five-factor personality description

Upload: thuy-huong-ton

Post on 06-Jul-2018

226 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/18/2019 Five-factor Personality Description

    1/29

    Psychological Bulletin

    1995 Vol 117, No. 2,  187-215

    Copyright 1995 by

     the

     American Psychological  Associat ion, Inc.

    0033-2909/95/S3.00

    A Contrarian View of the Five-Factor Approach to

    Personality Description

    Jack Block

    University

     of California, Berkeley

    The

     5-factor approach

      (FFA) to

     personality description

      has

     been represented

      as a

     comprehensive

    and

     compelling rubric

     for

     assessment.

      In

     this

     article,

     various misgivings about

     the FFA are

     delin-

    eated. The algorithmic method of factor analysis may not provide dimensions that are incisive. The

      discovery

    of th five

     factors

     may be  influenced  by

     unrecognized constraints

     on the

     variable sets

    analyzed. Lexical

     analyses

     are

     based

      on

     questionable conceptual

      and

     methodological assumptions,

    and

     have achieved uncertain results. The questionnaire version of the FFA has not demonstrated the

    special

     merits

     and

     sufficiencies

      of the five

     factors settled upon. Serious uncertainties have arisen

     in

    regard to the

     claimed 5-factor structure

     and the

     substantive meanings

     of the

     factors. Some implica-

    tions of

     these problems

     are

     drawn.

    During

     the last decade, the Big-Five approach has begun

    to loom large in the field of personality psychology. It is being

    said that rapid progress  has  been made toward  a  consensus

    on personality structure

    (Costa

     & McCrae,  1992d, p. 344).

    Goldberg (1992) has talked of a quiet revolution occurring in

    personality

      psychology.... An

     age-old scientific

     problem has

    recently begun to look tractable.. . . Gradually, agreement has

    been growing about the number of orthogonal factors needed

    to

      account

     for the

     interrelations among English-language trait

    descriptors

    (p.

     26). The contention is that, via the mathemat-

    ical method of factor analysis, the basic dimensions o f person-

    ality

     description have been

      discovered : Their

     number is five,

    and their nature can be summarized by the broad concepts of

    Surgency, Agreeableness,

     Conscientiousness, Emotional Stabil-

    ity

      versus Neuroticism, and Openness to Experience (John,

    1990,

      p.

     96).

     Digman (1990),

      reviewing

      the field, also cele-

    brates  the  emergence of  the  five  robust factors  of

    personality.

    Personality psychologists

     are

     being asked

     to

      accept this spe-

    cific set of five

     orthogonal factors

     and to use

     these factor dimen-

    sions as the conceptual structure for descriptively representing

    different  personalities. Widely,

     frequently,

      and  enthusiastically

    promulgated  by vigorous, resourceful, talented, ingenious ad-

    herents,

      and  with

      support mustered

      from

      many

      and

      various

    studies, the five-factor approach  (FFA) has achieved apprecia-

    ble

     popularity; a tide seems underway in the field. Adoption of

    the FFA as the universal

     framework

      for personality description

    This article benefited greatly

      from  the

      counsel

      of a

     number

     of

     col-

    leagues.

     I must exculpate them regarding

     its rema ining deficiencies.

     M y

    especial thanks go to Lew Goldberg,

     David Harrington, Robert Hogan,

    Oliver

     John, Robert McCrae, Philip Shaver,

     and  Auke

     Tellegen, among

    others.

    Preparation

     of this

     article

     was

     supported

     in

     part

     by

     National Institute

    of

     Mental Health

     Grant MH

      16080.

    Correspondence concerning this article should be addressed to Jack

    Block,

     Department

      of

      Psychology,

      University  of

     California, Berkeley,

    California 94720-1650.

    would,

     of course, fundamentally shape the subsequent course of

    thinking and

     research.

    In

     the words of advocates of this argument, the five factors

      are both necessary and reasonably  sufficient  for describing  at

    a global level the major features of personality (McCrae &

    Costa, 1986,

     p. 1001); the five-factor approach

     provides

      a

     uni-

    versal descriptive fr am ew or k. . . for the comprehensive assess-

    ment

     of

     individuals (McCrae, 1989,

     p.

     243); the

     five-factor

    model developed in studies of normal personality is fully  ade-

    quate to account for the dimensions of abnormal personality as

    well (Costa & McCrae, 1992d, p.

     347).

     Why are there five and

    only

     five

     factors?

      We

     believe

     it is an

      empirical  fact,  like

      the

    fact that there are seven continents on earth or eight American

    presidents

      from

      Virginia (McCrae

     & John, 1992, p. 194).

