PERSONALITY RESEARCH, METHODS, AND THEORY

A Festschrift Honoring Donald W. Fiske

cditcd by

Patrick E. Shrout New York University

Susan T. Fiske University of Massaeh usetts at Anlherst

~ LAWRENCE ERLBAUM ASSOCIATES, PUBLISHERS 1995 Hillsdale, New Jersey Hove, UK

CHAPTE(\

3

WHAT THE HELL TOOK So LONG? DONALD W. FISKE AND THE l3IG-FIVE

F AGOR STRUGURE

Lewis rt Goldberg Universiry of Oregon ond Oregon (\eseorch Insrirure

If a 11l0dern-day Rip van Winkle had been asleep during the 40-year interval between 1949 and 1989, he would have been slartled upon a\vakening by the alnount of scientific progress that had occurred in genclics and nledicinc, space science, nlaterials technology, conlpute." science, and even in lhe cognitive, behavioral, and neurosciences. On lhe olher hanu. if he had becn pat1icularly inlerested in the n10st basic and fundamental problenl in personality psychology­lhe slruclure of hunlan pcrsonality lraits-, he would \vonder why it had laken 40 years to begin to develop a scientific consensus about an appropriate taxononlic nlodel. In this chapter, I briefly describe sonle of the events leading to loday's en1erging consensus on a structuraJ represenlation for phenolypic personality truits. highlighting lhe historieal significance of Donald Fiske's (1949) senlinal analyses. 1 thcn point out sonle of the obstncles to scicntific agreenlent that occurrcd betwccn lhe publication of Fiske's report and the .present-day enlerging consensus. nlC overall ainl of this chaptcr is to provok~,lconjcctures about those praeticcs and procedures that could help ensure that fulure research on lhe "cutting edge'· in pcrsonaJity psychology will nol always take so torturously long lO be acecpted.

THE DIG-FIVE FAcrOR STRUcrURE

Thc seareh for a taxononly of individual differcnces is an aneient one: Humans have continuously sought to nlake sense of the nlyriad characteristics t!ley observe in others around thenl. However, until the present eentury t efforts to c<;lnstruct

29

30 GOLDDERG

taxononlic rcpresentations for human differences have relied on individual intuitions, such as those of Aristot)e, who developed an early c1assification of human character traits. To develop a truly scientific nlodcl of individual differences, two fundamental problems had to be solv~d: ('l) how to obtain a representative, if not comprehensive, set of such attributes; and (b) how to c1assify or categorizc those attributcs into a structural nux]el. The first problenl was solved by Sir Francis GaIton, who may have been the first scientist lo recognize explicitly the fundanlental "Iexical hypothesis" (viz., that the most important individual differences in hunlan transactions will come to be encoded as single tenns in sonle or all of the world's languages). Moreover, Galton (1884) was certainly one of the first scientists to consult a dictionary as a means of estimating the nunlber of pcrsonality-descriptivc terms in the lexicon, and to appreciate the extent to which trait tertns share aspects of their nleanings. Galton's eslinlate of lhe nunlber of personality-related tcrms in English was later sharpcned elnpirically, firsl by AlIport and Odbcrt (1936), who culled such temlS froln the second edition of Webster' s Ullabridged Dictionary, and later by Norman (1967), who supplemented the earlier list with tenns fronl lhe lhird edition. Galton's insight conceming the relations anlong personality temlS has been nlirrored in efforts by later investigators to discover the nature of those rclations, so as to construct a structural representation of personal ity descriptors.

What Galton Iacked were the procedures for constructing such a taxonolnic nlodel. That problem was sol ved with the development of the broad c1ass of statistical techniques refelTed to generically as factor analysis. Moreover, the father of factor analysis, L. L. Thurstone, becanle the first scientist to use lexical nlaterials to test his new technique; he administered 60 conlnlon trait adjectives to 1,300 subjects who used thenl to describe sonleone they knew well. Ana)yses of the cOlTelalions anlong the 60 terms across lhe 1,300 subjects revealed five broad factors, which led Thurstone (1934) to assert that u ••• the scienlific description of pcrsonalily nlay not be so hopelessly conlplcx as it is sonletimes thought 10 be" (p. 14).

Curiously, 111urslone never followed up on his ¡nitial analysis of personality­trail temls. Inslead, the first scientist lo devote systenlatic attention to lhe problenl was Raynlond B. Cattell, who began his personality explorations with a pcrusal of the approximately 4,500 trait-dcscriptive temlS inc1uded in lhe Allporl and Odbcrt compendium. Callell (1943) used this trail list as a starting point, eventually devcloping a set of 35 highly conlplex bipolar variables, each poJe of which includcd a composite set of adjectives and phrases. These variables were lhen elnployed in various sludies, in each of which lhe correlations an10ng the variables were factored using oblique rotational procedurcs (e.g., Cattell, 1947).

Cattell has repealcdJy c1ainled to have identified at lcast a dozen oblique factors. However, when Cattell's variables were later analyzed by others, only five factors proved lo be repJicable (e.g., Dignlan & Takenlolo-Chock, 1981; Fiske, 1949; Nomlan, 1963; Smith, 1967; Tupes & Christal, 1961). Similar

3. FISKE ANO THE BIG-FIVE FACfOR'STRUCfURE 31

five-factor structures bascd on othcrscts of variables have been reported by a number of other investigators (e.g., Borgatta; 1964; Digman & Inouye, 1986; Goldberg, 1990, 1992; McCrae & Costa, 1985a, 1987), and these studics havc now becn revicwed extensivcly (c.g., Digman, 1990; John, 1990; McCrac & John, 1992; Wigg~ns & Pincus, 1992; Wiggins & Trapnell, in prcss).

TIlese "Big-Fivc" factors have traditio'nally becn numbcrcd and labcled: (1) Surgency (or Extraversion) , (II) Agreeableness, (In) Conscientiousness, (IV) Emotional Stability (vs. Neuroticism), and (V) Culture} More rccently, Factor V has bccn rcinterprctcd as InteUect (e.g., Digman & Takemoto*Chock. 1981; Peabody & Goldbcrg, 1989) and Openness lo Experience (c.g .• McCrae & Costa, 1987).

