Lay Epistemo-Logic—Process and Contents: Another Look at Attribution Theory A Theory of Lay...

Psychological Review1980, Vol. 87, No. 1, 70-87

Lay Epistemo-Logic—Process and Contents:Another Look at Attribution Theory

Arie W. KruglanskiTel-Aviv University, Ramat-Aviv, Israel

A theory of the lay epistemic process is outlined. According to this theoryknowledge-seeking behavior is initiated by a purpose that a person has forreaching a given inference and consists of the stages of problem formulationand of problem resolution, the latter stage being governed by the principle oflogical consistency. Major attributional formulations are interpreted within thelay epistemic framework. It is concluded that such formulations have typicallyaddressed particular instances of epistemic (inferential) behavior rather thanthe underlying epistemic process. In this sense, the attributional formulationsmay be considered special cases of the present model applied to specific con-tents of knowledge. The present lay epistemic paradigm thus provides an inte-grative framework that allows us to consider diverse attributional models incommon theoretical terms and to derive the necessary applicability conditionsof different such models.

The present article features a theory of theprocess of knowledge acquisition that pro-vides an integrative paradigm for major at-tributional models proposed in recent yearsby social psychologists (e.g., Jones & Davis,1965; Kelley, 1967, 1971, 1972; Weiner etal., 1971). In what follows, the theory is firstset forth. Subsequently, central attributionalnotions are reinterpreted in present theoreticalterms. It is suggested that different attribu-tional formulations address different contentsof knowledge but share the same epistemicprocess explicated herein. Thus the presentanalysis reduces numerous disparate notionsin the field of attribution to a few underlyingconstructs in which terms human epistemicbehavior may be understood.

A Theory of Lay Epistemology *

The Epistemic Process

Problem Formulation

The present theory deals with the way inwhich individuals go about acquiring knowl-

I am indebted to Icek Ajzen for comments on anearlier draft.

Requests for reprints should be sent to Arie W.Kruglanski, Department of Psychology, Tel-AvivUniversity, Ramat-Aviv, Tel-Aviv, Israel.

edge of themselves and the surrounding world.Knowledge-seeking behavior is assumed tofollow a certain general sequence. This ispresently referred to as the epistemic process.Essentially, the epistemic process is assumedto consist of two major stages: the stage ofproblem formulation and that of problem res-olution. The epistemic problem is conceivedof as a set of mutually exclusive propositionswhose validity a knower might wish to assess.Henceforth, I call those the problem's con-stitutive propositions. Obviously, only oneamong mutually exclusive propositions maybe valid. The proposition finally adjudged asvalid would be the problem's resolution, thatis, a new knowledge or inference on sometopic. For example, a knower might wonderwhether a given event (say, Paul's enjoymentof a movie) was exclusively caused by (a)the movie's unique properties, (b) Paul'sunique personality, or (c) Paul's modality ofinteraction with the movie (e.g., viewing it onTV). A resolution of the problem in favor ofone of these propositions would constitute acausal inference that the knower may reach.

1 Conceptual precursors to the ideas elaboratedherein may be found in Kruglanski, Hamel, Maides,and Schwartz (1978), and in Kruglanski and Bar-Tal (Note 1).

Copyright 1980 by the American Psychological Association, Inc. 0033-295X/80/8701-0070$00.7S

70

LAY EPISTEMO-LOGIC 71

Not all epistemic problems refer to causalquestions. Thus, one might formulate a prob-lem regarding the direction in which a givenspot might lie (to the right, the left, or di-rectly ahead), the time at which a given oc-currence took place (at 7:00 or 7:15 p.m.),and, in general, problems concerned with allpossible types of propositions including thoseabout one's self (cf. Bern, 1972).

It is noted in passing that within the pres-ent framework the somewhat formal termproposition refers to the content of a thought,a belief, or a feeling. (For a similar usage, seeMcGuire, 1968.) For example, we may meanroughly the same thing when we say thatJohn thinks he is in love, believes that he is,or feels that way. In all three cases the refer-ence is to the content of the same proposition.The notion that feelings may be conceived ofas inferences similar in kind to other thoughtsor beliefs was stressed in the pioneering workof Schachter and Singer (1962) and was de-veloped further by Nisbett and Valins (1971).

In formulating the epistemic problem, theindividual is assumed to draw on his/herstore of available knowledge to generate plau-sible alternative propositions on some topic.Various psychological processes may play apart at this stage, in particular long- andshort-term memory, momentary saliences ofstimuli, and the knower's attentions. Althoughundoubtedly important, those various pro-cesses are outside the scope of the presentwork and will not be dealt with further. Whatwill be stressed, however, is that the formula-tion of an epistemic problem may be under-stood as a motivated behavior, prompted byan interest that a knower may have in agiven bit of knowledge. Such interest couldbe intrinsic such as when a child, out of sheercuriosity, attempts to divine the workings ofa TV set. Alternatively, the interest could beextrinsic such as that of a traveler seeking tofind out what time it is in order to catch aplane.

It follows that the epistemic problem for-mulated by the knower has to be ideologicallyfunctional2; that is, the knower assumes thatthe problem's resolution will serve some (in-trinsic or extrinsic) end of importance to thisindividual. Data relevant to the foregoing

proposition were reported by Kruglanski,Hamel, Maides, and Schwartz (1978). Sub-jects in an experiment were presented withone of two different objectives and were askedto state which of two epistemic problems theypreferred to resolve. The problems were soconstructed that one appeared to advance(was teleologically relevant to) one of theobjectives, and the second, the remaining ob-jective. For example, some subjects were pre-sented with the objective of deciding "whetherto invite John to a Saturday-night party,"and others were presented with the objectiveof deciding "whether it is worthwhile to buytickets to a given movie for oneself and afriend." All subjects then stated their choicebetween obtaining a resolution to (a) theproblem of whether John's decision to attenda movie on Saturday night was caused by Msunique personality or by the movie's proper-ties (i.e., a problem cast in the "person"-"stimulus" categories of Kelley's, 1967, anal-ysis of variance) and (b) the problem ofwhether John's decision to attend the moviewas an end in itself or a means of combattingloneliness (categories reminiscent of Krug-lanski's, 197S, endogenous-exogenous frame-work).

Note that the person-stimulus problem isteleologically functional to the objective ofdeciding about the movie tickets, for if John'sacquisition thereof had been prompted by themovie's attractive properties (stimulus attri-bution), the film might be worth viewing byoneself as well. On the other hand, the means-end problem is teleologically functional to theobjective of deciding about the party invita-tion, for if John's contemplated visit to themovies were a mere means of combattingloneliness, he might well attend the party in-

2 Note that the present notion of teleologicalfunctionality is unrelated to the issue of teleologicalexplanation and its place in naive epistemology. (Fora recent debate of the latter issue, see Buss, 1979;Harvey & Tucker, 1979; and Kruglanski, 1979.) Aspresently defined, teleological functionality refers tothe relation between the epistemic problem and theknower's end. It is independent of the type of epis-temic problem being dealt with, for example, com-posed of teleological or causal propositions, and soon.

72 ARIE W. KRUGLANSKI

stead. The results of the experiment lendstrong support to the notion of Ideologicalfunctionality. Across diverse hypotheticalsituations subjects exhibited significant andconsistent preference for resolutions to teleo-logically functional versus nonfunctional prob-lems, that is, to problems whose resolutionsappeared to advance the subjects' experi-mentally induced objectives.

Later it will be shown that distinct attribu-tional formulations address distinct episte-mic problems. The foregoing discussion sug-gests, therefore, that any given formulationshould apply to an individual's epistemicactivities only when the problem addressed bythe formulation is Ideologically functional tothis person.