    The claims of an emerging consensus about the FFA have

    also,  after  a lag, prompted some expression  of concerns about

    the FFA

     (cf.,

     e.g., Ben-Porath &

     Waller,

     1992a,  1992b; Eysenck,

    1992;

      Hough, 1992; McAdams, 1992; Mershon & Gorsuch,

    1988; Tellegen, 1993; Waller & Ben-Porath,

      1987),

      about the

    claims of the approach, its nominal empirical basis, and its sci-

    ence-directing implications. What is meant by the phrase, ma-

    jor

      features

     of

      personality,

    and

      what criteria should serve

     to

    ensure their identification? What does

     the

     term,

      global,

    de-

    note?

     How

     does

     one

     conclude that

     a

     description

      is

      reasonably

    sufficient or  comprehensive or

      fully

     adequate ? How com-

    pelling

     and indisputable were the procedures by which the five

    factors were discovered and settled upon? What is the role of

    concept

     and

     theory

     in the field of

     personality psychology?

    The present article tries to make explicit and bring together

    some of the reasons—methodological, empirical, semantic, the-

    oretical—underlying my own discontent with the FFA. Some

    prefatory remarks are necessary in order to convey the personal

    perspective  from  which I shall be speaking of the FFA. Not all

    personality psychologists

     will

     subscribe to my views.

    My  concern is, like the concerns of FFA advocates, with the

    problem—it is really an ideal—of establishing a set of con-

    structs for the scientific description of personality (see, e.g.,

    Block,  1961; Block & Block, 1980). For a set of constructs to be

    187

  • 8/18/2019 Five-factor Personality Description

    2/29

    188

    JACK BLOCK.

    scientifically

      sufficient,

      a  number  of  criteria  will  need  to be

    jointly  served.  Th e  psychological  constructs—often  primitive

    terms—should  receive  sufficient  elaboration (necessarily,  of

    course,

     via

     words)

     so

     that their meaning

     can

     attain  consensual-

    ity  among psychological scientists.  Th e  constructs should  be

    sufficient  in number so that th e personalities of individuals and

    the

      dynamics

     of

      behaviors

      can be

     represented

      in

      articulated

    ways a nd discriminating explanations and predictions  may be

    formulated. They should

     demonstrate

     a

     superior

     usefulness in

    prediction  or in economy of conceptualization over competing

    sets  of constructs.  Th e  constructs should  be theory-reflecting

    rather than constructs issued

     a nd

     controlled

     b y

     confounded

     so-

    cietal evaluations. As scientific constructs, they should be for-

    mulated with no necessary regard  fo r their usability and under-

    standability  by laypersons. These several  criteria constitute a

    tall  order

      of

      achievement

      and it is not

      suggested that their

    achievement is

      imminent; rather, they serve

      as a

      guiding

    aspiration.

    Some words are also d ue about  the terminology used in this

    essay.

      The Big

     Five

      are  often

      called the

      five-factor  mo el  of

    personality

    and are

     referred

     to as

     providing

     an

      understanding

    of

     the

      structure

      of

     personality.

    As the

     term  model

    is

     used

    in conventional parlance among psychologists, it means a theo-

    retically  based, logically coherent, working representation

      or

    simulation that,

     in

     operation, attempts

     to

     generate psychologi-

    cal phenomena of interest. However, no identifiable  hypotheses,

    theories, or models guided the emergence of or decision on this

    five-fold  space (although some have been  offered post hoc).

    There is no theoretical reason why it should be these five [ factors]

    rather than some other

     five. . . .

     [There were]

     no a

     priori predic-

    tions

     as to  what factors should emerge,  and a coherent an d  falsifi-

    able  explanation  for the five factors  has yet to be put  forward.

    (Briggs, 1989, p. 249)

    Because the Big Five formulation was entirely atheoretical,  us-

    age of the term

      model

    may be premature.

    Moreover, Big Five research  has  been based only  on the re-

    lations among

     a set of

     variables across individuals, what Cattell

    (1946) ,

      in his incisive formulation of the  data box, termed

    the R-technique (or

      variable-centered )

     approach

     to the

     anal-

    ysis

      of

     personality data.

      As he

      observed, although

      the

      R-ap-

    proach

     to

     data analysis

     is

     important, there

     are

      other important

    ways of looking at personality. In particular, it should b e recog-

    nized  that  no  matter  ho w

      satisfying

      on  descriptive  or  other

    grounds

     the variable-centered factor structure  of the FFA may

    be, it  cannot represent  a  personality structure. Personality

    structures  lie within individuals (see also Block, 1971, p. 13;

    John,  1990,

     p.

