioural measuresin the assessmentof charactertraits.40.sometimes havior different on reaching foundthatthetendency to Hartshorne May(1928) and Forexample. A suitable regularityin someone's behaviour.

In effect. given some suitable regularityin a person'sbehaviour. 326). the situationists'observations of various temporal stabilities in people'sbehaviourleave room for competing explanations.that the best explanationof it appealsto a corresponding trait.Errors aboutErrors 51 her behaviour exhibits a certain temporal stability. It denies the explanandum. 4).it may be helpful at least to illustratethe second half ('an underdeveloped situationism'. the fact that Jezebel'sbehaviour is temporallystablein a certainrespectdoes not sufficeto establish that she has the corresponding temporally stable trait. Similarly. the mainstay of situationism's critique consists in the claim that people's behaviour is not cross-situationallyconsistent. But nothing of substance rests on this assumption. togetherwith [a] failureto recognize the importance of situational factors in affecting behavior' (p.which both Harman (1999)and Ross and Nisbett (1991) adduce.In generalterms.p. In the experiment. this may be describedas an 'inflatedbelief in the importance of personalitytraits and dispositions. A good illustration is providedby the experimentthat Darleyand Batson (1973)modelled on the parableof the Good Samaritan. 509). Jezebel's possession of a stable psychological disposition that produces such behaviour (Harman 1999. While I shallbe emphasizingthe first half of this error('an overeagerdispositionism'). As we shall see. namely. The students also completed a questionnaire concerning the basis of their interest in religion. As they left for the other building. in themselves. and yet others .others were told they had just enough time to get there. 507-8). p. and not the explanans. some studentswere told to hurrybecausethey were late. I shallproceed to ignore it. pp. Darley and Batson had students from the Princeton Theological Seminary prepare a short talk that was to be recorded in another building.Harman and Doris actuallypart company here. as my aim is only to disarm a critique of character traits. Doris explicitly embraces temporallystabletraits (1998. Since there might be alternativeexplanations for it. as it also offers a particular explanation for that temporal stability. 130).without having alwaysto distinguishit. My simplifying assumption serves to isolate this denial.I make no pretenceof presentinga completepositive case for their existence. I shall simply assume.pp. 317-18. whereasHarmanunderscores the availabilityof non-trait explanations for temporally stable behaviour (1999.implicit in the assertion of a cross-situationally consistent trait. We are now ready to consider the fundamental attributionerror.p. I do so solely for the sake of simplicity. Having acknowledgedthis distinction.Doris 1998.

But only the degree of lateness-that is. More generally. This assumption derivesfrom the fact that charactertraits are supposed. pp.63 percentof those running early stopped to help. Turn now to cross-situational consistency. as did 45 percent of those running just on time.) Homer will not pocket the change next time he gets the chance either (Kunda and Nisbett 1986. in consideringthe parable. They attribute the (temporally stable) trait honesty to Homer. One mistakepeople are liable to make is to conclude that Homer is an honest person.To warrantpredictionsabout Homer of the order on which people typically make them. p. that is. 279). and take this trait to license very confident predictions that (e. and it is reflectedin the evidence base requiredto warrantan attribution. Imagine that one day we observe Homer declining to steal the change on the table in the empty classroom. a situational variable-turned out to be of any significance. each student encountered a man slumped in a doorway. simply on the basis of his honest behaviour on this one occasion.At bottom. A character trait. who coughed twice and groaned.we are apt not to recognize the extent to which the priest and the Levite may simply have been runninglate. what people apparentlyfail to appreciateis that the reliabilityof their predictionsdepends upon the numberof observationsunderlyingtheir trait attributions. to explain (at least some of) their possessor'sbehaviour. and degree of lateness. and recall our single observationof Homer'shonesty in the stealingsituation.g.52 GopalSreenivasan were told they would arrive a little early. The pair of mistakespeople make can be perspicuouslyrepresented as corresponding to the two dimensions of trait reliability we distinguished earlier. this derives from a pair of mistakes that people make in their assessment of the evidence base required to attribute a character trait to someone. religious being assumed here to be grounds for predicting how the person to whom the trait is attributed will behave in various particular situations (cf. The suggestionis that. whereas only 10 percent of those running late stopped to help.Taketemporal stability first. 209-11). as we have seen. it should be said explicitly. Flanagan 1991. En route.Perhapshe even reportsthe situation to the principal.A second mis- . More relevantfor our purposes is the overeagerdispositionism half of the fundamental attribution error. Which of the students stopped to help? There were three variables-content of assigned talk. an attribution of honesty to him would have to be based on evidence aggregatedover numerous observations.