Problem Resolution

Consider a person who, for some reason,formulated the epistemic problem of whetherthe coffee he/she just drank at a dinner partywas Brazilian or Colombian. One way of find-ing this out might be to ask the hostess andtrust her reply. If she asserted that the coffeewas Brazilian, the problem would be resolvedin favor of the Brazilian hypothesis and viceversa if she asserted it to be Colombian. Thissimple instance of inference is now treatedas prototypical in the following probe of theunderlying logic employed in the resolutionof all epistemic problems. First note that theknower likely considers the hostess's reportas relevant to the problem at hand. Given theknower's subjective world of meanings andconcepts (shared to a large extent by othermembers of the same cultural community),the hostess's report seems to bear on the prob-lem in some direct way, whereas other infor-mation (say, the weather report) may notappear to have such a bearing. That peopleprocess information in accordance with itsrelevance to their intuitive hypotheses andtheories has been recently stressed by Ajzen(1977) and has been an important theme inthe works of Asch (1946), Heider (19S-8),and Kelly (1970).

But the knower in our example is likely toview the hostess's report not only as vaguelyrelevant to the two alternative hypotheses

being evaluated but also as consistent withone such hypothesis and inconsistent with theother. For example, if the hostess assertedthat the coffee was Brazilian, this is consistentwith the Brazilian hypothesis and inconsistentwith the Colombian hypothesis. Therefore,the hostess's report may assist our knower indeciding in favor of the former and againstthe latter hypothesis. We may thus summarizeby saying that in resolving the epistemicproblem, the knower deduces implicationsfrom the alternative propositions as he/sheintuitively understands them and seeks evi-dence relevant to, that is, consistent or incon-sistent with, those implications.

A further theoretical point must now bemade. In our example evidence consistentwith a given proposition assisted the knowerin problem resolution. But not all consistentevidence is likely to be of such assistance. Inparticular, evidence may at the same time beconsistent with several alternative proposi-tions and thus furnish a poor basis for dis-criminating among them. For example, thesymptoms of arousal and insomnia would beconsistent both with having drunk Brazilianand Colombian coffee. Evidence regardingsuch symptoms is likely to be of little help toour knower, precisely because it does not dis-criminate among the competing propositionsthat he/she may have formulated. By con-trast, the hostess's aforementioned assertionthat the coffee was Brazilian would be con-sistent with the Brazilian hypothesis and atthe same time inconsistent with the Colombianone. We can say that this assertion is non-commonly deducible from (or implied by)the two competing hypotheses considered bythe knower and so it sets them apart. Let usnow define more formally some of the centralnotions employed in the above analysis, inparticular the notions of relevance, consist-ency, and noncommon deducibility.

Definition 1. Two cognitive elements aremutually relevant if and only if either is be-lieved to entail the other, or the negation ofthe other. By entailment is presently meantthe "if-then" relation. For example, onemight surmise that if "it is five o'clock," then"a reliable watch would indicate five"; thiswould render mutually relevant the two cog-


nitions just considered. On the other hand,the "five o'clock" cognition may be believednot to entail anything in particular about theweather conditions in Malta, so the cognitions"it is five o'clock" and "it is raining in Malta"would be irrelevant to each other.

Definition 2a. Two relevant propositionsare consistent if and only if their conjunctionis compatible with (i.e., is not negated by)their relevance relation. Thus, if it being 5:00p.m. entails that a good watch would indicatethis, the cognitions "it is five o'clock" and "areliable watch is indicating five" are mutuallyconsistent.

Definition 2b. Two mutually relevantpropositions are inconsistent if and only if thenegation of their conjunction is entailed bytheir relevance relation. Thus, if it being fiveo'clock entails that "a good watch would notindicate a quarter of six," the propositions "itis five o'clock" and "a good watch is indicat-ing a quarter of six" would be inconsistent.

This discussion suggests that propositionsare tested by examining their consistency orinconsistency with evidence. An item of evi-dence is useful to the extent that it distin-guishes between alternative propositions or isnoncommonly deducible therefrom as definednext.

Definition 3. An implication is noncom-monly deducible s from a set of propositions ifthe state of affairs it affirms is consistent withsome of the propositions and inconsistent withthe remaining propositions. By contrast, acommonly deducible implication is one thataffirms a state of affairs consistent with allthe alternative propositions. As we have seenin the earlier example, a commonly deducibleimplication is not of much use to a personattempting to choose among competing prop-ositions.

Earlier it was suggested that the epistemicproblem formulated by the knower is likely tobe teleologically functional; that is, the prob-lem's resolution would be hoped to contributeto the knower's ends in some way. We maynow identify yet another functionality re-quirement, this time to be met by evidence inwhich a knower would be interested. Suchevidence would have to be validationally func-tional; that is, it would have to be adjudged

as contributing to the resolution of the episte-mic problem by being noncommonly deduciblefrom the problem's constitutive propositions.

Noncommon Deducibility and DiverseContents of Knowledge

The noncommon deducibility logic is con-ceived of as an aspect of the epistemic processapplicable to all types of knowledge regardlessof contents. Let us illustrate the working ofthis logic in reference to propositions whosecontent is causal and noncausal, respectively.Turning first to the validation of causal prop-ositions, consider a tennis player attemptingto decide whether the effective corner shotthat he/she just executed is causally attribut-able to a correct flip of the wrist or to thebending of the knees. Now one implication ofthe concept of causality (as intuitivelygrasped by most people) is covariation of thecause and the effect. In our example, covaria-tion of the effective shot with wrist flippingwould, therefore, be consistent with the prop-osition that flipping was the shot's cause andinconsistent with the proposition that kneebending was the exclusive cause. Thus, thecovariation just mentioned is noncommonlydeducible from the alternative propositionsconsidered by our knower and may well facil-itate a choice among them.

Remaining with causal propositions, notethat covariation evidence need not be syn-onymous with noncommon deducibility. Letus revisit our scientifically minded tennisplayer attempting now to resolve a new prob-lem, notably whether the wrist flipping caused,versus was caused by, the well-executed stroke

3 Similarity should be noted between the presentnotion of noncommon deducibility and the Bayesiannotion of diagnosticity (Ajzen & Fishbein, 1975). Ac-cording to the latter notion, a datum is diagnosticof a hypothesis to the extent that its occurrence ismore strongly entailed by this hypothesis than by thehypothesis's negation. This is called the "likelihoodratio" and is the ratio of the conditional probabil-ities P(D/H) (probability of datum D under hy-pothesis H) and P(D/H) (probability of D underthe negation of H). In other words, a datum wouldbe diagnostic to the extent that it was noncommonlydeducible from the H and the not H hypotheses.


just accomplished. In this particular case co-variation (of the stroke with wrist flipping)would be consistent with both alternativepropositions among which choice is desired.In present terms, such covariation would becommonly deducible from these propositions.Hence, such evidence would not afford thebasis for deciding among the propositions. Inthe case at hand, one may well resort to evi-dence regarding the temporal order of flipversus the stroke. This would be noncom-monly deducible from the alternative prop-ositions being dealt with: Priority of the flipis consistent with the flip's causality and in-consistent with the stroke's causality a,nd viceversa.

So far, we have seen how a knower may usethe noncommon deducibility logic with prop-ositions whose content is causal. In somecases this particular logic may lead to interestin covariation evidence, whereas in othercases (as in the preceding example) the non-common deducibility logic would not impli-cate an interest in covariation. No particularinterest in covariation is expected either, whenthe epistemic problem involves a choiceamong noncausal propositions. Covariation isimplied by the notion of causality (as under-stood by most individuals), but is not so im-plied by numerous alternative notions inwhich a knower might be interested. Thuswhen interested in deciding whether "Acaused X" or "B caused X," one might in-deed look for a possible covariation of A orB with X. However, when attempting to de-cide whether A is "taller or shorter" than X,or whether X is "in love" with A or with B,one is not likely to be interested in the co-variation (conjunctive presence-absence of Aor B with X), but rather in evidence noncom-monly deducible from contents of the partic-ular noncausal propositions among which achoice is desired, notably, propositions assert-ing the "taller" versus "shorter than" rela-tion, or the "in love" with A versus "in love"with B relation. For example, the "tallerthan" concept may imply to most people ahigher reading to be yielded by a measuringrod; the "in love" relation may imply to somepeople a dreamy expression, a loss of appetite,and so on.