     96).

     It is the

     personality structure

     of an

     individ-

    ua l

     that, energized by  motivations, dynamically organizes per-

    ceptions, cognitions,

     and

     behaviors

     so as to

     achieve certain sys-

    tem goals.  No functioning psychological  system, with  its

    rules and bounds,  is designated  or implied by the  Big

      Five

    formulation;

      it

     does

     not  offer  a

     sense

      of

     what goes

      on

      within

    the  structured, motivation-processing, system-maintaining

    individual.

    On this analysis, it

     follows

     that a more appropriate,  and more

    limiting,

     label for the  Big Five body o f work is something like

    the  inordinately cumbersome phrase, the five-factor,

     variable-

    centered

     approach

      to personality descriptions. Of

     practical

     ne-

    cessity, in the following discussion reference will be made to the

      five-factor

      approach, but it  should  be  understood that this

    phrase is shorthand for a lengthier, more awkward, but  perhaps

    more

     fitting,

     wording.

    In

     what follows,

      I

     often  deliberately write

     in the first

     person

    instead  of  using the  traditional third person.  In  much  of the

    literature presenting

      and

      evaluating

      the

      FFA,

     opinions—in-

    formed  opinions  bu t  nevertheless  opinions—inevitably  have

    been, are,

     and

     will continue

     to be

     offered.

     In my own

     remarks,

    rather than using

     the

     seeming objectivity

     of

     third-person phras-

    ing,

     the

     egocentric

      I

    will serve

     to

     identify personal views.

     Be-

    cause previous interpretations of the FFA have come primarily

    from  its proponents, it may be useful  to provide a more dubita-

    tive

     reckoning.

     FFA

     advocates certainly will wish

     to and are en-

    couraged to dispute my interpretive account. The ensuing dia-

    logue can only be constructive and advancing regarding the is-

    sues involved.

    No  apologies are  offered  for the specifics to be presented.  Of-

    ten  slighted, they provide a basis for evaluating certain central,

    highly influential studies.  If the shaping of the field of personal-

    ity

     is to be

     based

      on

     reasoning

      and

     evidence rather than claim

    and

     counterclaim,

      we

     must

     be

     religious about

     the

     details.

    The  structure  of this essay is as

      follows.

      Because  it has been

    said that the five-factor  theoretical

     model.

      . . [is]

      rooted

     in

    factor analysis

    (McCrae &

     Costa,

      1989a, p.

      108), some forgot-

    ten or

     slighted knowledge regarding this extraordinarily

      useful

    method

     is

     brought forward.

     I

     maintain that

     the

     task

     o f

     evolving

    the   theoretical constructs necessary  for the scientific study  of

    personality cannot  be entrusted solely to the  pervasively useful

    method  of factor analysis. The  origins and early history of the

    FFA are

     then reviewed. Concerns

     and

     questions

     are

      raised

      re-

    garding th e origins of the FF A in the work, primarily, of Cattell

    (e.g.,

      1943a, 1943b,  1945),  Tupes

      and

      Christal  ( 1992) ,

      and

    Norman (1963). I will suggest that, to the extent this early work

    on  ratings

      is

     said

      to

      provide

     a

      foundation

     for the

     FFA, subse-

    quent  edifices

      may be

     shakier than

     is now

     recognized.

      The ex-

    tensive lexical labors

     of

     Goldberg (e.g., 1977, 1981, 1982, 1990,

    1992)

      are

      then considered.

      The

      widely ranging research

      of

    Costa  and McCrae (e.g., 1985, 1992b; McCrae  & Costa, 1985,

    1987), which brought the FFA, until then based exclusively on

    lexical

     analyses

     and

     adjective ratings

     of

     self

     or

      others, into

      the

    questionnaire realm, is next examined. It is primarily these last

    two research programs, vigorously pursued and promoted, that

    have moved

     the FFA

     toward

     its

     current prominence

      in

      person-

    ality  psychology.  On the  basis  of  this series  of  evaluations,

    largely methodological  in nature, th e current status of the FFA

    is appraised,

      its

     logical

     and

      psychological problems discussed,

    and

      some positively oriented suggestions

     offered.  It

      should

      be

    noted that

      a

     number

     of the

     ideas brought forward

     in

     this

     effort

    are not

     novel; many come

     from

     others, generally cited

     but

     some-

    times perhaps

     not

     because

     I do not

     remember their origin.