. We seriously and routinely underestimate. as we might put it. and an ambiguous setting.) Homer will not cheat on the classroomtest next time he gets the chance. The first reservationI should like to enter is that.g.Errors aboutErrors 53 take people are liable to make is to predict.that is. and cheating situations. strictly speaking.g. of the observationsunderlyingthe relevant trait attribution. but it is cross-situationally consistent as well (Kundaand Nisbett 1986. This failureis nicely brought out in a study on attributionsof extroversion reportedby Ross and Nisbett. More generally. 209-11).pp. one would require evidence of his honesty that was distributed across stealing situations.Not only is the honesty traitthey attribute to Homer temporallystable.) 'the trait-consistent inferences should be stronger for the ambiguous setting when the behavioral information came from academic and social settings alike than when it was confined to a single type of setting' (p. (pp. 130). the new prediction suffersfrom the same evidentialweaknessas the old one. Subjects two academic were settings in then askedto predictthe target's behavior an academic a social setting. and their philosophical advocates concur. that is not how the subjects actually predicted. since it is equallybased on a single observation. 129-30) The study found that the subjects' predictions were completely insensitive to the situational specificity in their evidence also suffersfrom a furtherweakness. lying situations. Situationists allege that ordinary people-such as you and Icommit the fundamental attribution error left and right. then. both the number of observations and the distribution within them that is required to warrant the attribution of crosssituationally consistent character traits. Forsome of the subjects. Let us suppose that this is correct.that (e. andone socialsettingor viceversa. behavioral the information drawnfromjust was or threesocialsettings. Subjectswere given information about the behaviourof certaintargetpersons in three situations. while (e. was either subjects. on this same basis and also very confidently. In part. such as honesty. Thus..which is connected to people'sfailureto distinguish between temporalstabilityand cross-situationalconsistency. setting. 2. something else that people apparently fail to appreciate is that the reliability of their cross-situational predictions depends upon the distri- bution. either threeacademic For settings the behavioral information drawnfromboth contexts. the fundamental attribution error is irrelevant to the question of . To warrant the attribution of cross-situationally consistent honesty to Homer. so it is said. other one context.

A neat demonstration of this point is implicit in the material we have already reviewed. 734-35). The issue is whether there is. Likewise with character traits. From the fact that people happen to add badly. which I have simply granted for the sake of traitsare spoiled by argumenthere. warrant to attribute traits and not whether the trait attributions that people commonly make are actually warranted. however. It requires us to bracket one source of apparent evidence for the existence of traits.What I am sayingis that if our ordinaryviews about character that error. in fact. The predictive inefficacy of everyday trait attributions does not itself tell us anything either about the predictive efficacy of warranted trait attributions or about the prospects for acquiring the relevant warrants. It is therefore at best misleading of Doris. p. 509). 7Here I elide the distinctionbetween a suitableregularityin someone'sbehaviourand the correspondingtrait.7 In one study of conscientiousness. . even cross-situationally consistent ones. that the situationists themselves concede that there are temporally stable traits. 81 do not say that our ordinaryviews about characterare spoiled by the fundamentalattribution error. even if elsewhere people attribute temporally stable traits without having acquired a warrant. our ordinary views about character. to say that 'trait attribution does not ground confident predictions of particular behaviors' (1998. Of course. 329). p.65 (Mischel and Peake 1982. these views 'can be explained without supposing that there are such traits' (1999. pp. when considering whethercharactertraitsexist. do exist. One source of the fundamental error in people's attributions was their disregard of the need to aggregate the behavioural evidence over a sufficient number of observations. none of this is strictly evidence against the thesis that character traits. observed between 2 and 12 times each. This applied equally to their attribution of temporally stable traits. for example. in which there were 19 different behavioural measures. then we would have to bracket the evidence they appear to offer. for example. Recall.we would have to enter into the question of the validity and precise bearing of situationism'sempirical data. Still. namely.54 GopalSreenivasan whether anyone really has a character trait.To pronounce on the matter. the mean temporal stability coefficient after aggregation was . there is a sense in which the fundamental attribution error is indirectly relevant to the question of whether there are character traits. If we suppose that a trait has been attributed without warrant. So here there is a clear warrant to attribute various temporally stable traits.8 As Harman says. it will come as no surprise to learn that the predictions it licenses are frequently confounded. it does not follow that there are no sums. the ones subject to the error.

we actually have to put the fundamental attribution error as such to one side. Recall that a warrant to attribute cross-situationally consistent honesty to Homer requires that we collect behavioural evidence of his honesty that is distributed across the various situations.impulsivity. Say it is .. the average correlation 9In fairnessto Doris. Virtuallyno coefficients.. dependency. the averagecorrelationbetween differentbehavioralmeasures specifically designedto tap the same personalitytrait (for example.13. What we need to consider instead is whether a warrant to attribute cross-situationally consistent traits can be acquired. Once again. Here we come to the heart of the situationists' case. begin by considering the average correlation between (e. One assumption is that . the average correlation between stealing and lying was . I should say explicitlythat he adducesthis evidence also and gives it due emphasis.Cf. it can be shown that Hartshorne and May's data exclude any warrant for attributing cross-situationally consistent honesty to our friend Homer.eitherbetween individual pairsof behavioralmeasuresor betweenpersonalityscalescoresand individual behavioralmeasures.23. This implies that repeated observations of Homer's behaving honestly were recorded in at most one of the stealing and lying situations.30 'barrier'.p. To see this.between stealing and cheating .Errors aboutErrors 55 Hence. it will not be possible to collect the requisite evidence. 95) Since we have been using honesty as our central example. the average correlation between any given pair of the various behavioural measures of honesty they studied was . 326).exceededthe .31. and often was even lower .lo and .9 As Ross and Nisbett summarize it. and cheating situations.13. Overall. lying. Had repeated observations of his behaving honestly been recorded in both situations. to pursue our question. as we know it can for temporally stable traits (cf.honesty. . and between lying and cheating . including the stealing.20. note 7). Granted our assumptions.(p.23 is also the overall average correlation between any given pair of behavioural measures of honesty in Homer's own case. let me also report that.or the like) was typicallyin the rangebetween .13.) Homer's stealing behaviour and his lying behaviour. Harman (1999. in line with the first assumption.g. which consists in the fact that empirical investigation of cross-situational consistency coefficients has found them to be almost uniformly low. If we make two further assumptions. The other assumption is that Hartshorne and May's behavioural measures properly operationalize the character trait honesty. according to Hartshorne and May (1928). I shall simply suppose that the data are correct.