The foregoing analysis suggests that non-commonly deducible evidence would be ofgreater value to a knower than (a) evidenceirrelevant to the alternative propositions be-ing chosen from or (b) evidence that is com-monly deducible from those propositions. Databearing on those derivations were reported byKruglanski, Hamel, Maides, and Schwartz(1978). In one experimental study, subjectswere presented with one of two epistemicproblems concerning the causal explanationof an event. In one case the problem waswhether the "person" or the "stimulus" wasthe cause of some event (as in Kelley, 1967;McArthur, 1972). In the second case the prob-lem was alternatively (a) whether the eventwas an end in itself or a means to another end(as in Kruglanski, 1975), (b) whether it wascaused by the actor's ability or effort (as inWeiner et al., 1971), or (c) whether it wasintentional or unintentional (as in Weiner,1974).

In each case, the subject was required tostate which of two sets of information wasmore useful in resolving the person's problem.One set contained information regarding thepresence or absence of covariation of theeffect with the person and with the stimulus.This information was noncommonly deduciblefrom the person-stimulus problem; for ex-ample, contained data consistent with theproposition that the person was the cause ofthe effect and inconsistent with the proposi-tion that the stimulus was the cause of theeffect. This informational set was largelyirrelevant to the alternative epistemic prob-lems (based on the means-end, ability-effort,or intentionality-unintentionality categories).On the other hand, the second informationalset was noncommonly deducible from thealternative epistemic problem given to theperson and was largely irrelevant to (i.e.,nondeducible from) the person-stimulus prob-lem. A manipulation check verified that thesubjects perceived the appropriate covaria-tional set as significantly more relevant to theperson-stimulus versus the alternative prob-lem and the alternative set as more relevantto the latter versus the former problem.

For example, one event presented to sub-jects was that "John laughed at the come-


dian." Subjects given the person-stimulusproblem were asked to decide whether "some-thing about John probably caused him tolaugh at the comedian" or "something aboutthe comedian probably caused John to laughat him." Other subjects were given the"means-end" problem constructed around thesame event. They were asked to decide, forexample, whether "something about thecomedian's jokes probably caused John tolaugh at him" or "something about the come-dian's high status caused John to laugh." Ineach case the subjects were required to chooseamong two sets of information according totheir ability for resolving the epistemic prob-lem. One set (noncommonly deducible fromthe person-stimulus problem) might havebeen that "almost everyone who hears thecomedian laughs at him," implying high con-sensus, that is, low covariation between thespecific person (John) and the laughing re-sponse and that "John hardly ever laughs atany other comedian," implying high covaria-tion between the specific stimulus and theresponse. The second set (noncommonly de-ducible from the means-ends categories)might have been that "John does not usuallyenjoy humor," "the comedian is John's super-visor at work," and "John is afraid of losinghis job." Across a variety of problems andinformational patterns, subjects significantlyrated as more useful the noncommonly de-ducible as opposed to the nondeducible (orirrelevant) information.

Another pertinent experiment tested theidea that noncommonly deducible evidencewould also be preferred over evidence that iscommonly deducible from the problem's con-stitutive propositions. Subjects were presentedwith information about different events andabout the causal ascriptions of those events byattributors. In one condition (M for modal-ity), additional knowledge was injected thatmodality of the attributor's interaction withthe event might have biased this person'scausal attribution. In another condition (P,for person), the additional information wasthat the attributor had a vested interest inthe attribution rendered, thus casting doubton his/her objectivity. In the third condition(T for time), it was implied that the attribu-

tion was rendered on a particularly disruptiveoccasion so that this historical factor mayhave biased the attribution.

Thus, in the M, P, and T conditions of thisresearch, the respective categories of modal-ity, person, and time provided the basis forrival alternatives to the proposition that theattribution presented to the subject was ar-rived at in an unbiased way. More specifically,in Condition M the subject's epistemic prob-lem was to decide whether the attribution was(a) unbiased versus (b) biased by a faultyinteractional modality; in the P condition thedecision was between a versus c, possible biasdue to the attributor's personal circumstances,and in Condition T, between Proposition aand d, of possible bias due to time-contingentfactors.

In each condition the subjects replied tospecific questions that inquired whether be-fore accepting the attribution it would be de-sirable to have additional information regard-ing (a) consensus in the form of attributionmade by another person, (b) consistencyacross modalities in the form of attributionreached via a different modality, or (c) con-sistency across times in the form of attribu-tion reached on a different occasion. Notethat an answer to each of the above questionsis noncommonly deducible from one of theepistemic problems constructed about a givenevent (e.g., the problem in the M condition),whereas answers to the remaining questionsare commonly deducible from this epistemicproblem. Thus, evidence about consistencyacross modalities is noncommonly deduciblefrom the epistemic problem in the M condi-tion in which the knower's choice is betweenbelieving that the attribution was unbiasedand being biased due to a faulty modality.Specifically, constancy of the effect acrossmodalities is consistent with the hypothesis ofunbiased attribution and inconsistent with thehypothesis of defective modality. Hence, thispattern of data should augment the knower'sconfidence in the former as opposed to thelatter hypothesis. By contrast, evidence aboutconsensus, for example (consistency acrosspersons), is commonly deducible from theproblem's constitutive propositions in the Mcondition: If the attribution is unbiased,


various competent attributors should be com-monly capable of arriving at it. But also ifthe interactional modality is defective, thenvarious competent attributors should be sim-ilarly misled and should consensually arriveat an incorrect attribution. A similar case canbe made for the common deducibility statusof consistency across times with regard to theM and P problems, and for the idea that inthe P and T conditions, consensus and tem-poral consistency, respectively, are noncom-monly deducible and the remaining evidencetypes are commonly deducible from the cor-responding problems.

The results of the study just described lendstrong support to the present theory. In eachexperimental condition (and across diversehypothetical events), the subjects expressedsignificantly greater interest in noncommonlyversus commonly deducible information.Thus, in the M condition, interest in consist-ency across modalities significantly exceededinterest in the remaining information. In theP condition, interest in consensus informationsignificantly exceeded interest in the remain-ing criteria, and in the T condition interest inconsistency across time exceeded interest inthe remaining criteria (with marginal signif-icance). To summarize, experimental evidencesuggests that when faced with a given episte-mic problem, the knower attaches greatervalue to (or has a greater interest in) non-commonly deducible evidence as opposed toevidence that is commonly deducible or non-deducible from the alternative propositionsthat he/she may be considering.

Availability

According to the foregoing discussion, onlyepistemic contents (i.e., particular proposi-tions and evidence types) that are teleolog-ically or validationally functional wouldfigure in an epistemic episode. But considera-tions of functionality set only a loose con-straint on contents dealt with by the knower.Thus, a person could generate a large varietyof epistemic problems that are all teleolog-ically relevant to this knower's end. For in-stance, someone whose end may be to pur-chase a car may wonder about (a) the way

to the nearest dealership, (b) the comparativeadvantages and disadvantages of the variousmodels, and (c) the corresponding priceranges. Furthermore, granting a specific epi-stemic problem, the knower may seek evi-dence for different, noncommonly deducible(hence, validationally functional) implica-tions of the problem's constitutive proposi-tions. For instance, an individual who wishesto ascertain whether the way to the grocerystore lies to the left or to the right mightvariously seek evidence for the followingnoncommonly deducible implications of the"left" and "right" hypotheses: (a) If theway actually lay to the left (or to the right),a policeman would attest to this; (b) a mapwould indicate this; (c) pursuing the left (orthe right) path would actually lead to thedestination, and so on.

Momentary availability may determinewhich among the teleologically functionalproblems and the validationally functionalinformation would actually be addressed bythe person. By "availability" is meant theease with which a given cognitive contentcomes to mind, that is, the ease with which itis readily conceptualized or retrieved at agiven moment (Kahneman & Tversky, 1973;Tversky & Kahneman, 1974). This should de-pend on a highly particularistic array of fac-tors, such as the repertory of concepts that agiven individual has learned in the course of alifetime, the frequency with which a givencontent element has been recently employedin similar circumstances, or the saliency ofsuch content elements in the situation.