    A

     Viewpoint Regarding

     the

     Method

     and

     Possibilities

     of

    Factor Analysis

    Because

      the FFA is

      acknowledged

      to be

      rooted

    in the

    method

      of

     factor analysis,

     an

     evaluation

     of the

      approach must

    be  predicated  upon  an  understanding  of the  method,  its

    strengths

     and its

     vagaries. Although factor analysis

     is a

     method

  • 8/18/2019 Five-factor Personality Description

    3/29

    A  CONTRARIAN

     VIEW

      OF THE  FIVE-FACTOR APPROACH

    189

    usable in diverse ways  in a variety of contexts to analyze the

    correlations

     among a set of

     variables,

     the problems or

     arbitrari-

    ness

     of the

     method

     are

     often

     not

     given sufficient  recognition.

    The mathematics of the method generates a succession of la-

    tent variables or factors that, subsequently, can be used, in re-

    verse,

     to regenerate the initial

     correlation

     matrix to any desired

    degree of precision.

     If

     as many principal component factors are

    extracted

     as

     there

     are

     variables

     in the

     beginning variable set,

     the

    reproduction of the initial correlation matrix i s perfect.

    There would be little point to the factor analytic method,

    however, if all it

     provided

     was as

     many factors

     at the end of the

    process

     as one had

     variables

     at the outset. It is

     because

     the

     seri-

    ally

     extracted

     factors

     necessarily decrease in

     their influence

     on

    reproducing

     the initial matrix that the method  can offer cogni-

    tive

     economy.

     If the

     extraction

     o f

     relatively

     fe w

     factors permits

    the reproduction of the initial correlation matrix reasonably

    well,

      then one may consider the information contained in the

    starting correlation matrix to be expressible or

      explained

    in

    terms

     of

     these relatively

     few

     factors.

     One can

     then seek

     to

      pro-

    vide

      psychological meaning (usually

      via

      factor

      rotation)  for

    these

     summarizing

     latent variables.

    As

     a method  of  data reduction, exploratory factor analysis

    can   often simplify  and  make reportable masses  of data—net-

    works

      of interrelations—otherwise

     cognitively

      unencompass-

    able.

     For some, the promise of factor analysis has seemed to be

    that the method would permit easy empiricism rather than

    tough theory to develop our  scientific constructs. However, it is

    necessary to know and to remember the rigid logic of the

    method

      and its

     consequent quirkiness when

      it is

     unwarily

     set

    down in the real world. The method can issue marvelous, pre-

    viously obscure

     connections,

     it can also issue

     mindless  results

    (e.g., see

     discussion

     of

     this issue

     by

     Armstrong, 1967; Lykken,

    1971).

     In considering the results

     afforded

      by factor analysis, one

    must

     be

     mindful

     o f the

     ways

     the method may suggest more than

    is supportable.  Herewith are some cautionary  remarks

     about

    factor

     analysis that derive

     from

      long-available knowledge

     in the

    fiel

    The  correlations analyzed by factor analysis

     reflect

      what  is

    traditionally

      and  honorifically  called  communal  or  common

    v ri nce But

     communal variance

     may

     also

     be

     called, less con-

    ventionally

      and more pejoratively,

     redundant variance

    To the

    extent

     a

     variable

     correlates

     with

     other variables, it has

     commu-

    nal

     or redundant variance: it is said to be

      explainable

    by these

    other variables

     and

     conveys

     no

     unique information.

     To the ex-

    tent

      a

     variable does

      not

     correlate

      with

      other variables

      but is

    itself highly reliable,

     that

     variable is indexing some quality not

    otherwise

     being captured.

    Redundancy

      has its

      usefulness.

      In

      factor analysis, redun-

    dancy provides

     multiple

     indicators and, therefore, more reliable

    indices of

     underlying factor dimensions. However,

     the

     commu-

    na l

     or redundant variance observed to exist within the particu-

    lar set of

     variables subjected

     to  factor

     analysis

     may or may not

    be

     important

     in

     other

     domains or other

     settings. Therefore,

      the

    factors

      summarizing this communal

      or

      redundant variance

    may

     or may not be important when they are brought out into

    the

     larger world.

    As

     a

     contrived

     but

     instructive example,

     consider a correlation

    matrix

     based on 100 variables, 99 of which intercorrelate be-

    cause they are all fully reliable manifestations of the latent vari-

    able, shoelace-tying competency,

     in one

     form

     or

     another. Vari-

    able 100 is a fully

     reliable, fully

     valid but anonymous or unrec-

    ognized measure of the latent variable, general intelligence,

    which

     is presumed here to not correlate at all with the first 99

    measures. Factor analysis of this matrix will issue one general

    factor explaining

     all of the

      communal variance.