Since the argumentwe have just rehearsedactually delivers the conclusion at issue (at least in the case of honesty).in effect. The consistency coefficientof .23. Neither offersto explainhow the situationists'data lead to the conclusion that there are no cross-situationallyconsistent charactertraits (cf.let me start by entering a reservationbrieflyabout the first.It follows that the evidence of Homer'sbehaving honestly does not satisfythe requireddistributiveconstraint. for example.Hence the second assumption:that Hartshorne and May'sbehaviouralmeasuresproperlyoperationalizethe character trait honesty. however. they may well warrantattributingsome cross-situationally consistenttrait or other to Homer.If so. Of course. Repeatedobservationsof Homer'sbehavinghonestly must thereforehave been limited to a relativelysmall number of the various situations. is that no data warrant this attribution-that any such attribution would be false. we may conclude that consistent. Doris 1998.and these new data might yield ratherhigher averageconsistencycoefficients. Notice that something like the first assumptionis needed in orderto consistenttraitsto particudiscussthe attributionof cross-situationally lar individuals. We therefore need some guarantee. is an average over all the children whom Hartshorne and May studied. it follows that Hartshorne May'sdata do not warand rant attributing cross-situationallyconsistent honesty to Homer.there is no possibilitythat the trait we may thereby be warrantedin attributing to Homer is the trait honesty. the figures do . and likewise for the overall averagecoefficient of .this figure does not exclude there being some individuals for whom the correlation between the stealing and lying situations was much higher. I have reservationsabout each of its two further assumptions.pp.A similarconclusion applies to most pairs of Hartshorne and May'sbehavioural measures.56 Gopal Sreenivasan would obviouslyhavebeen higherthan .Yet. however. that Hartshorneand May'sdata are the only data that matter. By itself.That is becausethe resultsof the psychologicalresearch are reported at the level of a population aggregate. 506-7). given that in Homer'scase the averagecorrelationbetween any given pair of measuresis only . The conclusion we are looking for.13.grantedthe second assumption.23 for any pair of behaviouralmeasuresof honesty. While I shall concentrateon the second of these assumptions. Otherwiseput.New data about Homer might be collectedon the basis of somewhat differentbehavioural measures. no one'shonesty is cross-situationally I do not know whetherthis is the sort of argumentthat Harmanand Doris have in mind.If everyoneis in Homer'sposition. it may be instructiveto see where it goes wrong.13between lying and stealing.

(in " This may well be only a subset of the traitsstudied by psychologists. On the other hand. is to be understood. the core of Doris's formulation:a virtuous person 'will exhibitvirtue-relevantbehaviorin a given virtue-relevanteliciting condition' (1998. then its empirical presuppositions are not falsified unless it is reallytrue that next to no one's (virtue) traitsare cross-situationallyconsistent. or even next to none. 509). which These people are models of virtue and only some people need have. The scope of a theory of virtue presumablyvaries with the theory. if a theory of virtue only appliesin the firstinstance to 'some people'. If a theory of virtue does indeed take'most people' within its scope.23. This means that in order to articulate our understandingof a particulartrait.10 will only approximate them in varyingdegrees. In general terms. 511-12). may thus be regardedas an innocent could be arguedthat extroversion. 3. He has something to say against it. 10 Doris concedes this possibility (1998.That was one of my reasonsfor focusing on honesty. as a cross-situationally consistent character trait. I mention this reservationbecause.the presentationof the results reflects psychology's interest here in how most people behave. p." Recall. Yetthe furtherassumption necessaryto reachthis conclusion takesus well beyond the existing data. My reservationconcerns the transpositionof this interestto the case of a theory of virtue. I defend a theory of virtue along these lines elsewhere. we have to specify as which responsesare to count as 'honesty-relevant behaviour'. in my own view.In this way. 96-100) . such as honesty. My fundamental reservation concerns the assumption that Hartshorne and May'sbehavioural measures properly operationalize the character trait honesty. but I shall not pursue the matter here.for example.and dependency-all studied extensivelyby situationists (pp. The claim that there are no charactertraits. Since I cannot defend that second reservation. pp. .includordinarypeople view here.For example.talkativeness. then the possibilitiesleft open by the situationists'data need not be excluded. not fit this conception. I do not insist on the ing zero. in my Emotionand MoralJudgement preparation). the traits in which virtue theory is interestedmay be conceived as dispositions to respond in a distinctive way to a certain sort of situation. The assumption has the merit of forcing us explicitlyto confront the question of how honesty. well as to specifywhich situationsare to count as 'honesty-eliciting'situations. the correct theory of virtue is a theory of what Aristotle called full virtue.aboutErrors 57 Errors exclude there being many people whose own overall averageconsistency coefficientis much higher than .