Although therc is sorne disagreement about the precise nature of these fivc dornains, there is widcspread agreelTIcnt that some aspccts of the language of pcrsonality description can be organized hierarchically (e.g., Cantor & Mischcl, 1979; Hampsont John, & Goldberg, 1986). In such a representation, the Big-Five domains are Iocated at the highest levelthat is stin descriptive of bchavior, with only general evaluation locatcd at a higher and more abstract level (John, Hanlpson, & Goldberg, 1991). When lhus viewed hicrarchically, it shouJd be clear that proponcnts of lhe five~faclor nlodel have never intended to reduce lhe rich tapestry of personality to a nlere five traits. Rather, they scek lo providc a scientific,llly compelling framcwork in which to organizc the nlyriad individual differences tha1 characterize humankind.

Indeed, these broad donlains incorporate hundreds, jf not thousands, of traits: Factor 1 (SlIrgency or EXlraversion) contrasts traits such as Talkativeness, Assertiveness, and Activity Level with traits such as Silellce, Passiv;ty, and Reserve; Factor n (Agreeableness or Pleasalltlless) contrasts traits such as Kindness, Trust, and Warmth wilh traits such as Hostility, Selflslmess, and Dislrust; Factor III (CoIIscientiousness or Depenclability) contrasts traits such as Organizarioll, Thorouglmess. and Re lia bi/it y with traits such as Carelessness, Negligence, and Unrelia/iUity, Factor IV (Emotio1lal Stabi!ity vs. Neuroticisl1') contrast traits such as Imperturbability and Calmness wilh traits such as Nervousness, Moodiness, and TemperanJcllrality; and Factor V (lntellect or Opelllless lO Experience) contrasts traits such as Imaginatioll, Curiosit)', ano Creativity with traits such as Shallowness and buperceptiveness ..

lAs pointcd out by Peabody and Goldberg (1989), the interpretalion of Factor V as Cullure arose from a historieal acddent: Although Callell had initially constructed a subset of variables reJaling to ¡me/lcí:I, in lhe seminal sludies of Canell (1947) he omittcd all .hose variables in favor of an intelligence test. In tumo this test was omiucd from his I~tcr studies. leaving no direct rcpresentation of most ¡mcllccl variables. In lheir abscnce. Factor Vwas cancd Culturc by Tupes and Chrislal ( 1 % 1) and so me laler investigators. However. when variables related 10 ¡II/ellcel were reintroouced (e.g .• Goldbcrg. 19(0). il beca me clear lhat ¡lIlcllcel was lhe more appropriale label for lhe fiflh broad factor.

02 GOLDDERG

THE REMARKAOLE CONTRIOUTION OF DONALD W. FISKE

Whereas Thurstone (1934) found the correct number of broad pcrsonality factors, his collection of 60 trait adjectives was too idiosyncratically assembled lo have produced today's Big-Five structure. Instead, lhe honor of first discovery must be accorded to Fiske (1949), who analyzcd a set of 22 variables devclopcd by Caue1J, and found five factors that replicated across sanlp)es of self-ratings, observer ratings, and peer ratings. Fiske's labcls for his factors, like those proposed by subscquent invcstigators, can be viewed as never perfectly successful attenlpts to capture the prototypical content of these broad domains-Confidellt SelfExpression (Big-Five Factor 1), Social Adaptability (Factor I1), COllformity (Factor UI), Emotional Control (Factor IV), and Inqlliring ¡ntellect (Factor V). Of these five labcls, four could still be used loday; only ConJormily seenlS too narrow and undesirable lo capture lhe nlany aspccts of responsibiJity and dependability inherent in Big-Five Factor III (Conscientiousness).

In Fiske 's article, entitled "Consislency of the Factorial Structures of Personality Ratings from Different Sources," he acknowledged the earlier contributions of Cattell (1947), and noted that no one had yet compared lhe factor slructures obtaincd fronl different typcs of raters. He asked, "Do (he factors in self-ratings bear any resemblanceto the factors in ratings by pcers? Are lhe factors fronl either of these sources conlparable to those found in lhe ratings of trained el inicians?" (p. 329). To answer lhese questions, he analyzed the ratings of 128 111en \vho were about to enter graduate schools in cJinical psychology and who haú attended one of lhe week-Iong asscssnlcnt sessions held at lhe University of Michigan in the sumnler of 1947; the details of this Veterans Adnlinistration Selection Research Projectare reported in the c1assic vo)ume by Kelly and Fiske (195). The subjects described thenlselves and were described by three peers and three rnembcrs of the assessment staff with 42 variables, of which 22 had been adapted from lhe set developed by CaUell () 947); 4111 ratings were made on 8-step sca)es. The self-ratings, median peer ratings, and the pooled staff ratings for lhe 22 Cattell variables were factored separatcly and then conlpared.

That sounds easy. Today it would be. However, 45 years ago, before lhe conlputcr revolution, all analyses had to be carricd out by hand calculator. With lhe help of his wife, Barbara, the Fiskes hand-calculated the correlations anlong lhe 22 variables, hand-extracted five centroid faclors, hand-rotated those factors by Thurstone's nlethod of extended vcctors, and then hand-calcuJaled lhe transfomlation tnatrices between the unrotated and rotated factors, as \vell as lhe correlations anl0ng the rotated factors. Moreover, they did all this three tinles--once for each of lhe three data sources. Final1y, to compare the three solutions, they hand-correlated the rotated factor loadings from each data source with lhe loadings fronl the other two sources.

Fiske caBed lhe five factors he discovered recurrent factors to enlphasize their similarity across lhe three data sources. He lhen conlpared his own five recurrenl

3. FISKE ANO TItE BtO-AVE FACfOR STRUCfURE

factors with the much larger number rcportcd by CattelI, and expressed his feelings about the correspondence bctwcen llié lwO solutions: "A thorough study of these two sets of factors, one from Catten and one fron10ur ratings by teamnlates, leaves one with feelings of both optinlism and discouragement. Even in the face of no cpmplcte congruence, the similarities support a belief in the possibility of eventual agreement upon the basic variables in personality" (p. 341). Note the similarity between that last sentence and the one quoted earHer fron1 1l1urstone (1934). who also obtained five personality factors.