     An

     unsophisti-

    cated enthusiast of factor analysis would be impressed by the

    finding

     of

     so

     powerful

      a factor explaining so much of the co-

    variance. Further, the lonely, nameless, intelligence measure

    would

     likely be viewed as a

      residual

    and dismissively con-

    signed

     to obscurity. Yet, relating these factor analytic findings

    to the wider, external world beyond the mathematical solipsism

    of the particular

     factor analysis would reveal  that

     the power-

    ful

    general factor

      has no or

      trivial implications, whereas

    the ignored residual variable has momentous behavioral

    significance.

    Of course, this example is a contrived extreme. But the point

    made applies

     to the

     real world

     of

     data

      to an

      unrecognized

      or

    unacknowledged extent:

     the

     amount

      of

     variance

      explained

    internally by a factor need not

     testify

     to the external psycholog-

    ical importance

     of the

     factor.

    The mix of variables submitted to factor analysis can be var-

    ied

     with

     the

     consequence

     that the factors then obtained will, in

    effect, be entailed rather than represent

     substantive

     findings. A

    small factor can be

     made large,

     a

     large factor

     can be

     made small,

    residuals can be made into factors (or discarded as

    irrelevant). One can

     make

     for a

      simple structure

    by

     dropping

    factorially

     complex

     ( interstitial )

     variables or prevent

      simple

    structure

    (and even create a circumplex configuration) by

    adding factorially complex variables that

      blend different

    sources

     of variance.

     If

     the

     locations

     of

     axes

     in one

      simple struc-

    ture are not  satisfying, it is possible by judicious selection and

    deletion

     of

     variables

     to

     relocate

     the

     axes

     and

     have quite another

      simple structure. One can

     achieve such ends

     by just

     varying

    the degree and the

     location

     of redundancy within the variable

    set. In particular, a set of variables can be

      prestructured.

    That

    is, wittingly or unwittingly, the set of variables subjected to fac-

    tor analysis may have been previously selected so as to contain

    several quite different  subsets or

      clusters

    of redundant vari-

    ables. Such

      prestructuring

    preordains the  factors subse-

    quently

      found by the algorithmic operations of factor

    analysis.

    It  is my  view

      that

      the

      frequent

     presence  and the  powerful

    effects  of  prestructuring are

      often

      no t  sufficiently recognized

    by

     those using the method of exploratory factor analysis. In par-

    ticular, it will be suggested later that, in crucial ways, influential

    demonstrations

     of the sufficiency  of the FFA may

     have been

      un-

    duly influenced

      by prior prestructuring of the personality vari-

    ables used in these analyses.

    1

     If so, then the

      recurrence

    and

      robustness

    over diverse samples of factor structures may be

    attributable more to the sameness of the variable

     sets

     used than

    to the intrinsic structure of the personality-descriptive domain.

    Although

     the method of

     factor

     analysis

     has been

     used

     for al-

    1

     Protagonists of the FFA recently have been acknowledging the cru-

    cial

     influence

     on factor solutions of the particular sets of measures used.

    For

     example, Costa and

      McCrae

      observe that the axes chosen by a

    varimax rotation  will  depend completely on the selection  of  variables

    (1992a, p. 661) . Goldberg (1993) also acknowledges this problem.

  • 8/18/2019 Five-factor Personality Description

    4/29

    190

    JACK  BLOCK

    most a century, there is still not a clear, unequivocal basis for

    deciding on the number of factors to extract in a factor anal-

    ysis or how to obtain an optimum rotation of the particular

    set of factors

     settled

      upon. Various rules exist for these conse-

    quential

      decisions, based on various arguable assumptions or

    aesthetic preferences. One can debate th proper number of fac-

    tors; is a ma trix under-factored or  over-factored? Wha t are

    the dangers of un derfactorin g as compared to overfactoring?

    How

     should

     one

     rotate

     to

     achieve psychological meani ng

    and

    who

     is to be the judge? Should a

     rotational

     method be used that

    destroys a general factor (varimax), em phasizes a general fac-

    tor,

     or

     seeks

     to

     conform

     to a

      priori, perhaps theoretical, expec-

    tations? Should one impose math ematic al orthogonality on the

    rotated factors, or is this cognitively economical way of repre-

    senting factors psychologically inappropriate?