To conclude this portion of the article, letus recapitulate the central notions of thepresent epistemic theory. It is assumed thatthe sequence of epistemic activities can besubdivided according to the stages of problemformulation and of problem resolution. It isproposed that an epistemic problem gets for-mulated in accordance with the knower's pur-pose for acquiring a given bit of knowledge.The epistemic problem is conceived of as aset of alternative propositions in whose con-tents the knower happens to be interested.Those contents could include self-knowledge(e.g., regarding one's attitudes, feelings, orproperties) as well as knowledge about other


persons, objects, and events. In the course ofproblem resolution, the knower is assumed to(a) deduce from the alternative propositions(as he/she understood them) implicationsthat might differentiate among them, (b) seekevidence regarding those implications, and(c) come to feel greater confidence in prop-ositions consistent with the evidence andlesser confidence in those inconsistent withthe evidence. Contents of the propositions in-voked by a knower and of implications de-duced from those propositions are assumed todepend on their situational functionality andavailability. The invoked contents would haveto be considered by the knower as functionalto his/her situational objectives and to beavailable or mentally accessible to this indi-vidual at the given moment.

In the following sections the theoreticalnotions just reviewed are applied to majorformulations in the field of attribution. Inparticular, an attempt is made to show howthe present framework provides an integrativeconceptual scheme whereby diverse attribu-tional conceptions may be analyzed.

Attribution Theory; A Lay EpistemicInterpretation

The following reinterpretation of attribu-tional formulations in present theoreticallanguage stresses two major proposals. Ac-cording to one proposal, different attributionalformulations address the contents of differentepistemic problems. If so, it follows from thepresent notion of teleological functionalitythat a given attributional formulation willapply to an individual's epistemic activitiesonly when the particular problem it encap-sulates is functional to this person's situa-tional objective. According to the second pro-posal, the diverse attributional criteria as-sumed to determine the knower's confidencein specific inferences represent, at root, thelogic of noncommon deducibility. This partic-ular argument implies further that (a) agiven attributional criterion will have the pre-dicted impact on confidence only when theknower's epistemic problem renders the cri-terion a noncommonly deducible bit of infor-mation, and (b) the notion of noncommon

deducibility is the generative principle where-by one can derive the particular attributionalcriteria mentioned in the literature so far andalso other functionally equivalent criteria.

Problem Differences AmongAttributional Models

A brief review of major attributional for-mulations will suffice to demonstrate thatthey indeed address distinct epistemic prob-lems. In Kelley's (1967) analysis of variance(ANOVA) formulation, the attributor's prob-lem is to determine whether some effect wascaused solely by (a) the external entity, (b)the person (his/her unique properties), (c)the interactional modality, (d) the "time"at which the interaction occurred.4 Somewhatdifferent problems figure in the several causalschemata and attributional principles that thelayman was assumed to employ in lieu of thefull fledged ANOVA (Kelley, 1971, 1972). Forexample, the person-entity schema concernsthe problem of determining whether an effectwas caused by (a) the actor's personal prop-erties or (b) the entity's properties respondedto similarly by people in general. This is anarrower problem than contained in theANOVA cube, as the latter also includes theadditional possibilities that the effect wascaused by the modality or the time. In themultiple-necessary-cause schema (Kelley,1972), the attributor assumes that severalcausal factors were indispensable for an ef-fect's occurrence. This may be interpreted asa specific resolution to the problem of decid-ing whether, alternatively, only a single fac-tor may have been indispensable. In this par-ticular case the epistemic problem concernsthe number of causal factors (one vs. several)rather than the specific nature of such factors(as in the person-entity case). This hardlyalters the status of the necessary-causesschema as a resolution to a specific epistemic

4 Actually, Kelley's (1967) ANOVA matrix allows theadditional possibility that some particular combina-tion of entity and/or person, time, or modality wasthe cause of the effect. For simplicity's sake, however,this possibility will be disregarded throughout thepresent article.


problem in which a knower may be sometimesinterested. The pairing and grouping schemata(Kelley, 1972, p. IS) concern, respectively,symmetry in social relations and transitivityas well as symmetry in such relations. Fromthe present perspective both schemata referto specific propositions about social relationsthat could figure in some knower's epistemicproblems. In the Jones and Davis (196S)model of correspondent inferences, the per-ceiver's epistemic problem is to determinewhether or not a given act was prompted bya "correspondent" attribute in which corre-spondence (a) is defined in terms of the ex-tent to which "the act and the underlyingcharacteristic are similarly describable by theinference" (Jones & Davis, 1965, p. 223) and(b) is assumed to increase "as the judgedvalue of the attribute departs from the judge'sconception of the average person's standingon that attribute" (p. 224).

An attributional model proposed by Weinerand his associates (Weiner et al., 1971) ad-dressed the problem of deciding whether asuccess or a failure was prompted by one ormore of "ability," "effort," "'task difficulty,"and "luck." Finally, Kruglanski (197S) dis-cussed the .case in which the layman's epi-stemic problem is to decide whether a volun-tary action was an end in itself (representingan endogenous attribution) or a means to afurther end (representing an exogenous attri-bution).

Although not exhaustive, the foregoing re-view demonstrates that the various attribu-tional models differ in the particular epistemicproblems they feature. The formulations con-sidered thus far concern causal or explanatoryproblems, but other attributional analysesaddress noncausal inferences, for example,those about responsibility or freedom (e.g.,Harvey, 1976; Shaw & Sulzer, 1964).

That attribution-theoretic models have ad-dressed distinct epistemic problems may not,in and of itself, seem surprising. Yet thebroader, metatheoretic import of this hastypically gone unattended. Note that peopleare competent to generate vast numbers ofepistemic problems classifiable in a vast num-ber of ways. The problem ensconsced, say, inKelley's (1967) ANOVA formulation may be

considered merely one among a large numberof possible causal problems, and so with prob-lems of Jones and Davis (1965), Kruglanski(1975), or Weiner et al. (1971).

Thus, someone interested in explaining anact need not be interested in whether the actfollowed from a given "correspondent" attri-bute. For instance, a person might merely wishto identify the actor's momentary intentionwithout wondering about its further linkagewith a stable trait. Similarly, a knower inter-ested in the actor's intention need not formu-late an endogenous-exogenous problem, butmight wonder instead which of two equallyexogenous intentions has brought the actabout. A knower attempting to account for aperformance outcome (a success or a failure)need not phrase a problem in terms of thecategories of "ability," "effort," "luck," and"difficulty," but instead might wish to knowwhether a success was prompted by "diligentpreparation," "cheating," "the teacher's fa-voritism," and what not.

Granting the potentially vast number ofepistemic problems, the extant collection ofattributional formulations could be regardedas a vanishingly small sample of possible suchmodels differing in the particular problemsthey address. Thus it becomes of some im-portance to determine whether the problemsin current attributional models may claim aspecial significance for the layman versusconstituting a more or less arbitrary collec-tion of no unique distinction. The attribu-tional analyses so far have not attempted tojustify their choice of epistemic problems. Onreflection, however, any such attempt wouldseem a priori bound for failure. Any epistemicproblem could be of interest to someone insome circumstances, and the extent to whichit is recurrent or pervasive might vary widelywith the particular life vicissitudes of personsor groups.

Thus, there seems at least reasonable roomto doubt whether the epistemic problems con-tained in the attributional formulations de-serve special attention. But if they do not,focal attributional constructs as, say, the"ANOVA cube," "correspondent inferences,"the "endogenous-exogenous partition," andthe "ability, luck, effort, and difficulty" cat-


egories, all demonstrably reduced now to thespecifics of epistemic problems, might haveto abdicate their theoretical centrality. Theymight come to be regarded as a mere sampleof lay notions, of equal potential significancefor the knower to the multitude of alternativeconcepts in most individuals' repertories.