Ross and Nisbett call it the principle of construal. Suppose. Certain behavioural measures. namely. Homer will be scored as inconsistent across the stealing and lying situations. To predict the we the behaviorof a given personsuccessfully. that Homer believes in 'finders keepers' and so does not consider pocketing some stray change to count as stealing. low consistency coefficients between such measures-is not always good evidence that the person's behaviour is actually inconsistent across the situations in question. it would seem that the subject's own specifications should be the ones to count. for example. Accordingly. (1983. Hartshorne and May count not taking the change as the 'honest' response in this situation. situationists themselves are at pains to emphasize what amounts to the same point. From the standpoint of the aspiration to predict a subject's behaviour. however. Moreover. but a statementabout a disagreementbetween a group of individuals and an investigatorover which situations and behaviors may properly be classifiedinto a common equivalenceclass. For their part. each of which cuts against the assumption under examination. say. To begin with. and describe it as one of the fundamental tenets of social psychology. as we have already noted. one might reasonably think. The impact of any 'objective'stimulus situation depends upon the personal and subjectivemeaning that the actor attachesto that situation. 214) Ironically. that [t]he traditionalinferenceof inconsistencyis not an inferenceaboutindividuals. 11) It follows. however. Then he will regard pocketing the change as perfectly consistent with his conventionally honest behaviour in. which are fairly severe (cf.58 GopalSreenivasan Three important issues intersect here. (p. It is simply to acknowledge the practical difficulties of meaningfully predicting individual behaviour. the cheating situation. pp. to accept this is decidedly not to sever the important link between warranted trait attribution and prediction. . Since the behavioural measures in their study are 'objective'. 163-68). it is their specifications of honesty which are used to score the subjects. It also depends on the extent to which the subject and the observer agree about the relevance of the trait specifications which the behavioural measures incorporate. Next there is the issue of degrees of relevance to the requirements of honesty. that failure to predict a person's behaviour on the basis of'objective' behavioural measures-that is. This brings out a point that has been well put by Daryl Bem.p. there is the issue of whose specification is to count. must be ableto appreciate actor'sconstrualof the situation.

it hardly seems a central or paradigmatic case of theft.ConsiderHartshorne and May'sstealing situation again. Yet recall the particular situation with which they pair this response:it is one in which the intention to mislead servesto achievea genuine good. not every apparentlyequivalent variation on our concrete specification will preserve its central relevance. strictly.with the temporalstabilitycoefficientof . and cheating are all centrallyrelevantcases of the sort of behaviourexcludedby honesty.Errors aboutErrors 59 will be of greaterrelevance to the assessment of honesty than others. the lying situation. Assume for the moment that it does suffice. Hartshorneand May'sbehavioural measure of lying clearly appearsto be a paradigm case. the reason not to lie present . and this quite independently of any boost that might result to Homer's averageconsistency coefficientbetween the two situations. That is because. If I do not believe in 'finderskeepers'. for example.A reasonable argument could thus be made to substitute a shoplifting scenario for the stray change pocketing scenario within the 'stealing'situation. Nevertheless.apparentlyequivalentvariationson a concretesituation may preserve its identity under some mechanical description without preserving the significance of the reason for action present within it. namely. Then. Finally. bearing in mind that the subjectsin this study were children. Rather. Evenif we have a concretesituation-responsepair that is centrally relevantto the assessmentof honesty. Shoplifting.I may well acceptthat pocketingstraychangeis. for example. the responsesin question are not meant to be responsessimply to the situation as such.they are meant to be responsesto some reasonfor action present within the situation.By contrast. There is still a question about the relevanceof the more concretesituation-response pairsused to exemplifythese cases. In that sense. that stealing.changing one's test answers as one corrects them from the answer key does seem like a fairly central case of cheating-this was the behavioural measure.recall.there is the issue of the normativesensitivityof the specification. Let us stipulate that 'knowinglycommunicating a falsehood with the deliberate intention to mislead' specifies a behavioural responsethat is a paradigmcase of lying. while leaving the cheating situation as it stands. in the case of the traits that interesta theory of virtue. preventing another child from getting into trouble. strikesme as a much clearercase. Suppose we accept. Considerfirst another of Hartshorneand May'ssituations. obviously enough. lying.79.stealingit. The question thereforeariseswhetherthis good sufficesto justify the communication of the falsehood. As a couple of examples should make clear.