     Is an

     oblique

     ro-

    tation, which inevitably

     fits the

     data po ints better, truly

     a

     fairer

    representation of the u nde rlyin g reality, or is it a better fit only

    because of the fitting of error? Should one inject unancho red,

    spurious variance into th e analysis by c onsidering al l variables

    to be fully

     reliable,

     or should the

     analysis

     employ more

     realistic

    communality

     estimates that set more constraints on subsequent

    rotations and produce less striking factor structures? When ro-

    tating factors, should one build up a weak factor or residualize

    it? Is a given  factor really only a bloated specific (Cat tell,

    1973)

     or

      tautological

    (Eysenck & Eysenck,

     1969)

     facto r war-

    ranting residualization? Or is the factor a

      primary

    factor or a

    merging of prim ary factors into a complex or combinato-

    rial

    factor w ith broader but c onflated behavioral significance?

    Since

     complex, prim ary, and tautological factors can all emerge

    simultaneously in a given factor analysis (Comrey, 1978; Guil-

    ford,

      1975;  Lykken,  1 9 7 1 >

      how does  one

      recognize

      and re-

    spond to their

     different

      nature s and implications? These issues

    are of foundational importance because, with the method of

    factor

      analysis, the psychological  nature of the factors ob-

    tained very

     frequently

     changes

     fundamentally

     as the

     number

     of

    factors changes and as the rotational criteria are varied. Intu-

    itively

      and

     conceptually, these transmogrifications should

      not

    occur: A

      real

    factor should not change in meaning when an-

    other factor

      is

     introduced

     or

     when rotations

     are

     shifted.

     But in

    fact,

     this happens

     in the

     arbitrary hyperspace

     of

     factor analysis

    and is unse ttling as one tries to i mp ute psychological meaning

    to the factors robotically issued by the workings of the method.

    As

     Cliff

     (1983) remarks in a more general context,

      There

     a re

    typically

     an

     infinity

     of

     alternative sets

     of parameters [e.g.,

     factor

    loadings] which

     are

     equa lly consistent with

     th e

      data, many

     of

    which

      would lead

      to

      entirely

     different

      conclusions concerning

    the

     nature

      of the

     latent variables (pp. 122-123).

     The

      method

    of factor analysis cannot choose among the

     infinity

     of possibili-

    ties. The decision requires conceptual

     argument

     and empirical

    work; one

     must return

     to the

     task

     o f

     being

     a

     psychologist.

    To

     further encourage a wary perspective on the results issued

    by fac tor analysis, it is also

     useful

      to consider the starting basis

    or grist for the method. Because factor analysis processes corre-

    lation

      matrices, all the psychometric problems that afflict

      cor-

    relation coefficients have  effects  on the  subsequent results

    afforded  by the

      method. Thus, computed correlation

      coeffi-

    cients

     are seriously

     influenced

     by the degree of relation linearity

    that is present; by scaling

     considerations;

     by the validities of the

    measures used;

     by the

     reliabilities

     of the

     measures being corre-

    lated;

     by

     truncations , restrictions,

      or

     extensions

     of

     range

     on the

    variables used; by the form s of the score distributions for the

    variables

      being correlated; by maverick observations that may

    improperly raise or lower correlations; by method variance; by

    merging

     different

      kinds of samples m anifesting different  covar-

    iance

      patterns (e.g., males and

      females)

      into one large, hodge-

    podge sample wherein

      the

      resultant correlations

      are an

      irrele-

    vant function

     of the

     relative sizes

     of the

     samples being inapp ro-

    priately

      combined; by heterogeneity with respect to a third

    variable (e.g.,

     measures x and y m ay be

     unrelated

     but

     because

    each correlates with age,  x and y

     will

     correlate and  appear  on

    the

     same factor unless age is

     partialed);

     by whether the variables

    being correlated are represented by differential  scores or by cu-

    mulative  scores (Loevinger, 1948); and much more. The

      downstream findings issued by

     factor

     analysis may be

     funda-

    mentally

      affected

      by  these

      often

      unevaluated upstream

    influences.

    For example, it appears th at the B ig Five factors are reason-

    ably orthogonal when data from a homogeneous sample of sub-

    jects are  evaluated,  but when data from a more  heterogeneous

    sample are factored, the consequent factor structure may ap-

    preciably  lose its orthogonality (see, e.g., Costa & McCrae,

    1992c; Goldberg, 1992,

     p. 37;

     Mroczek,

      1992). For

     another

     ex-

    ample, consider that  introspectiveness and impulsivity

    correlate quite negatively in a sample of Air Force jet  pilots bu t

    quite positively in a sample of male graduate students, with the

    consequence that a factor analysis

     will

      locate these two impor-

    tant variables differently, depending on the whether the data be-

    ing

     evaluated are

     from

      the first sample, the second sample, the

    samples combined, or the two samples in various proportions

    (Block,

      1955).