PUITING IT ALL TOGETHER: THE DECADE OF THE 1960s

Before 1950, then, at least two separate five-factor models of phcnotypic personality traits had appeared in the scientific lilcrature-both with roots in !he lcxicographie tradition, but each using a differcnt set of trait-descriptivc variables. Fiske (1949) did nol cite Thurstone (1934). although he did cite Thurstone's textbook on multiple-factor analysis. Cattell (1947), who also did not cite Thurstonet reported the findings fronl ana1yses of his 35 variables. soon followcd by additional analyses oC the same or sinlilar variables in otber sanlples. In each of his analyses, CaUell extractcd and rotatedat least 12 factors. Noting that many of Cattell's factors wcre identified on the basis of quite low loadings, Fiske comnlentcd that he u ••• does not share this confidence in the significance of )OW

10adingsU (p. 342). History has proved the wisdom of Fiske's observation: Cattell' s factors have proved notoriously difficult to replicatc.

Among the nlany who tricd and failed were Tupes and Christal (1958, 1961), who analyzcd a variety of datasets, including those collected by Fiske. CatteU, and themselves. AH of the datasets included sets of variables from Cauell, and werc based on either peer or observer ratings. Using across-sample replicability as a criterion for deciding on the nunlber of factors. Tupes and Christal (1961) discovered five robust factors, which they labeled SllrgelJcy (Factor 1), Agreeableness (Factor 11), Dependability (Factor 111), EnJotional Stability (Factor IV), and Culture (Factor V). Their c1assic 1961 Air Force technical repart has now been reprinted (Tupes & Christal, 1992). along with an apprcciatory editorial by McCrae (1992) and a historieal introduction by Christal (1992).

Norman (1963), Borgatta (1964), and Snlith(1967, 1969) an reactcd to the work of Tupes and Christal (1958, 1961) with their own independent studies, and found much the same five-factor structure. Of these reports, Norman's article has been historically the most important, both bccause of his carefulness in analysis and reporting and because it was published in a highly visible and pres­tigious journal. Norman selected the four variables frotll the Tupes and Christal (1961) analyses that loaded most highly on each of the five factors, and then collected peer' ratings with those 20 variables in four new samples.He found such substantial factor replicability across the samplcs thathis five-factor solution has often been incorrcctly cited as the "Nonnan Five!'

04 GOLDBERG

Dorgatta used other variables in separatc analyses of maJe and female samples and found five replicated factors, which he Jabcled Asserriveness (1), Likability (II), Responsibility (IlI), Enlotionality (IV), and Inrelligellce (V). Using composite scores on each of the five factors, he compared seIf-ratings, self-rankings, and pccr rankings (the laUer two in samples conlposed of cIose fricnds, as wcll as in samplcs of acquaintanccs) in a multitrait-multimethod (MTMM) dcsign (Campbcll & Fiskc, 1959). Using the variables of CatteIl (1947), Smilh (1967) conlpared lhe structures derivcd from three large samples (N = 583 t 521, and 324) and found five robust factors, which he labeled EXlraversioll (I), Agreeableness (Il), Srrengrh ofCharacter (In), Enlotionality (IV), and Refinemellt (V). Moreover, Smith found that scores derived fronl the Character (ni) factor corrclatcd .43 with coUegc grades. In a latcr pccr-nomination study t Snlith (1969) rccovercd the four factors other than Refinen,ellt in nlale and female sanlples at bOth the junior high school and high school levels. Moreover, he found high correlations with snl0king status, smoking being negatively rclated to Factors n (Agreeableness) and III (Strenglh of Character) , and positively related to Factor I (Exlraversion).

TIME OUT: THE 1970s

D Y 1970, then, one might have assumed that the five-factor nlodel had anlassed more than enough evidence to ensure its viability as Ihe taxononlic framework for phenotypic personality traiL~. Dut it was nol lo be. Other forces dominatcd this decade, unlcashcd by a Zcilgeist that eschewed the very concept of personality traits, if nol of pcrsonality itself (c.g., Mischel, 1968). The 1970s witnessed the virtual abandonment of major segments of personality research, including the investigation of personality-trait structure.

THE RENAISSANCE: THE 1980s

In the spring of 1978, John Digman was to teach a COUI"Se in factor analysis, for which he obtained a number of corre)ation nlatrices frorn classic studies of abilities and personality traits, including those previously analyzcd by Tupes and Christal (1961) and Nonnan (1963). Before providing them to his students for reanalysis, however, he chccked them careful1y and found clerical errors in the matrices of two of Cattell's studies. More importantly, he discovered that when six or more factors werc rotated from lhe various matrices, the factors did not corrcspond; whereas when five factors werc rotated, there was striking interstudy correspondencc (Digman & Takemoto-Chock, 1981). Later rcsearch by Digman and Inouye (1986) again obtained the Big-Five factors in teachcrs' ratings of the children in their classrooms. This same period saw the publication of two of nly

3. FISKE AND TI-IE BIG-FIVE FACTOR STRUCTURE 35

ehapters (Goldberg, 1981 t 1982), both stimulated by the lexical hypothesis. TI1C

first provided sorne eonjectures about the possible universality of the Big-Fivc factor strueture, and the sccond described a program of research focusing on English trait-deseriptive adjeetives and nouns.