Finally, insofar as distinct attributionalformulations address distinct epistemic prob-lems, the present theory suggests under whatconditions any given such formulation maynot represent the individual's knowledge-seeking activities. This property follows di-rectly from the notion of ideological func-tionality whereby the epistemic problem (ad-dressed by a given attributional formulation)would not be tackled by a knower unless itspotential resolution was assumed to advancethis person's situational objectives. Thus, inthe research referred to earlier (Kruglanski,Hamel, Maides, & Schwartz, 1978), subjectsexpressed a preference for a resolution to theperson-stimulus problem (reminiscent of Kel-ley's, 1972, person-entity schema) only whensuch a resolution was made functional to thesubjects' experimentally induced objectives,and the same was true for alternative episte-mic problems modeled after alternative attri-butional formulations (e.g., the end-means orability-effort problems modeled after Krug-lanski, 1975, and Weiner et al, 1971, re-spectively).

Attributional Criteria and NoncommonDeducibility

If distinct attributional formulations ad-dress different epistemic problems, it followsfrom the present analysis that the types ofevidence that may be used toward the resolu-tion of such problems should also differ. Suchevidence should be noncommonly deduciblefrom the alternative propositions making upthe problem in each case. Accordingly, thefollowing review attempts to demonstrate thatthe various attributional criteria featured inthe various formulations may indeed be in-terpreted as noncommonly deducible implica-tions of problems featured in the respectiveformulations.

Kettey's (1967) criteria for external attri-bution. According to Kelley (1967), a con-fident causal attribution of effect to externalentity would be rendered if the following fourcriteria were satisfied

(1) Distinctiveness: The impression is attributed tothe thing if it uniquely occurs when the thing ispresent and does not occur in its absence; (2) Con-sistency over time: Each time the thing is present,the individual's reaction must be the same or nearlyso; (3) Consistency over modality: His reactionmust be consistent even though his mode of interac-tion with this thing varies; (4) Consensus: At-tributes of external origin are experienced the sameway by all observers, (p. 197)

Let us see now in what sense the foregoingevidential pattern is noncommonly deduciblefrom the epistemic problem addressed by Kel-ley's (1967) model. As we have seen this isthe problem of determining whether the effectwas caused exclusively by (a) the externalentity, (b) the person interacting with theentity, (c) the modality of the interaction, or(d) the time at which the interaction tookplace.

It should first be noted that distinctivenessof an effect with regard to external entity isconsistent with the hypothesis that the ex-ternal entity was the cause of the effect; dis-tinctiveness or conjunctive presence or ab-sence is implied by the causality concept ascommonly understood. One implication of theproposition that in a given situation A wasthe cause of B is that under those circum-stances the removal of A would eliminate Bas well (conjunctive absence), whereas anintroduction of A would lead to the appear-ance of B (conjunctive presence).

Further, it should be noted that constancyof the effect (e.g., across people, times, ormodalities) may be conceived of as the ab-sence of distinctiveness. To say that an effectis constant across persons, for example, meansthat it is present even though the specificperson is absent, thus violating the conjunc-tive presence/absence criterion of distinctive-ness. Therefore, constancies across persons,modalities, and times may refute the implica-tions that the effect was distinctive with re-gard to the specific person, modality, or time.Insofar as distinctiveness is implied by the


causality notion, the above constancies areinconsistent with the propositions that thespecific person, modality, or time was thecause of the effect.

Thus the evidential "package" identifiedby Kelley (1967) and including distinctive-ness of the effect with respect to external en-tity and constancy of effect across persons,modalities, and times is consistent with oneof the alternative propositions implicit inKelley's cube (that regarding entity causa-tion) and is inconsistent with other suchpropositions (regarding causation due to per-son, modality, or time) .B In present terms theforegoing evidential pattern is, therefore, non-commonly deducible from the alternativepropositions suggested by Kelley's model.

Although it is true that Kelley's (1967)criteria can be viewed as noncommonly de-ducible from the alternative propositions im-plied by the ANOVA cube, application of thenoncommon deducibility notion to those par-ticular propositions does not necessarily yieldthose particular criteria, but could yield othercriteria as well. For example, in addition tothe property of conjunctive presence-absenceof cause and effect, the notion of causality iscommonly taken to imply the temporal prece-dence of the former over the latter. Accord-ingly, an evidential pattern whereby someeffect was preceded by an entity, but, in turn,preceded the particular person, modality, andtime would be consistent with entity causa-tion hypotheses and inconsistent with the per-son, modality, or time causation hypotheses.This also would be noncommonly deduciblefrom the epistemic problem represented byKelley's cube.

Finally, the notion of noncommon deduc-ibility highlights the idea that Kelley's(1967) criteria for entity attribution mayonly apply when the epistemic problem con-tains all jour of the possibilities suggested byKelley's model, notably that the effect mayhave been caused by the (a) entity, (b) per-son, (c) modality, or (d) time. But considernow a more restricted problem in which theknower has generated only two (instead offour) of those alternative propositions, say,that the effect was caused by (a) the specificperson or (b) the environmental entity. In

such a case, consistency over times andmodalities would be of little use to the knowerbecause it is commonly rather than noncom-monly deducible from the alternative proposi-tions at stake. Thus, if the person were thecause, the effect should appear consistentlyacross occasions in which the specific indi-vidual was present. But also, if the externalentity were the cause, consistency over timesshould be expected granting the entity'spresence at those times. Thus, consistencyover times, for example, may not differentiateamong the person and the entity hypotheses.It lacks noncommon deducibility from thosetwo propositions, and hence it should notaffect the confidence of someone attemptingto choose among them. The very same argu-ment can be made for the common deducibil-ity status of consistency-across-modalities cri-terion as far as the person-entity problem isconcerned. In sum, when the knower's prob-lem is to decide whether the person or theentity was cause of the effect, the criteria forentity attribution would not encompass theentire set of four propositions mentioned byKelley (1967), but would be limited to thetwo noncommonly deducible ones, in this casedistinctiveness of the effect with respect tothe entity and constancy of the effect acrosspersons (i.e., consensus).

The covariation principle. The criteria forexternal attribution (Kelley, 1967) were in-timately tied to specific causal categories:The distinctiveness outcome was reserved forthe entities category, and the constancy out-come, for the times, modalities, and personscategories. However, in a subsequent article,Kelley (1971) articulated the distinctiveness-consistency rule in terms relatively uncom-mitted to specific categories. This was labeled

5 Strictly speaking, the "attributional criteria" pro-posed by Kelley (1967) (i.e., distinctiveness of theeffect with respect to the entity and constancy of theeffect across persons, times, and modalities) are com-patible with the idea that some combination of en-tity, person, modality, and the time has caused theeffect, as well as with the idea that only the entityhas caused the effect: In ANOVA terms, the appear-ance of only one main effect (analogous to the dis-tinctiveness outcome) in a multivariate matrix doesnot preclude the additional existence of interactioneffects.


the covariation principle, whereby "an effectis attributed to the one of its possible causeswith which, over time, its covaries" (Kelley,1971, p. 3). Strictly speaking, covariation de-notes distinctiveness or conjunctive presence/absence of entity and effect across numeroustimes. In the present terminology, covariationis noncommonly deducible from the proposi-tions that (a) the particular entity (x) wasthe effect's sole cause and, (b) the particulartime (t) played a causal role in the effect'sappearance: Covariation across times is con-sistent with the first proposition and incon-sistent with the second.

In 4he preceding example, covariation wasnoncommonly deducible from the knower'sepistemic problem. But it is reiterated thatthe noncommon deducibility rule may lead toimplications other than covariation. Evengranting the same epistemic problem of decid-ing whether entity x or time t was the causeof the effect, one could employ the noncom-mon deducibility notion to generate a criterionother than covariation, using instead the tem-poral precedence implication of causality.Temporal precedence of entity x relative tothe effect and temporal precedence of theeffect relative to time t together add up toan evidential pattern that is noncommonlydeducible from the problem at issue and couldbe used to resolve it in favor of an entityattribution in the same way as covariation.