70 percentof those in a hurry. make a significantdifferenceto the results. Half the new subjects were told that their performanceof the task awaiting them in the second building while the other half were told that theirswas 'not eswas 'of vital importance'to the experimenter. . But that does not reallymatter. for instance. So the value of Hartshorne and May'slying situation as a behavioural measure of consistency is spoiled eitherway. the effect would only be to transformthe case into one in which a subjectand the observermight well disagreeon this score. sential'.Batsonet al. I should have thought. it follows that lying to prevent another child from getting into trouble is not a paradigm case of lying after all. It does not follow that a failureto help someone in distressalwayscontra-indicates depends upon a comparison of variousfactors. It does not follow because. But since it articulates an eminently comprehensible view.p. Now consider a second example.stopped to help (versus80 percentof those neither in a hurrynor on their way to something important). Whatever the the right assessment.but for something of little importance. I thank PeterVranasfor this reference. indeed. We could rescind this assumption. I could be wrong. These conclusions have been reached under the assumption that lying can be justified (and is. (1978)found that the importanceof what the subject is hurryingfor did. If I am right.that it belongs to a paradigmcase of lying that the reason not to lie is there a decisive reason for action.12 point to notice here about their experiment is that its adequacyas a behaviouralmeasure of compassion dependson 12In a subsequentvarianton this experiment. the reasonto help someone in distresscan be defeated. here).like the reasonnot to lie. although it satisfiesthe stipulateddescription. Suppose we stipulate that 'helping someone in distress'specifiesa situation-responsepair that constitutes a paradigmcase of compassion.since Hartshorneand May'slying situation would still not be a good behaviouralmeasureof honesty. It seems to me.Wheneverthat reason is defeated.or alwaysindicatesinconsistencyin an otherwisecompassionate person. Of course.but still on their way to something important.60 GopalSreenivasan within 'the lying situation' is not a decisive reason.failureto help someone in distressis perfectlyconsistent (in that case) with the trait of compassion. a differentcharactertrait. unless we wanted to define a model of honesty as someone who behaves 'honestly' even when the balance of reasonsstandsagainstit.Naturally.stoppedto help (versus io percentof those hurryingfor something important). however. 50 percentof those not in a hurry.that the fact that one is in a hurry can defeatthe reasonto help someone in distress.such as the importanceof what one is hurryingfor and the nature of the victim's distress. 108). I do not know how these factors should be assessed in the particularcase of Darley and Batson'sGood Samaritanexperiment:they themselvesflag the possibility that 'conflict' explains some of the failures to stop (1973.

We should therefore restrict our attention to a certaingenus of character traits-traits that areboth responsiveto reasons and capableof properlylicensingpredictions. its power to highlight the predictive relevance of the degree of latenessdoes not depend on the statusof the reasonto help. Our previous discussion enables us to identify three genericrequirements behaviouralmeasuremust satisfyin order a to operationalize a character trait of this kind.Errors aboutErrors 61 the subject'sdegreeof latenessnot defeatingthe reasonto help. From a constructive perspective. By contrast. Others arise mitment to moral-theoreticaladequacy-its commitment to the character traits it describes being genuine virtues.We have been taking it as given all along that a character trait must secure a proper licence to predict behaviour. Let us begin with the empiricaldesiderata. As this assumption is a premiss of the argumentwe rehearsedto the conclusion that no one's honesty is cross-situationallyconsistent. (i) each behaviouralmeasuremust specify a responsethat represents a centralor paradigmcaseof what that trait requires. moreover. (ii) the concrete situation each specifies must not have any features that defeatthe reasonon account of which that trait requiresthe response in question. From a criticalperspective.each issue lays the ground for a principled objection to the relevanceof various of Hartshorne and May'sbehavioural measures to the nature of honesty as a consistenttrait. which havebeen the focus of our discussion.and . in a character trait that is not responsive to reasons. Let me summarize the importance of these three issues.A theory of virtue has no interest. For a given properly charactertrait. Some of these desiderata arise in connection with a theory of virtue'scommitment to empiricaladequacy-its commitment to there being people who actuallypossess the character traits in connection with a theory of virtue'scomit describes. We can do so in either of two ways. consideration of the same issues suggestscertaindesiderataour behaviouralmeasuresmust satisfyif we are to be able adequatelyto investigatethe cross-situationalconsistency of someone's honesty. It will be useful to considerthe two sorts of desiderataseparately.The cross-situationally consistent charactertraits described by any theory of virtue will belong to this genus. Togetherthese objectionsvitiate the cross-situationally assumption that Hartshorneand May'sbehaviouralmeasuresproperly operationalize the character trait honesty. we may therefore reject that argument.