TIle 1980s wit~essed the emergenee of the two dominant variants of lhe five-factor nlodel",one developed by McCrae and Costa (1985a, 1987) ano operationalizco in the Neuroticism-Extraversion .. Opcnness Personality Inventory (NEO-PI; Costa & MeCrae, 1985), and the other associated with studies baseo on the lexieal hYPolhesis and operationaHzcd in the sets of factor nlarkers provided by Norman (1963), Pcabody and Goldberg (1989), Goldbcrg (1990, 1992), John (1989), Trapnel1 and Wiggins (1990), and Digman and his assoeiatcs (e.g., Dignlan, 1989; Dignlan & Inouye, 1986). Much is the sanle in both variants: (a) lhe nunlber of dinlensions is identical, namely five; (b) the eontenl of Factor IV is essentially lhe sanle, although it is orientcd in the opposite direclion in the two Jnodels, and is thus so labcled (Emorional Srability vs. Neuroticisln); and (e) therc is considerable sinli1arity, although not identity t in the eontent of Factor III (Collsciclltiousllcss). On· the olher hand, al leasl two of the differences bctween lhe véuianls are quite striking: (a) lhe localions of Faelors 1 and 11 are systenlatically rotateo, such thal Warmlll is a faeel 01' Exlraversioll in lhe NEO-PI, whereas it is a facet of Agrecableness in lhe lexical nlodcJ;2 and (b) Factor V is eoneeived as Openncss lo E~7}erienee in the NEO-PI and as Inlel/cet, or Imaginario" in lhe lexical nlodeI.

These differenees stenl fronl the history of the NEO .. PI, which starled out as a questionnaire nleasurc of a thrce-faetor nlodel. including only Ncurolicisl1J. EXlraversioll, and Openness 10 Expericncc. Whcreas olher three-faetor lheorists sueh as Eysenek (1991) have stood firm as proponents of their original representations t Costa and MeCrae reacted to the events of the early 19805 \Vilh sueh relllarkable Openness lo E~7}eriellce that,by the cnd of the deeade, lhese investigalors had beeome the world's nlost prolific and influential proponents of lhe five-faelor nlodel. TI1C prodigious oulpouring of rcports by McCrae and Costa probably did Jnore lo form the modem eonsensus aboul pcrsonalily structure lhan anything eIse that oeeurred during the 1980s. Specifieally, they used the NEO-PI seales as a franlework for integrating a wide varietyof other questionnaire scales, incIuding those developcd by Eysenek (McCrae & Costa, 1985b), Jackson (Costa & McCrae, 1988a), Spielbcrger (Costa & MeCrac, 1987), and Wiggins (McCrac & Costa, 1989b), as wcll as the scales included in lhe Minnesota Multipha~ic Personal ity Inventory (MMPI; Costa, Buseh, Zondcnnan, & MeCrae, 1986) ano the Myers-Briggs Typc Indicator (MeCrae & Costa, I989a).

2Actually. (he trait descriptor Warm has been classified in racel 11+/1+ in sorne AB5C (Hofslee. de Raad. & Goldberg. 1992) analyses in (he lexical model. whereas Warmlh is considered a 1+/11+ facel in lhe McCrnc and Cosaa model. suggesting more agreemenl when both primary and secondary loadings are considercd (han when one cOllsiders lhe primary loadings alonc.

06

THE MINORITY REPORTS: PRESENT-DA y OPPONENTS AND CRITICS

GOLDBERG

An enlergi'ng consensus is not the same as universal agreenlent 1l1ere are those who do not accept the Big-Five factor structure. Indeed, the two most famous holdouts, Cattell and Eysenck, share Hule but their opposition to the five-factor model. Catte11 renlains convinced that there are far nlore faclors than five, whcreas Eysenck is certain that five is too many. Specifically, Eysenck (1991, 1992) has argued that Faclors n (Agreeableness) and 111 (Conscientiousness) in the Big-Five represcntation are nlere)y facets of the highcr level construct of Psychoticism in his Psychoticism-Extraversion-Neuroticisnl (P-E-N) nlodcl.

SO WHY DIO IT TAKE SO LONG?

Despite lhe crities, any alert reader of the primary literature in personality psychology would be struck by the extent to which the Big-Five factor structure has gained recognilion and grudging acceptance as the major representation of phenotypic personality traits. Moreover, this franlework has begun to win converts in industrial/organizational contexts. Indeed, bOlh qualitative (e.g., Schnlidt & Ones, 1992) and quantitative (e.g., Barriek & Mount, 1991) reviews of lhe literature have concluded that personality nleasures, when classified within lhe B ig-Five donlains, are systenlatically related to a variety of criteria of job pcrfornlance. If it is fair to say that lhe five-factor nlodel is now gencrnlly seen as lhe mosl useful organizalion of personality lraits (e.g., Dignlan, 1990), lhen it is rcasonable to ask, what took it so long to be accepted? What force s kept the field fractionated for al1 lhose ycars? In the remainder of lhis chapter, 1 brieny outline sonle of the pressures against consensus that characterized lhe history of personalily psychology in the years since Fiske (1949). My hope is that, by understanding these pressures, we can mitigate deJays in the acceptance of rulure nlodcls and procedures, thus allowing the field to proceed on its scientific course Inorc expeditiously.

Ooing Ir AII by Hand

Certainly one of the forces that worked against ear)y rcplications of lhe five-factor nl0del was the sheer c.lifficulty of the calcuJations prior to the widespread availabiJity of reasonably efficient conlputers and standardizcd COtllputcr progranls. From Thurstone (1934) through Fiske (1949) to Tupes and Christal (1961), factor analyses were carried out by hand, which virtuaIly guaranteed the possibility of sorne clerical errors, as Digman and Takemoto-Chock (1981) discovered when they reanalyzed sonle of the datasets provided by Cattell. Tupes anc.l Christal included lhe findings fronl one computer analysis in their report,

3. FISKE ANO THE BIG-FIVE FACTOR STRUCfURE 07

and were able to show its conlparabiJity with tlleir painstaking hand .. ca1culations. However. by the time of Norman (1963), computers were hcre to stay, and today this problem has all bUl disappeared.

lock of Follow .. Up ,

Had Thurstone (1934) more fully enlbraccd lhe lexical hypothesis and realized the significance of his ¡nitial discovery, lhe history of the Big-Five model might have been quite different. But Thurstone devoted hinlself to other scicntific pursuits, as did Fiske, Borgatta, and Smith. Seenlingly. tllesc investigators did not reaBre lhe inlportance of their discoveries. Why? Here are sorne spcculative possibiHties.