The difference between the notions of non-common deducibility and of covariationstands out particularly in those cases in whichcovariation is commonly rather than noncom-monly deducible from the alternative proposi-tions considered by the knower. Consider thecase in which the knower's problem is to de-termine the direction of causality, as in choos-ing among the alternative propositionswhether "A caused B" or "B caused A." Eventhough we are dealing here with a causalproblem and covariation is deducible from theidea of causality, in the case at hand, covaria-tion is commonly rather than noncommonlydeducible from the alternative propositionsconsidered by the knower: The proposition"A caused B" implies that A and B wouldcovary across times, but the alternative "Bcaused A" implies the very same covariation.

Thus, because it lacks noncommon deducibil-ity in this case, covariation evidence wouldnot be useful to the knower in attempting todecide between the alternative propositionsconcerning the direction of causality.

It is also reiterated that whereas covaria-tion is deducible from the notion of causality,it is not so deducible from numerous other no-tions that the knower's epistemic problemscould sometimes address. In the latter casesnoncommonly deducible evidence would bequite dissimilar from covariation evidence.Suppose that at a professional meeting con-ducted many years ago a young psychologistp was greatly impressed by professor P witha distinct German accent and the youngpsychologist wondered whether the professormight be Kurt Lewin or Wolfgang Kohler.Suppose further that the curious youth eaves-dropped on P and overheard him say that hewas born in Prussia in the year 1890. Beingacquainted with biographies of the personal-ities involved, our knower might quicklyrealize that the evidence was consistent withthe Lewin hypothesis and inconsistent withthe Kohler hypothesis. In present terms theevidence would be noncommonly deduciblefrom the alternative propositions of interest,although it hardly qualified as covariation; itwould seem somewhat incorrect to say thatLewin "covaried" with the specific date andplace of birth so that whenever he was presentor absent so were they.

In short, covariation evidence could some-times be noncommonly deducible from thealternative propositions making up a knower'sepistemic problem. But in other cases covaria-tion evidence would be commonly deducibleor completely nondeducible from such prop-ositions. According to the present analysis,only in the first case and not in the latterones would covariation evidence be of interestto the knower or affect the knower's con-fidence in a proposition or an inference.

Discounting. According to the discountingrule (Kelley, 1971, p. 8), "the role of a givencause in producing a given effect is discountedif other plausible causes are also present." Inother words, the knower's confidence regard-ing the causal role of a given factor would begreater if it alone appeared as a plausible


cause. Note that the discounting rule may beinterpreted to refer to the impact on aknower's confidence of noncommonly deduc-ible information inconsistent with a causalhypothesis: A plausible competing cause is(trivially) consistent with itself, and at thesame time is inconsistent with the alternativecause. According to the present theory, thisshould reduce confidence in the latter cause asimplied by the discounting rule (Kelley,1971).

The present concept of noncommon deduc-ibility has implications that have not beenobvious from the discounting notion. In par-ticular, if discounting is a special case of non-common deducibility, it should occur onlywhen the alternative hypotheses consideredby the knower appears truly inconsistent ormutually exclusive. This has not been as clearin extant discussions of the discounting rulethat stressed the number of causal hypothesesrather than their contradictory nature. Yetintuitively it seems that the discounting phe-nomenon would not occur if the alternativehypotheses entertained by the knower over-lapped. For instance, imagine a patient toldby one physician that his/her particular symp-toms were the result of influenza, and told byanother physician that the symptoms werethe result of a virus. It seems that the secondsuggestion would not cause any appreciablediscounting of the first because the "virus"hypothesis does not exclude that of "influ-enza," so no inconsistency is involved. Al-though intuitively appealing, the present in-terpretation cannot, so far, claim systematicdata in its support, and further research isneeded to conclusively establish its validity.

Furthermore, in its extant formulation(Kelley, 1971), the discounting rule refers tothe content of causal hypotheses; its presentportrayal as a special case of noncommon de-ducibility suggests that the discounting effectshould also occur with competing hypotheseswhose content is not causal. For example, aperson considering the possibility that theroad to Jerusalem lay directly ahead mightlose some confidence in this proposition if asign indicated instead that the road lay tothe right. Similarly, someone considering thepossibility of the time being five o'clock might

lose some faith in this hypothesis if a glanceat a friend's watch indicated five-thirty in-stead. In both cases the introduction of anincompatible alternative may reduce theknower's confidence in a given inference eventhough the hypotheses involved are not causalin nature.

Jones and Davis's (1965) criteria of corre-spondent inference. Jones and Davis (1965)articulated two major criteria whose fulfill-ment was assumed to lead to correspondentinferences. First, the number of noncommoneffects separating the action from the un-chosen alternative and second, the assumedsocial desirability of those effects. On closeexamination, both criteria turn out to re-flect the noncommon deducibility logic asapplied to the specific problem content ad-dressed by the correspondent-inference model.

According to the noncommon-effects cri-terion, "the probability [of any effect of be-ing the perceived goal of the action] shouldvary inversely as a function of the numberof other effects competing for the perceiver'sattention" (Jones & Davis, 1965, p. 228) .Assuming that "probability of effect beingthe goal" means roughly the same as con-fidence in its being the goal, and the "effectscompeting for the perceiver's attention" areplausible alternative hypotheses about theaction's underlying goal, the similarity of theeffects principle to the discounting rule be-comes apparent (as observed by Kelley, 1971,p. 8). Thus, the earlier interpretation of dis-counting as an instance of noncommon de-ducibility extends to the number of effectscriterion as well. For a given perceiver theproposition that effect x was the likely goalof some action is (trivially) consistent withitself and inconsistent with the hypothesisthat effect y was the exclusive goal; becauseof the inconsistency, the perceiver's con-fidence should be lowered as compared to thecase where only x or only y appeared as alikely goal.

In addition to their number of effects prin-ciple, Jones and Davis (1965) viewed as amajor condition for a correspondent inferencea low assumed social desirability of the act's(noncommon) effects. Recall (a) that a cor-respondent inference was defined in terms of


the degree to which "the judged value of theattribute departs from the judge's conceptionof the average person's standing on that at-tribute" (Jones & Davis, 1965, p. 224), and(b) that the epistemic problem featured inthe Jones and Davis (1965) analysis was todecide whether an act was caused by a cor-respondent attribute. It should be clear thatlow (assumed) social desirability is con-sistent with the correspondent-attribute hy-pothesis and inconsistent with the noncorre-spondent-attribute hypothesis: If the actor'sreason for a behavior is rare within a culture(i.e., low in assumed social desirability), somust be the underlying attribute. This wouldrender correspondent the inference in ques-tion. In the foregoing sense then, assumed so-cial desirability constitutes a noncommonlydeducible implication of the propositions thata correspondent versus a noncorrespondentattribute prompted a given act.

For example, in the classical experiment byJones, Davis, and Gergen (1961), subjectsinferred with greater confidence that an actorhad a particular personality characteristicwhen he/she behaved consistently with thecharacteristic and contrary to role require-ments (or, out of role) that presumablywould have elicited the opposite behaviorfrom most people. The low assumed socialdesirability of the particular behavior in thiscase is consistent with -the proposition thatthe actor had an unusual degree of the par-ticular characteristic or inconsistent with theproposition that he/she was about averageon that particular characteristic: Simply, theknower (the perceiver) may reason that anunusual degree of the characteristic impliesthat it may manifest itself despite inhibitivesituational constraints, whereas a merelyaverage degree of the characteristic impliesthat it would not be manifest under such con-straints. Thus, the actor's out-of-role behaviorin the Jones et al. (1961) experiment is non-commonly deducible from the propositionsthat the actor deviates versus does not deviatefrom the average person on some trait, or inJones and Davis's (1965) terms that the in-ference is correspondent or not. Note that al-though the noncommon deducibility principlecan be used to generate the assumed social

desirability rule, it can also be used to gen-erate alternative criteria for validating thecorrespondence of an inference. For instance,insofar as a correspondent attribute is aboveall an attribute, that is, a trait steadily ownedby a person, it affords the deduction of stabil-ity, or recurrence of the same act on numeroussimilar occasions. Such stability would differ-entiate the correspondent attribute hypothesisfrom the one alternative that the act wascaused by a momentary whim. Furthermore,from the specific content of any correspondentattribute, one could deduce a variety of differ-ent acts whose collective appearance in theappropriate circumstances would convergentlyvalidate the attribute in question (Ajzen &Fishbein, 1975).