I emphasize this because it brings out the crucial point that the requirementof agreementbetween subjectand observer in no way limits or conditions the moral substanceof a given virtue. that it does so in order to conform to the third generic requirement-perhaps.Not every charactertrait minimally eligible to be a virtue is in fact a virtue. These On the one hand.Whetherit is depends.Suppose it upholds that principle. The furtherrequirementson being a virtue arise in connection with a theadequacy. 11). on whether 'finderskeepers'is a correctprinciple. a theory of honestywill either uphold or repudiatethe principle of 'finderskeepers'. If pocketing straychange is stealing.62 GopalSreenivasan mustagreeon these characterizations and the observer (iii) the subject of the specifiedresponsesand situations. it may still be does not cease to be stealingwhen subjects disagree.for all that. 13 Of course. the agreementrequiredbetween subjectand observerhas to be a suitablylimited one: it has. indeed.Nor does the generic requirement(iii) imply otherwise. .I shall illusory of virtue'scommitment to moral-theoretical trate them briefly to put the generic requirements on behavioural measuresof cross-situational consistencyin some perspective.but not a sufficientcondition.among other things.13 situationiststhemselvesconcede (p. even though the generalidea behind it is clearenough in principle. there arevariousvirtue-specific requirements. then the charis actertraitit operationalizes not even minimallyeligibleto be a virtue. thereby departing from Hartshorne and May. Now there are variousways in which an empiricallyadequatetheory of virtue may neverthelessbe defective. In that case.behaviouralmeasures that fail requirement (iii) operationalize a trait that cannot properly as license predictions. have to do with how a given theory of virtue understandsthe specific nature of individualvirtues.and that is a separatequestion. As the phrase suggests. If a set of behaviouralmeasuresfails these requirements. behavioural measures that fail requirement (ii) do not operationalizethe responsivenessto reasons (or normative sensitivity) that manifestlybelongs to any virtue. everyonewho is cross-situationallyconsistent qualifiesas consistent only on the supposition that finders are keepers. as it were. minimum eligibility is a necessary. 163-68).in the sense of describing a charactertrait that people actually possess. For example. such as honesty. Yet. both to enablepredictionand yet to leave room for it at the same time. Suppose. For example. this theory of honesty may well turn out to be empiricallyadequate. The third requirementis thereforelikely to prove a somewhat delicatematterin practice(pp. so far as honesty is concerned. Likewise.

my Emotionand MoralJudgement preparation).Since these are independent desiderata.In fact. .do I think we have been given any reason to believe that it will be realized. however. So there is a sense. are genuine virtues)? The generic requirementsI have articulated aim to clarify what it takes to answer this question properly. We encountered elements of this influence in our discussion of normativesensitivity. and not to cheat. if my view.15 Situationismraisesthe more basic question of whether empiricaladequacycan be securedfor any theory of virtue at all: does anyone actually have a charactertrait some theory of virtue describes (quite apart from whether the traits people have.'4If they were. that no such theory of virtue has as yet actuallybeen falsified. A theory of virtue aspires. and this will influence the account it gives of 'individual'virtues. For my part.But that is not the possibility with which we have been primarilyconcerned. the possibilitydescribedin the text would sufficeto undermineany theory of virtue.let alone be defeated by them. on being a virtue.the requirements of honesty could not conflict with the requirementsof beneficence. What a theory of virtue presupposes.Nothing in the earlierdiscussion restsupon it. then. no person with a cross-situationallyconsistent character trait would have a virtue unless the behavioural measures acrosswhich her trait was consistent incorporatedthe requirementsof every virtue. 16 do not mean to suggest that a theory of virtue must recognize the virtue of honesty as we have been discussing it.16 theory of virtue committed to some such presupposition is vulnerable to empirical falsification. to both empirical adequacy and moral-theoreticaladequacy. as we might call them.Errors aboutErrors 63 On the other is at least possible that empirical adequacycan only be secured at the expense of moral-theoretical adequacy. I which the position taken by situationism'sphilosophical advocates is more radicalthan it needs to be. such as honesty. More to the present point: if the virtues were strongly unified. Still. is that there is some non-trivial number of people whose honesty is cross-situationally consistent across a range of behavioural measures that satisfy all three generic requirements. interestingly. For example. so far from fal141 defend this in presumption. not to lie. there are also various theory-specificrequirements. I doubt that there is a single charactertrait that respondsto the reasonsnot to steal.and likewise for such other virtues as it recogA nizes. These have to do with how a given theory of virtue understandsthe moral nature of virtue in general. for that matter.along with other theory-specificrequirements. (in 15 Nor.where my assumptions about certainconflictsbetween reasonsreflectedthe presumption that the virtues are not stronglyunified. a theory will take some position on the question of the unity of the virtues.