     Because of the man y kinds of unrecognized or

    unattended-to influences  on  correlation

      coefficients—notably,

    in my

      view,

      the  usage of  psychologically different  subject sam-

    ples— the findings of factor analysis, especially with regard to

    the factors emerging later in the factoring sequence,

     often

     do not

    replicate

     from

      one analysis to another or

     will

     appear in

     different

    form.

     I n

     p articular,

     on e

     mu st guard against easy

     and

     convenient

    dismissal of factors that may not replicate or have not yet been

    replicated. The

     factors being

     dismissed may

     well

     be

     reliable,

     im -

    portant within each of the analyses, and may well be crucial to

    recognize when the aspiration is to develop a comprehensive set

    of

      constructs for the description and understanding of

    personality.

    At a

     deeper,

     conceptual

     level, there

     is the question of

     whether

    the ubiquitously used correlation  coefficient  is, by its nature,

    able

     to

     represent pivotal features

     of

     personality functioning.

     For

    example, correlation coefficients

     fail

      to adequately represent

    asymmetricality

      of

      relations (e.g., although  wittiness

    and

      intelligence correlate,

     and wittiness necessarily entails  intel-

    ligence, intelligence

     does no t

      entail wittiness; although

      a  talk-

    ative

    person

      is

     necessarily

      gregarious, a

      gregarious person

    need

     not be talkative).

     There

      is

     also

     th e

     complication

      of

     condi-

    tionality o f relations (e.g.,  agreeableness with other individu-

    als may be

     m anifested

      if and

     only

     if the

     individuals involved

      are

    at the

      same

      or

     higher status

      or

     social level; spontaneity

     may be

    manifested

      only in safe conte xts). Correlation

      coefficients

    imperfectly

      reflect

      such relations. To the extent these relations

    exist,

     the

     psychological interpretation

     of

     subsequent factor anal-

    ysis

     findings

     will

     be

     misdirected.

     Of course, recognition of the

  • 8/18/2019 Five-factor Personality Description

    5/29

    A

      CONTRARIAN

     VIEW  OF THE

      FIVE-FACTOR APPROACH

    191

    insufficiencies   of

     correlation

      coefficients

     should

     n ot

     cause their

    abandonment. Rather,

     there needs

     to be

     more

     awareness of their

    inadequacy

     for various conceptual

     purposes.

    For reasons such as these, and others, a new look in factor and

    covariance

      structure analysis may be emerging. MacCallum

    and Tucker (1991) have presented a more realistic framework

    for

      the common-factor model,  on e  that recognizes  the  differ-

    ence between

     measured

     variables and

     modeled

     variables. Their

    approach, which awaits

     further

     development, respects the non-

    linear

      influences

     of common factors on measured variables or

    the presence of large numbers of minor common factors (p.

    503). Cudeck and

     Henly

     ( 1991) discuss various sources of dis-

    crepancy between models

     of

     covariance  structures

     and the

     sam-

    ple matrices

     from

     which these models are derived. They present

    sobering remarks on the model selection problem and re-

    mind

     us that such models are only descriptive or summarizing

    devices. Their concluding paragraph is worth quoting here:

    Yet

     even with

     models that are intended to describe or summarize,

    the problem of making comparisons inevitably arises when two or

    more structures apply

     to the same data and one

     wishes

     to

     evaluate

    their

     relative performance.

     Often

      the

     best that

     can be

     done,

     which

    is

     actually

     a

      useful  accomplishment whose value should

      not be

    minimized, is to

     state

     clearly the criteria that are used in the com-

    parison, in conjunction with descriptions of the models, character-

    istics of the

     data,

     and the

     purpose

      for

     which

     the

     models were con-

    structed. Evaluations of this kind inevitably include elements o f

    prior

     experience

     and

     individual preference. These personal points

    of  view  should

     be

     articulated.

     Many

     believe such

      subjectivity in

    model development  is somehow unscientific and undesirable,  as if

    it

     could

     be

     avoided

     by

     pretending

     it

     does

     no t

     exist.

     (p .