Doing Our Own Things

111C ficld of personality psychology has faced continucd difficulties in establishing a curnulative scientific record. Perhaps bccause of lhe academic reward system, personalily psychologists lend lo function nlore Iike a nlotley array of superstars lhan. like a srnooth .. functioning lea m. Each of us has our own agenda. and intcgralion is not our long suil. Indeed, we do nol even agree about lhe mosl inlportant scientific questions, much less the answers. We estcenl our own contributions, and we are quick to pick holes in the contributions of others. Each of us scems to live by lhe moUo, If I didn't do U, it wasn't done right.

One result of this egocentric bias has bcen the proliferation of taxononlic nlodels. Very rarely does lhe developer of a hlodel admil that a rival nlorlel is superior. As a conscquencc, we engage in wars of attrition: Models disappcar only when lhcir originators fade away. Thosc who are not ¡nvolved in rescarch on lhe structurc of individual rlifferences arc justifiably bewildered by lhe diversity of our conlpeting models and the vitriol of our arguments in favor of our own conlributions. Whom are they to be1 ieve?

For we are bctter critics than integrators, bcttcr at finding holes lhan palching thenl. One could argue that the gradual acceptance of lhe Big ... Five modeI grew out of our continued attenlpts toreplaee it with sonlething more altractive. aH of which faBed. For examplet Nonnan, whose 1963 article is the single-nlost ciled reference to the Big-Five structure, spent much of his carly research career as a skeptic. ParadoxicaIly, altl10ugh his inlportance in the history of the five-faetor mode1 is universal1y acknowledgcd, his refusa) to bccome a truc believcr has typically becn overlooked. Yet after his seminal studics confirming the five-faclor nlodel with a se)celcd set of Cattell variables (Nomlan, 1963), he instituted an cxtensive research program ainlcd at replacing that rnodel with a nlore eonlprchensivc one. He began by cxpanding the corpus of English personaJity tenns assembled by Al1port and Odbert (1936), lhen c1assifying lhe ternlS in lhe expanded pool into su eh categories as slales, traits, and roles, and finally collccting nonnativc ¡nfonnation about sonlC 2,800 trait-deseriptive temlS

08 GOLDBERG

(Nonnan, 1967). Nonnan was convinced that, because of the inevitable conlputational and other technicallimitations of research in the 1 930s and 1940s, CattclPs variables len much to be desired, and therefore studies utilizing a rcprcsentative subsct of thc total English pcrsonality-trait lexicon would uncovcr sonle broad dinlcnsions beyond thc Big-Five struclure. AlLhough Nornlan never tested this appealing conjecturc, others did (e.ge, GoJdberg, 1990), and it hRIi proyed to be wrong.

Failure to See the Forest From rhe Trees

Digrnan espoused a lO-factor modcl of child personality structurc as latc as 1977 t and Cattell still espouses an even more complex representation. lloth invcstigators followed a "bottonl-up" strategy for conslructing a hierarchical rnodel, llsing oblique rotations lo oblain correlated factors, which in tum could then be analyzed obliquely al successively higher levcls. In contrast, investigators such as Tupes and Olrislul (1961), Nonnan (1963), and Goldberg (1990, 1992) followed a "top-down" slralegy, using orlhogonal rotations so as to firsl establish the oyerall dirnensionality of their variable sets, and only later atternpting to discover lhe facets wilhin each broad factor dornain (Hofstee, de Raad, & Goldbcrg, 1992). By analogy, the fonner strategy builds up the forest on a tree-by-tree bRc;is, whereas lhe latter.strategy first locales lhe outlines of the forcst and lhen focuses on lhe individual trees. Using a geographical analogy, the fOlmer strategy first finds such distinclive features of lhe landscape as lakes, rivers, and nlountain rimges, which in turn are used to discover lhe continents, where,lS lhe laller strategy first locales lhe continenlli and later nlaps lheir distinctive features. Which is lhe best way lo proceed?

If the history of the five-factor rnodel is to be one's guide, one should place one's bets on lhe top-down strategy. As Dignlan (1990) noted, when factors are rotated obliquely to a simple-structure criterion, they are likely lo be associaled with relatively hornogeneous clusters of variables, rather than lhe basic dirnensions of lhe multivariate space. Whereas il is possible lhat the two strategies could converge on a cornrnon representation, this would be nlost unlikely unless lhe ¡nitial selection of variables had been carried out with extraordinary prescience. That is, to select the initial variables so thoughtfully requires knowledge of the nurnbcr of overall factor dornains and lhe naturc of lhe facets located within each dorna¡n. One is not like)y lo know all this in advance.

Failure ro Separare Signal From Noise

As early as 1961, Tupes and OlfistaJ caBed attention lo lhe difficulties invo)vcd in dcciding when two structural reprcscntations were lhe sanle ano when they truly oiffcrcd. In discussing their own finding that five factors replicated across the diverse datasets that they analyzcd, they cOO"lmented:

3. FISKE ANO iHE B1G-FIVE FACTOR STRUCTURE 09

In many ways it sccms rcmarkable that such stability shouJd be found in an area which to date has granted anything but consistent results. Undoubtedly lhe consistency has always beCn thcre, bUl it has becn hiddcn by inconsistcncy of factorial lcchniques and philosophics, the lack of replication using identical variables, amI disagrccment among analysts as lo factor titles. None of lhe factors idcntificd in this s,tudy are new. They have becn idcntificd many times in previous analyscs, although they havc nol always becn callcd by thc sanlC names. (Tupes & Christal, 1961, p. 12)

Littlc did those investigators realize that it would lake yet another 25 years for lhe ficld to separate lhe Big-Five signal from the noise of alternative factor representations. Inconsistency of factorial techniques and philosophies have renlained, and have increased over tinle. On the other hand, lhe widcsprcad availability of cotnputer progrmns forconducting a standard analysis (such as principal cornponents rotated by varimax) has servcd to decrcase lhe use of idiosyncratic proccdurcs. In any case, wc now know that nlost reservations about the fragility of factor solutions are unfounded, provided that the datasets ¡nelude substantial nunlbcrs of variables and sanlples of largc sizc. Por example, Goldbcrg (1990) dcnlonslratcd lhe virtual idenlity of Big-Five factor loadings across 10 factorial procedurcs, incJuding bOlh factor anuconlponent analyses, and both orthogonal and obl ¡que rotational algorithrns.