64 GopalSreenivasan sifying this presupposition. measures which furthermore exhibit no normative sensitivity and which may be drawn from the widest range of measures minimally associated with the trait by common sense. Nor should they be so read. I should like to emphasizethat there is no need to read my conclusions as implyingthat the psychologists'experimentswere ill-designed. and it is certainly contradictedby the empirical findings of Hartshorne and May. Neither does philosophy.Hartshorneand May'sstudy does not even put it to the test.they have alreadydone so. let us call this the personologist's conception of a character trait.1982). Mischel and Peake. To some extent.virtue theory has a variety of principled grounds for rejecting the personologist's conception of a character trait. the personologist's conception is nevertheless a perfectly legitimate target. On the assumption that this is a fair portrayalof personology. this is the conception that situationists impute to their opponents in the psychologicaldebate. 1982). that various situationists also affirm the point which motivates the first requirement(e.There is nothing to prevent psychologistsfrom affirmingthe three correspondinggeneric requirements on behavioural measures of cross-situational consistency.g. the traditionalpersonalitytheorists or personologists (pp.It turns out. Their experimentscan thus be read as refutingpersonology on its own terms. it is not refuted by refuting that conception.See note 11. It is perfectly possible to conceive of a 'charactertrait' as licensing predictions simply on the basis of 'objective' behavioural measures. however.17 4. While it is not endorsed by the situationists themselves (p. 90-94.I am afraidthat this crossing of purposes must be laid at the door of situationism's philosophicaladvocates. moreover. Indeed. Hence. 163). As we have seen. EarlierI mentioned that social psychology's principle of construal affirmsthe basic point which motivates the third requirement.18 17Asimilarconclusion appliesto the other studies adducedby Ross and Nisbett (1991). in addition. among others.though it bearsrepeatingthat the traitsthere in question are not reallyof the sort in which virtue theory is interested. There is accordinglya sense in which the confrontation between situationism traits'operatesat and a theory of virtue over the existenceof 'character cross-purposes. Mischeland Peake. It is also worth recognizingthat virtue theory has no specialclaim on the principled grounds we have observed for rejectingthe personologist'sconception. .

Mischel and Peake conclude that the perception of consistency in behaviour is rooted in the temporal stability of an individual's behaviour with respect to some bundle of measures-but not necessarily many-that are highly prototypical for a given trait (1982. Their results encourage virtue theory in its expectation that cross-situational consistency coefficients calculated on the basis of adequately specified behavioural measures would be rather higher than the coefficients we have encountered thus far (cf. pp. note 12). In a second study. rather than that of the individual concerned-he found that they then failedto producehis result. a particularwaiterwho 'didn'tdo also reinforcesthe impressionthat situationism's purposeswith a theory of virtue.. I shall call attention to two studies in particular.Errors aboutErrors 65 The present relevance of this ecumenical obeisance lies in some empirical work undertaken by these fellow-travelling psychologists.meaning that '[w]ith families.71). [w]ith adolescentson dates. as a group. for example. Charles Lord (1983) found that an individual's cross-situational consistency in conscientiousness was significantly higher when the pair of situations (behavioural measures of conscientiousness) in question was regarded as similar by the individual himself or herself.he was haughtyand intimidating'.'9 This result supports the expectation generated by our third generic requirement. Simply asking the subject how similar various pairs of situations seemed. . 750-52).he was warm and homey . that is. nor is cross-situational consistency reported as varying significantly with self-perceived consistency (regardless of the prototypicality of the behavioural measures). than did individuals who perceived themselves as less consistent in their conscientiousness (. They remark on 19 Interestingly. 18 There is even some recognition of the issue of normativesensitivity. to which I referred previously.the result depended upon the method used to assess the subject'sview of the similarity between situations.20 They found that individuals who perceived themselves as highly consistent in their conscientiousness had much higher mean temporal stability coefficients. on the more prototypical measures (. as rated by the subject population as a whole. 164).47)..Ross and Nisbett relate an anecdote from a study concerned with waiters' strategies for maximizing tips. While this recognizesthe broaderphenomenon of teleological sense of'inconsistency'lies at crosssensitivity.did not produce the result. Mischel and Peake (1982) divided their behavioural measures of conscientiousness according to whether the measures were more prototypical or less prototypical of the trait conscientiousness. 20 Note that when Lord (1983)employedhis previouslysuccessfulmethods of similarityassessment at this same level-the aggregate level.and so on (p. both of the trait conscientiousness. In one study.exceptseek to maximizetips'. The stability coefficients of the two groups did not differ significantly on the less prototypical measures.

Mischel and Peake'sself-perceivedhigh consistency subjectsalso had high cross-situational consistency coefficients for conscientiousness. They constitute the empirical constraint under which moral philosophy labours here. It is merelya vestigeof the personologist'sconception.the behaviourof an actual model-even a reliableone-will be consistent with his or her paradigm performances. p. Mischel 1990.It follows that. at least for subjects who perceive themselves as highly consistent in the relevant respect. pp.21 Moreover.66 GopalSreenivasan This result supportsthe conjecturethat adheringto our first generic requirementwould boost averageconsistency coefficients. It cannot be assumedthat. they will be considerablynarrowerthan the rangeof behaviouralspecificationspermissiblyassociatedwith a given trait by common sense (cf.there will be nothing post hoc about that bundle. it is an empirical ques- tion how wide the widest bundle of paradigmatic behaviouralmeasures is that is consistentlyinstantiatedby an actualperson. 131). what the relation is between this range of paradigmaticbehavioural measures and all the other behavioural measuresproperlyassociated with the same virtue. cross-situationally consistent charactertraits will be narrowerthan imagined.Anyone who has high (positive) temporal stabilitycoefficients on each of two behaviouralmeasuresof a given trait will perforcehave a high cross-situationalconsistency coefficient as between those same measures. Second. in view of our third generic requirement. They are also very much in line with the conclusions reachedby Owen Flanagan (1991. Harman (1999. question. 326) criticizes Flanagan for not citing Mischeland Peakedo not themselvesdrawthis conclusion. this wider rangehas no specialauthority(cf. within the rangeof the particularhighly prototypical measuresthat made up their respectiveindividual 'traitbundles'.and it is another empirical question how many individuals exhibit the trait with that These persons will turn out to be our models of the relevant range. p.22 virtue. However.the models for a given virtue will have to be nominated relativeto some 'normalbackground'range of situations. For any given virtue trait. and a neglected one. but then they do not focus on the restrictedrange given by a subject'strait bundle. who discussed this literature a decade ago. note 20). I believe these lessons are both instructive and readily assimilated. 22It is a theoretical 21 . Certainly. let me simply state two lessons that a theory of virtue can learn from situationism.if the measuresin a subject'strait bundle correspond to his or her own ratings of high prototypicality. in literallyany novel situation.their consistency coefficientsare calculated over the wider rangegiven by all the measuresof conscientiousnessratedas more prototypical by their subject population. In closing. 279-81 and 290-92). Rather.How to specify this restrictionis a task to be dischargedby particulartheories of virtue. First.