Augmentation. A major attributional prin-ciple featured in Kelley's (1971) analysis isthe augmentation rule whereby

if for a given effect both a plausible inhibitory causeand a plausible facilitative cause are present, the roleof the facilitative cause in producing the effect willbe judged greater than if it alone were present as aplausible cause for the effect, (p. 12)

In different terms (Kruglanski, Schwartz,Maides, & Hamel, 1978), if an effect occurreddespite a potent counterforce (inhibitorycause), the prompting force (facilitativecause) would be believed of considerablemagnitude. Note that occurrence of an effectdespite a hindrance is noncommonly deduciblefrom the epistemic problem of whether thecause was of a considerable versus a slightmagnitude: Occurrence in the face of a hin-drance is consistent with the "considerablemagnitude" hypothesis and inconsistent withthe "slight magnitude" hypothesis. Directexperimental evidence for the augmentationeffect has been furnished by Kruglanski,Schwartz, Maides, and Hamel (1978).

Note again that the noncommon deducibil-ity idea suggests that it is possible to derivecriteria other than the augmentation principlefrom the "considerable causal magnitude" hy-pothesis. For example, all else being equal,the hypothesis that the cause was of consider-able magnitude implies a large magnitude ofthe effect. Thus high magnitude of effectwould be evidence noncommonly deducible


from the propositions that the cause was of aconsiderable versus a slight magnitude.

Weiner et al.'s (1971) conditions forachievement-related attributions. In theirattributional analysis of behaviors in achieve-ment contexts, Weiner et al. (1971) identifiedsome stimulus conditions assumed to result inattributions to ability, luck, effort, and taskdifficulty. For instance, attribution of successto task ease is assumed likely if most persons,in addition to the actor, also succeeded at thetask. Attribution of success to effort is as-sumed likely given evidence regarding "(1)covariation of performance with incentivevalue or ... (2) with cues as perceived mus-cular tension, or task performance" (Weineret al., 1971, p. 5). The epistemic problemaddressed by Weiner et al. was to decideamong the propositions that a performanceoutcome (a success or failure) was causedexclusively by outstanding ability, effort, luck,or task ease (difficulty). The kinds of evi-dence just listed are noncommonly deduciblefrom these propositions: Success on task bymost other persons is consistent with thehypothesis that task ease may constitute theexplanation and is inconsistent with the hy-pothesis that ability, effort, or luck has to bethe exclusive explanation. Similarly, "covaria-tion of performance with incentive value" isconsistent with the effort hypothesis and isinconsistent with the ability, luck, or easehypotheses. Note that here, also, the noncom-mon deducibility principle could be used togenerate other kinds of evidence than thoselisted by Weiner et al. Thus, in addition tothe success of many persons, task ease mayalso imply a speedy performance or successwithout preparation; the effort hypothesismight sometimes imply long hours at the task,experienced weariness in the aftermath oftask completion, and so on; in short, all pos-sible implications of the effort, ease, ability,and luck hypotheses could occasionally con-stitute noncommonly deducible items of evi-dence and be used to validate the hypothesesin question.

Discussion

In the preceding sections I attempted todemonstrate that various attributional criteriamay represent the noncommon deducibilitylogic applied to specific epistemic problems.Thus, the notion of noncommon deducibilityprovides the general construct whereby sep-arate attributional notions as apparently dif-ferent as say, "discounting," "assumed socialdesirability," or "covariation" are integrated.

But the generality of the noncommon de-ducibility concept extends beyond the attribu-tional criteria articulated so far: (a) Giventhe same epistemic problem that the originalcriterion was meant to resolve, the noncom-mon deducibility notion may be used to gen-erate alternative, functionally equivalent cri-teria. As has been shown, such criteria mayrefer to implications of the knower's hypoth-eses that the original criteria did not address,(b) The notion of noncommon deducibilityextends beyond the epistemic problems cov-ered by extant attributional formulations.Those formulations addressed specific episte-mic problems, whereas the logic of noncom-mon deducibility is assumed applicable to allepistemic problems regardless of content.

Finally, the notion of noncommon deduc-ibility renders explicit the idea that an itemof information (e.g., an attributional cri-terion) may validate a hypothesis (e.g., prop-osition, or attribution) only when it differen-tiates it from the competing propositions thatthe knower may consider plausible. In otherwords, criteria for the same attribution wouldvary as a function of the total epistemic prob-lem confronted by the knower. To reiteratean earlier example, the knower with a problemof deciding whether A caused X or B causedX might well use as a criterion for the A-caused-X attribution evidence of covariationbetween A and X and an absence of covaria-tion between B and X. But a knower con-fronted with a problem of deciding whetherA caused X or X caused A would not likelyuse as a criterion (for the same A caused Xattribution) evidence regarding A's covaria-tion with X, as it would not differentiate thatattribution from the competing X caused A


attribution of concern to the knower at themoment.

The present theory addressed the processof naive inference. It was suggested that thisprocess begins with a purpose that an indi-vidual may have for acquiring a given itemof knowledge, and that it consists of thestages of problem formulation and resolution,in the latter of which a major role is playedby the logical principle of consistency. Inaccordance with this outline, it was proposedthat requirements of functionality may con-strain the epistemic problems formulated andthe evidence types employed to resolve them:(a) The epistemic problem formulated bythe knower must be teleologically relevant tothis individual's purpose such that the resolu-tion must be believed to advance the purposein some way, and (b) the evidence employedby the knower must be validationally func-tional to the knower's problem. The evidencemust be noncommonly deducible from theproblem's constitutive propositions so as tofacilitate a choice among them. Beyond con-siderations of functionality, the epistemiccontents dealt with by the layman (the prob-lems and evidence types attended to) woulddepend on their momentary availability, thatis, on the ease with which they may come toa person's mind in the situation.

Extant attribution-theoretic formulationswere reviewed from the present epistemic per-spective and the following major generaliza-tions were arrived at. First, the various at-tributional models differ in the particularepistemic problems and the particular attri-butional criteria with which they deal. In thissense, the attributional models concern par-ticular instances of inference rather than theunderlying inferential process. Second, thenecessary applicability conditions of the at-tributional models are ideological and valida-tional functionalities of the particular episte-mic problems and informational criteria fea-tured in the models: To adequately depict theknower's epistemic activities in some situa-tion, the epistemic problem featured in agiven model must be functional to the per-son's situational objective, and the attribu-tional criteria specified in the model must benoncommonly deducible from the problem in

question. Thus, the present paradigm pro-vides an integrative framework that enablesconsideration of diverse attributional modelsin common theoretical terms and that de-lineates the applicability conditions of vari-ous such models.

The empirical support for the presenttheory may be classed into two categories:(a) Much of the evidence directly supportiveof "veteran" attributional formulations sup-ports the epistemic framework to the extentthat it indeed subsumes them as special cases,and (b) evidence for unique implication ofthe novel framework not readily derivablefrom previous analyses. Support in the firstcategory is represented, for example, by Mc-Arthur's (1972) results regarding the effectsof consensus, distinctiveness, and consistencycriteria introduced by Kelley (1967). In thesame category also falls evidence for the dis-counting and the covariation principles (sum-marized by Kelley, 1971) and for the corre-spondent-inferences theory (summarized byJones & McGillis, 1976).