A sccond problenl identificd by Tupes and Christal (1961) was a )ack of replications using identical variables. Indecd, it hasproved 10 be extrcrnely difricult lo cunlulatc concJusions fronl single-shot studics with uniquc scts of variables. In lhe history of lhe Big-Five factor SlruClure, the studies thal have provcd to be most persuasive to critics of the nlodel were those that conlpared the factor slructurcs from similar sets of variables across divcrse samplcs or procedurcs (e.g., Costa & McCrae, 1988b; Dignlan & Takemoto-Chock, 1981; Fiskc, 1949; Goldbcrg, 1990; McCrae & Cosla, 1987; Norman, 1963; Tupes & Christal, 1961).

lllC final problelll identified by Tupes and Christal waS that of disagrecnlcnt anlong analysts as to factor tilles. Everyone scems to have a favorite ·labcJ. espccially for the Big-Pive factors, which are c1early loo broad to be captured wcll by any single trait-descriptive teml. Critics of lhe rnodel are quick lo nole the discord an10ng investigators in lheir factor Jabcls, and to attribute such diffcrences in labcls to fundamental differences in the undcrlying factor structures. However, n10st of the diffcrences in labcls rcsult from individual styles or tastes (e.g., COllscienliollslless vs. WiU for Pactor 111). One difference inlabcls docs renect a sonlewhat different conception of the underlying factor: Factor V is labcJcd as Openness lo Experience in the McCrae and Costa variant of the five-factor nlodel, and as Illtellect in the lexical variant. Actually, as suggestcd by Saucicr (1992), neither Openness nor Illlellect captures weJl the central cluster of traits thai define Factor V. Perhaps a more apt tabe) is Imaginarion for a dornain incJuding such traits as Crealivily and Curiosity.

40 GOLDDERG

A Malignant Zeitgeist

In 1961, Tupes and Christal had no way of knowing that the 1970s would witness a virtual moratorium on the investigation of many fundamental problems in personality psychology, inc1uding the search for a compclIing taxononly of personality traits. However, the 1970s turoed out to be a decade of controversy about the scientific status of traits in general and broad traits in particular, stimulated by such critics as D' Andrade (1965), Mischel (1968), and Shweder (1975). Many of the issues ¡nvolved in this controversy may have now been resolved (e.g., Borkenau, 1992; Funder, 1991), with optinlistk implications for the eventual transformation of personality psychology into a nlore cumulative seien­tifk enterprise.

CODA

Fiske was nluch ¡nvolved in the controversies of the 1970s, and he was profoundly infiuenced by the behavioristic Zeitgeist that was swccping the field at thallinle. Perhaps as a reaction to lhe "nattering nabobs of ncgativism" who were punlnleling personality psyehology t he lost ¡nlerest in broad pcrsonalily traits (e.g., Fiske, 1973, 1974), and began to argue in favor of more bchavioral unils of anaIysis. so as to ensure high levcls of interjudge agreement on lhe bask elements of pcrsonality. Moreover, he began to feel quite pessimistic about lhe likelihood of ever reaching agreetnent on a taxonomy of personality traits, with the result that he did not further pursue his seminal study of trait factor struetures. Today t Fiske is renowned for his methodological and philosophkal contributions to the ficld of personality. Wouldn't it be ¡ronie if historians of lhe future also judged his 1949 empirical study as bcing of such signa1 importance that it served as a verification of his earlier "belief in the possibility of eventual agreement upon the basic variables of personal ity" (Fiske, 1949 t p. 341)1 If so, he has lhe right lo ask, What lhe hell look so long?

ACKNOWLEDGMENTS

Preparation of this chapter was supporled by Grant MH-49227 from the Nalional Institute of Mental Health. 1 am indebted to A. Timothy Church, Jacob Cohen, Paul T. Costa, Jr., Lec J. Cronbach, Robyn M. Dawes, Qlarlcs F. Dkken, John M. Digman, Donald W. Fiske, Susan T. Fiske, Sarah E. Hampson, Willem K. B. Hofslee, Oliver P. John, Daniel Lcvitin, Maurice Lorr, Clarence McCormkk, Ivan Mervielde, Warren T. Norman, Delroy L. Paulhus, Lawrence A. Pervin, James A. Russell, Gerard Saueier, Mkhael F. Scheier, Patrick E. Shrout, and Auke Tellegen for lheir thoughtful suggestions. Portions of this chaptcr have

J. FISKE ANO TI-IE BlG-FlVE FACfOR STRUCfURE 41

bccn adaptcd from Goldbcrg (1993), which can be consultcd for furthcr dctails about thc Big-Fivc factor structurc and its historical dcvclopn1cnt.

REFERENCES

Allport. G. W .• & Odbcrt. H. S. (1936). Trail-names: A psycho-Iexical sludy. Ps)'dlOlogical MonogrofJhs, 47 (1, Wholc No. 211).

Barrick. M. R •• & Mount. M. K. (1991). The Big Fivc pcrsonality dimensions and job performance: A meta-analysis. PersO/Ille! PsydlOlogy, 44. 1 ... 26.

Borgatla. E. F. (1964). Thc structure of pcrsonaJity characteristics. Behavioral Science. 9. 8-17. Borkenau. P. (1992). Implicil personality theory and lhe five-faclor model. JourtuJl of Personality.

60. 295-327. Campbell. O. T .• & Fiske. O. W. (1959). Convergent and discriminant validalion by (he

multitrait-multimethod matrix. PsydlOlogic:al Bullel;n, 56. 81-105. Cantor. N .• & Mischel. W. (1979). Prototypes in person perception.ln L. Berkowitz (Ed.), Ad\·ance.f

ill experimelltal sodall'syt:llOlogy (Vol. 12. pp. 2-52). San Diego. CA: Aeademic Prcss. Callell. R. B. (1943). The deseription of personality: Baste traits resolved into cluslers. Jourllal of

Abllormal ond Sodal PsydlOlo¡:y. 38. 476-506. Canell, R. U. (1947). Confirmation and clnrificalion of primary personality faclors. PsydlOmelril.:a.