and two anonymous referees.1983: on ations'. Or. Blackburn. as we have seen. John 1998: 'Persons. to be precise. pp.I am especiallyindebted to John Doris and GilbertHarman for instructive exchanges. PeterVranas. Daniel Batson 1973:"'FromJerusalemto Jeriin cho":A Studyof Situationaland DispositionalVariables Helping Behavior'. Oxford: Oxford University Press. and Virtue Ethics'. Situations. . 108. John M. I am gratefulto Paul Benacerraf.MarshallF Biederman. dence to cite in favourof cross-situationally properlyconceived (cf. pp. Moreover. and C. 23 For helpful comments.JamesL. also note 7). Building10. 'Toward ResponseStyleTheory of Personsin Situa Bem. 1982. USA gsreenivasan@nih. Page (ed) Nebraska Symposium Motivation. -in press:Lackof Character: Personalityand MoralBehavior.Errors aboutErrors 67 empirical evidence in support of these conclusions. Ryan. 32. Monte M. Daryl J. 504-530.23 Department of Clinical Bioethics GOPALSREENIVASAN Warren MagnusonClinicalCenter G. 97-101. PamelaJ. 100- Doris. there is some empirical eviconsistentcharactertraits. Room 1C118 NationalInstitutesof Health MD 20892-1156 Bethesda.SarahBuss. evidence which is in need of a rebuttal. This gives the impression that some evidence alreadystands against them. Cochran. and Bruce Vogt 1978:'Failure to Help When in a Hurry: Callousness or Conflict?'Personalityand Social Psychology Bulletin. that impression does not withstand scrutiny. Yet. C. Lincoln: University of Nebraska Press.An earlierversion of this paper was given as a philosophy colloquium at Washington Universityin St. pp. volume 30. Journal of Personality and Social Psychology. 4.there is evidence that there wouldbe evidence for their existence. Blosser.Cambridge:CambridgeUniversityPress. 27. I thankthe audiencefor a stimulatingdiscussion.if only we looked in the right place. Simon 1998:Ruling Passions. Darley. Maurice J.Nous. References Batson.

supp. Gopal in preparation: . 730-755. Ross.Cambridge. 195-224. Gilbert 1999: 'Moral Philosophy Meets Social Psychology: Virtue Ethics and the FundamentalAttribution Error'. of Perspectives SocialPsychology.Proceedingsof the Aristotelian Society. Consistency'. Searchfor Cross-Situational Review. May 1928: Studies in the Nature of Macmillan. 100. pp. Nisbett 1991:The Person and the Situation: New York: McGraw-Hill. 223-226.New York: Ziva and RichardE. day life' Cognitive Lord. Peake 1982: 'Beyond Deja Vu in the Mischel. 1076-1088. and MoralJudgement. --2ooo000: 'The Nonexistence of CharacterTraits'. Lee and Richard E. vol.89. Psychological Pp. Emotion Sreenivasan. Hartshorne. A View afterThree Decades'.MA: HarvardUniversityPress. Theory of Personality: Walter and Philip K. pp. 79-106. and Research. 1982: 'Predicting Behavioral Consistency From an Individual'sPerception of Situational Similarities'. Hugh and Mark A.42. Nisbett 1986:'The psychometricsof everyKunda. 1. 18. Journalof Personality and Social Psychology. Psychology. Studiesin Deceit. 315-331.CanadianJournalof Philosophy. Mischel. Railton.Owen 1991: Varietiesof Moral Personality. pp. 99. pp. pp. Proceedings of the Aristotelian Society. 21. Peter1995:'Made in the Shade:Moral Compatibilismand the Aims of Moral Theory'. Charles G. Walter1990:'PersonalityDispositions Revisitedand Revised: A.68 GopalSreenivasan Lawrence Pervin (ed) Handbook New York: GuilfordPress. vol. Harman. Character.