Support in the second category has in-cluded the previously described findings re-garding Ideological and validational func-tionality as they may affect the knower'sinterest, in particular, epistemic problemsand/or evidential patterns. Admittedly, fur-ther research on the unique implications ofthe present theory is well indicated. Somesuch research might further test the presentimplications regarding conditions under whicha given attributional principle (e.g., discount-ing or covariation) would or would not affectthe knower's inferences. Alternatively, onemight apply the epistemic model to contentsnot previously addressed by attributionalformulations. For example, the discountingrule (Kelley, 1971) has been enunciated inreference to causal hypotheses, but the pres-ent analysis suggests it to be applicable tohypotheses of whatever type, including non-causal ones as well.

The present article necessarily addressedonly a portion of issues relevant to humanepistemic behavior. Some major remainingones are now indicated in passing.

1. Inferential biases and errors. These havebeen typically viewed as deviations from epi-


stemic rationality, possibly deriving fromreliance on various suboptimal schemata(Kelley, 1972), strategies, or heuristics(Tversky & Kahneman, 1974). By contrast,the epistemic process outlined herein, al-though conceived of as rational, allows at thesame time for the possibility of biases anderrors stemming, for example, from the mo-mentary unavailability of hypotheses or in-formation to the knower.

2. Motivational influences on the epistemicprocess. A special category of inferentialbiases may stem from the influence of needsand desires on a person's confidence in vari-ous propositions. Writers on attribution havelisted various such influences (e.g., Jones &Davis, 1965; see Kelley, 1967, 1971; Miller& Ross, 1975), but their place in the layepistemic framework remains to be explicated.

3. Relation oj the lay epistemic theory toother conceptualizations in cognitive socialpsychology. Contemporary social psychologyhas laid considerable stress on the role inhuman affairs of the acquisition and utiliza-tion of lay knowledge. Theories of cognitiveconsistency (see Abelson et al., 1968), of so-cial comparison processes (Festinger, 1954),and of attitude formation and change (Fish-bein & Ajzen, 1975; Himmelfarb & Eagly,1974; Insko, 1967) all contain a distinctaspect of knowledge acquisition. Insofar asthe present epistemic framework is assumedto address the process whereby all knowledgeis gained, it could provide the basis for link-ing together the epistemic aspects of severaldisparate formulations in cognitive socialpsychology.

Reference Note

1. Kruglanski, A. W., & Bar-Tal, D. Emerging con-ceptions of attribution: An integrative review.Unpublished manuscript, Tel-Aviv University.

References

Abelson, R. P., et al. (Eds.). Theories of cognitiveconsistency: A sourcebook. Chicago: Rand Mc-Nally, 1968.

Ajzen, I. Intuitive theories of events and effects ofbase-rate information on prediction. Journal of

Personality and Social Psychology, 1977, 35, 303-314.

Ajzen, I., & Fishbein, M. A Bayesian analysis of at-tribution processes. Psychological Bulletin, 197S,82, 261-277.

Asch, S. E. Forming impressions of personality.Journal oj Abnormal and Social Psychology, 1946,41, 258-290.

Bern, D. J. Self-perception theory. In L. Berkowitz(Ed.), Advances in experimental social psychology(Vol. 6). New York: Academic Press, 1972.

Buss, A. R. Causes and reasons in attribution theory:A conceptual critique. Journal of Personality andSocial Psychology, 1979, 37, 1458-1461.

Festinger, L. A theory of social comparison process.Human Relations, 1954, 7, 117-140.

Fishbein, M., & Ajzen, I. Belief, attitude, intentionand behavior: An introduction to theory and re-search. Reading, Mass.: Addison-Wesley, 1975.

Harvey, J. H. Attribution of freedom. In J. H.Harvey, W. J. Ickes, & R. F. Kidd (Eds.), Newdirections in attribution research (Vol. 1). Hills-dale, N.J.: Erlbaum, 1976.

Harvey, J. H., & Tucker, J. A. On problems withthe cause-reason distinction in attribution theory.Journal of Personality and Social Psychology,1979, 37, 1441-1446.

Heider, F. The psychology oj interpersonal relations.New York: Wiley, 1958.

Himmelfarb, S., & Eagly, A. (Eds.). Readings inattitude change. New York: Wiley, 1974.

Insko, C. A. Theories oj attitude change. New York:Appleton-Century-Crofts, 1967.

Jones, E. E., & Davis, K. E. From acts to disposi-tions. In L. Berkowitz (Ed.), Advances in experi-mental social psychology (Vol. 2) . New York:Academic Press, 1965.

Jones, E. E., Davis, K. E., & Gergen, K. I. Role-playing variations and their informational valuefor person-perception. Journal oj Abnormal andSocial Psychology, 1961, 63, 302-310.

Jones, E. E., & McGillis, D. Correspondent inferencesand the attribution cube: A comparative re-ap-praisal. In J. H. Harvey, W. J. Ickes, & R. F. Kidd(Eds.), New directions in attribution research(Vol. 1). Hillsdale, N.J.: Erlbaum, 1976.

Kahneman, D., & Tversky, A. On the psychology ofprediction. Psychological Review, 1973, 80, 237-2S1.

Kelly, G. A. A brief introduction to personal con-struct theory. In D. Banister (Ed.), Perspectivesin personal construct theory. New York: AcademicPress, 1970.

Kelley, H. H. Attribution theory in social psychology.In D. Levine (Ed.), Nebraska Symposium onMotivation. Lincoln: University of NebraskaPress, 1967.

Kelley, H. H. Attribution in social interaction. NewYork: General Learning Press, 1971.

Kelley, H. H. Causal schemata in the attributionprocess. New York: General Learning Press, 1972.


Kruglanski, A. W. The endogenous-exogenous parti-tion in attribution theory. Psychological Review,1975, 82, 387-406.

Kruglanski, A. W. Causal explanation, Ideologicalexplanation: On radical particularism in attribu-tion theory. Journal of Personality and Social Psy-chology, 1979, 37, 1447-1457.

Kruglanski, A. W., Hamel, I. A., Maides, S. A., &Schwartz, J. M. Attribution theory as a specialcase of lay epistemology. In J. H. Harvey, W. J.Ickes, & R. F. Kidd (Eds.), New directions in at-tribution research (Vol. 2). Hillsdale, N.J.: Erl-baum, 1978.

Kruglanski, A. W., Schwartz, J. M., Maides, S. A., &Hamel, I. A. Covariation, discounting, and aug-mentation: Toward a clarification of attributionalprinciples. Journal of Personality, 1978, 46, 176-189.

McArthur, L. A. The how and what of why: Somedeterminants and consequences of causal attribu-tion. Journal of Personality and Social Psychology,1972, 22, 171-193.

McGuire, W. J. Theory of the structure of humanthought. In R. P. Abelson et al. (Eds.), Theoriesof cognitive consistency: A sourcebook. Chicago:Rand McNally, 1968.

Miller, D. T., & Ross, M. Self-serving biases in theattribution of causality: Fact or fiction. Psycho-logical Bulletin, 1975, 82, 213-225.

Nisbett, R. E., & Valins, S. Perceiving the causes ofone's own behavior. New York: General LearningPress, 1971.

Schachter, S., & Singer, J. E. Cognitive, social, andphysiological determinants of emotional state.Psychological Review, 1962, 69, 370-399.

Shaw, M. E., & Sulzer, J. L. An empirical test ofHeider's levels in attribution or responsibility.Journal of Abnormal and Social Psychology, 1964,69, 39-46.

Tversky, A., & Kahneman, D. Judgment under un-certainty: Heuristics and biases. Science, 1974, 185,1124-1131.

Weiner, B. Achievement motivation and attributiontheory. Morristown, N.J.: General Learning Press,1974.

Weiner, B., et al. Perceiving the causes of success andfailure. Morristown, N.J.: General Learning Press,1971.

Received August 25, 1978Revision received July 24, 1979

Lay Epistemo-Logic—Process and Contents: Another Look at Attribution Theory A Theory of Lay...

Documents

Transcript of Lay Epistemo-Logic—Process and Contents: Another Look at Attribution Theory A Theory of Lay...