/2, 197-220. Christal. R. E. (1992). Aulhor's note On "Recurrent persona lit y ractors based en traíl ratings." Joumal

of Personality. 60. 221-224. Costa. P. T .• Jr., Busch, C. M., Zonderman. A. B., & McCrae. R. R. (1986). Correlations of MMPI

faclor scales with measures of the five factor model of personality. JuurtuJl of Ptrsonality Aues.fmelll. 50. 640-650.

Costa, P. T .• Jr., & McCrae, (~. R. (1985). Tire NEO Persollality Im·tlllory malllllll. Odessa. FL: Psychological Assessmenl Resources.

Costa. P. T .• Jr .• & McCrae. R. R. (1987). Personality assessment in psychosornatie medicine: Value of a lraÍl taxonomy. In G. A. Fava & T. N. Wise (Eds.). Advanc'es illl'syc:Jwsomalk medicine: Vol. /7. Researd, I,aradi¡:"'s illl,sydlOsomal;c medic;lte (pp. 71-..82). Basel, Swilzerland: Karger.

Costa, P. T .• Jr •• & McCrae. R. R. (1988a). From eatalog lO classification: Murray's needs and lhe five-factor model. )oufl,al of Persolla/ity and Social Psyehology. 55. 258-265.

Costa. P. T., Jr .• & McCrac. R. R. (1988b). Personality in adullhood: A six-year longiludinal study of self-reports and spouse ralings on lhe NEO Personalily Inventory. Journal of Persolfalily and Scx..'ial Psyelroloxy. 54. 853-863.

O' Andrade, R. G. (1965). Trail psychology and eomponential analysís. AmericlllJ AltlhrOI)()lo¡:isl. 67.215-228.

Oigman. J. M. (1989). Five robusl trail dimcnsions: OeveJopmenl, stabilily. and utility. Jourltal uf /)er.wnalily. 57. 195-214.

Digman. J. M. (1990). Personality structure: Emergence of 'he five-faclor model. In M. R. Rosenzweig & L. W. Porter (Eds.). A""ual review of psyclw/u¡:y (Vol. 41. pp. 417 ... 440). Palo Alto. CA: Annual I~eviews.

Oigman. J. M .• & Inouye. J. (1986). Further spccification of lhe five robust faclors of personality. Juumal of Personality all(1 Scx..·ial PsydlOlo¡:y. 50, 116-123.

Digman. J. M •• & Takemolo-Chock, N. K. (1981). Factors in lhe natural languageof personality: Re-analysis. comparison. and interpretation of síx major studies. Mulli,'ariale Belra";oral Researe". /6. 149-170.

Eysenck. H. J. (1991). Dimensions of personality: 16. S. or 3?-Criteria for a taxonomic paradigm. Per.wnalily.alUl lI,di"idllal Dij)erences. 12. 773-790.

42 GOLDBERG

Eysenck. H. J. (1992). rour ways five faclors are 1101 basic. Personality and J"d;~'id'lQl DW'crenc:cs, IJ.667-673.

Fiskc. D. W. (1949). Consislcncy of lhe factorial struclures of personalily ralings from differcnl

sources. 10urllal of Ab"ormal a"d Sodal PsycJlOlogy. 44, 329-344. Fiske. D. W. (1973). Can a personality construct be validated empirically? PsycllOlogic:al Bulle/i",

80.89-92. Fiske. D. W. (1974). The IimUs for lhe convcnlional scicnce of personaJity. Joumal 01 Pcr.'iOIlality,

42, 1-11. Funder. D. C. (1991). GlobaltraÍls: A neo-A 11 portian apprCh1ch lo pcrsonality. Psyc:110108ical SdCIICC,

2.31-39. Galton. F. (1884). Measuremcnt of charaeter. Forllf;8hlly Review, J6. 179-185. Goldberg. L. R. (1981). Language and individual differences: The search for universals in personality

Icxicons. In L. Wheeler (Ed.), RevietV cifl'ersollalily a"d sodal psychology (Vol. 2, pp. 141-165). Beverly Bilis. CA: Sage.

Goldberg. L. R. (1982). From ace lo zombie: Sorne exploralions in lhe language of personality. In C. D. Spiclbergcr & J. N. Butcher (Eds.), Allvalu:c./f in I'cr.wllalily aS.U.umCIII (Vol. 1, (lp. 203-234). HiIlsdale. NJ: L1wrenee Erlbaum Associates.

Goldhcrg, L. R. (1990). An altemalive uDescription of pcrsonalily": ll1e Dig-Five factor stnacture. Joumal of I'crsollalily all¿ Social Psychol08Y. 59. 1216-1229.

Goldberg. L. R. (1992). 111e devcJopmcnt of markers for lhe Big-Five fáclor slructure. l'sYc:/10108ic:al A.uc.umclfI, 4, 2()-42.

Goldhcrg, L. Ro (1993). ll1e slruclure of phcnolYpic pcrsonality tmíls. Amcricm, I'sydw/08i:t1. 4t't. 2()-34.

Ilmnpsoll. S. E.o John, O. P .• & Goldbcrg, L. It (1986). Calcgory brcadlh and hierarchieal Slnaclure in personalily: Studies of asymmelries in judgmenls of lraÍl implications. JOllrl.al o/ Persollalily alld SIH.:ial P syellology. 5/, 37-54.

Ilofslec, W. K. B .• de Raad. B., & Goldberg. L. R. (1992). Illlegralion of lhe Big-Five and cireumplcx apprChlches lo (mil struelure. Joumal of Pcrsollalily and Sodal Psyd10108Y. 6J. 146-163.

Jolm, O. P. (1989).' Towards a laxonomy of personalily descriptors. In D. M. Buss & N. Cantor (Eds.). /'ersollalily l,sycI10108Y: Rece," Irelldf a"d cmcr8i"8 dircl'liolls (pp. 261-271). Ncw York: Springer-Verlag.

Jolm. O. r. (1990). Thc UBig Five" factor taxonomy: Dimcnsions of pcrsonality in lhe naturallanguage and in questionnaircs. In L. A. Pervin (Ed.), /Ialldhook (Jll,crsollalily: r¡'eory alUl rescare" (pp. 66-100). New York: Guilrord.

John. 0.1' .. Hampson. S. E .• & Goldbcrg, L. R. (1991). ll1e basie levcl in personality-trail hierarchies: Studies of trail use and aceessibility in differenl conlexlS. JOllmal of Pcrsollalily allll Sodal

Psydlolo8Y. 60, 348-361. Kelly. E. L., & Fiske. D. W. (1951). rile predic:lioll o/I'er/ormmu:e in dillkal I's)'('/10108Y. Ann

Arbor. MI: University of Michigan I'ress. McCrac. R. R. (1992). Editor' s introduClion lo Tupes and Chrislal. JOllma/ (if Per.wllalilJ'. 60.

217-219. McCrae. R. R., & Costa, P. T., Jr. (1985a). Updating Norman's "adcquate laxonomy": Intelligcnce

and personality dimensions in natural languagc and in questio.nnaires. Joumal (Jf Pcr.wnalily and Sodal P./fyC:/10108Y, 49, 710-721.

McCrae. R. R, & Costa, P. T., Jr. (1985b). Comparison or EPI and psycholicism scalcs wilh measures of lhe (¡ve-factor model of personalily; Persollalily a"d Indh'idlllJl Dij)'erenc:cs. 6. 587-597.

McCrac~ R. R., & Costa, P. T., Jr. (1987). Validation of lhe five-ractor modcl of personality across instruments and observcrs. Jounwl 01 Persollalily alld Social Psychol08Y, 52. 81-90.

McCrae. R. R .• & Costa. P. T •• Jr. (1989a). Reinterpreling lhe Myers-Origgs Type Indicator from lhe pcrspectivc of lhe five-faclor model or personaJity. Jourllal o/ Per.'iOIlality, 57, 17-40.

McCrae. R. R .• & Costa. P. T .• Jr. (1989b). The struclure of interpersonallrails: Wiggins' circumplex :too lhe five·faetor model. Joumal o/ PersOIUllity a"d Social Psyc¡'oI08Y, 56, 586-595.

3. FISKE AND TI-IE BIG-FIVE FACfOR STRUcruRE 43

McCrae. R. R., & Jolm, O. P. (1992). An introduction lo lhe nve-factor model and its applications. J ourlUll 01 Persollality, 60, 175-215.

Mischel, W. (1%8). Pcrsonality and asscssnrenl. New York: Wiley. Norman, W. T. (1%3). Toward an adcquate laxonomy of personalily atlributes: Replicalcd factor

structure in pec'r nomination pcrsonality ratings. Journal 01 Abnormal and Social Psychology, 66. 574-583.

Norman, W. T. (1967). 2800 pcrsollality Irail dcscriptors: Normalive 0l)(!raling characlcristics lor a university IXlIJUlalioll. Ann Arbor, MI: Dcpartmcnlof Psychology, University of Michigan.

Peabody. D., & Goldbcrg. L R. (1989). Sorne delcrminants of factor structures from pcrsonality-lrait descriptors. Journal 01 Persotrality alld Social Psyclwlogy, 57.552-567.

Saucicr, G. (1992). Openness versus ¡ntelleca: Much ado about nothing? European Journal 01 PersonalilY, 6, 381-386.

Schmidl, F. L.. & Ones, D. S. (1992). Personnel selection. In M. R. Rosenzweig & L. W. Portcr (Eds.), A""ual rcview OllJSyclwlogy (Vol. 43, pp. 627-670). Palo Alto. CA: Annual Reviews.

Shweder. R. A. (1975). How relevanl is nn individual dirrcrcncc theory of personality? Journal 01 PersolJality. 43. 455-484.

Smilh, G. M. (1967). Usefulncss of pcer ralings of pcrsonality in cducational rcscarch. Educalional and PsycllOloJ:ic:al Measureme,,', 27, 967-984.

Smilh. G. M. (1 %9). Rclations bctwecn personality and smoking bchavior in pre-adult subjects. Journal tI/CoII.mlli"R and Clínical P.fycllOlogy, J3, 710-715.

Thurstone, L. L. (1934). The veclors of mind. Psych(JloRical Review, 4/, 1-32. Trapnell. P. D .• & Wiggins. J. S. (1990). EXlensionof lhe Inlérpersonal Adjcclive Scales lo ¡ndude

lhe Uig Five dimcnsions of pcrsonality. JourfUll 01 I'trsollality and Social !'sycllOloRY. 59. 781-790.

Tupes. E. C .• & Christal, R. E. (1958). Stability of personality traÍl raling factors obtaincd undcr diverse conditions (USAF WADC Tech. Note No. 58-61). L.ackland Air Force Base. TX: U.S. Air Force.

Tupes. E. C., & Chrislal. R. E. (1961). Recurrent personality factors bascd on trait ratings (USAF ASD Tech. Rep. No. 61-97). L.ackland Air Force Base. TX: U.S. Air Force.

Tupes, E. C .• & Christal. R. E. (1992). Recurrent personalily factors based on trail ratings. Journal 01 Persollalíty. 60. 225-25 l.

Wiggins, J. S., & Pincus, A. L. (1992). Personality: Slruclure and assessmcnt. In M. R. Rosenzwcig & L. W. Porter (Eds.), Amulal rcview 01 psychology (Vol. 43. pp. 473-504). Palo Alto, CA: Annual Rcviews.

Wiggins, J. S •• & Trapncll, P. D. (in press). Personality structure: The retum oC the Big Ave. In R. Bogan, J. A. Johnson, & S. R. Briggs (Eds.), lIalldhook 01 persona/ity psychology. Orlando. FL: Acadcmic Prcss.