     518)

    Finally,

      as an anthropological observation, an unremarked

    oddity in the field of

     factor

     analysis warrants noting: Regardless

    of

     the

     content domain being evaluated,

      the

      great majority

     of

    published

      factor

      analyses report relatively  few

      factors—three

    through perhaps six or seven factors. Rarely does the

     factor

     an-

    alyst

      settle

      on  more factors  in

      reporting

      substantive

      results.

    There

     can be

     valid reasons

     for

     such

     findings:

     There

     may

     indeed

    be   only

     a small number of latent dimensions underlying the

    starting

     correlation matrices. But the underlying motivation of

    many

      factor analysts should be  considered  as well. They wish

    to

     achieve latent dimensions explaining large amounts

     of the

    covariance present. Otherwise, the method could be said to lose

    its raison d'etre. Unfortunately, however, this ambition of the

    factor  analyst

     is

     opposed

      by

     logical constraints residing

      in the

    nature of the

     procedure.

      To

     have many factors would necessar-

    ily diminish

     the

      explanatory power

     of

     each

      and

     result

      in a

    cognitively complicated

      picture of the

     world.

      To

     enhance

      the

      importance of the factors in an

     analysis,

     it is

     therefore prag-

    matically preferable to restrict their number and to discourage

    the  existence of numerous, small factors. If one allows for the

    level  of communality

      usually

      existing  in  empirical matrices

    (attenuated by unreliability and by the specific  variance of the

    measures being analyzed), each of five factors would be con-

    strained to  explain on the average not more than

     about

     8% or

    10 of the total variance. This is on the verge of explaining

    too

     little

     variance to be

     satisfying

     to many factor analysts. The

    human preference for schematizing the world a t a certain  level

    of

     complexity (Miller, 1956)

     may be

     operative  here. Such cog-

    nitive preferences,

     subsequently

     firmed

     into commitments,

     may

    underlie

     the

     motivation

     of

     some factor analysts toward sparse

    solutions rather than solutions involving more numerous and,

    therefore, more

     complicating latent

      dimensions.

    The preceding disquisition is by way of saying that factor

    analysis is a highly useful technique for the study of personality,

    but it is not a method for all reasons. As Costa and McCrae

    (1992a) say,  Used intelligently,

     it can

     yield valuable insights

    (p. 654). But abject deference to the method and the results

    it

     issues is not warranted (see, especially, Lykken,  1971). It is

    unlikely indeed that

     the

     logical structure

      and

     sequence

     of

     oper-

    ations underlying the method of factor analysis and correla-

    tional data can reflect the way individuals evolve, articulate, and

    conditionally

     use

     descriptive terms

     to

      characterize themselves

    and others. To the extent that the method of factor analysis, per

    se,

     is said to provide

     sufficient

     or strong justification for the five-

    factor  approach to the description of individual  differences  in

    personality, we must remember to look for and evaluate the un-

    acknowledged assumptions

     or restrictions that in fact undergird

    the

     data

     used, the correlation

      coefficients

     subsequently gener-

    ated, the factor analytic logic, the factor analytic heuristics, and

    the interpretations of the consequent factor analytic findings.

    The

      faith

      of

     FFA

      adherents—their premise and their prom-

    ise—that

     the field of personality psychology can confidently rely

    on the factor analytic algorithm as an appropriate and

      sufficient

    basis

     for

     objectively deciding

     on the

     theoretical constructs

     to be

    used by personologists is, in my

      view,

     unwarranted, naive, and

    limiting.

      Certainly,

      the

      results  afforded

      by

      factor analysis

    thoughtfully

      and  fairly applied should

      often

      be

     influential.

      But

    also influential should be understandings gained

      from

      experi-

    mental investigations,

      from

      intimate and prolonged observa-

    tions of other people,

      from  neurophysiological

      recognitions,

    from

      psychiatric insights,

      from

      personal  introspections,

      from

    formal  cognitive efforts  to

     create

     a theoretical system that en-

    compasses a chosen domain of phenomena,  from the thoughts

    flicking

      through

     one's mind as one

     drifts

     toward sleep,

     from

     any

    and

     all possible

      contexts

     of discovery (Reichenbach, 1951).

    None of these sources

     will

     be fully  dependable in the seeking of

    incisive, generative, and coherently related concepts, but all can

    contribute  to a  context  of  theory  construction that then

    warrants

      the  context  of

      justification (i.e.,

      validation)

    (Reichenbach,

     1951)

     and, in a helical process, takes the science

    onward

     to a new  context  of discovery.

    A

     Revisionist History of the  Discovery of the Big Five

    Approach to Personality Description

    Previous historical accounts of the FFA have come only

     from

    proponents of the FFA. Therefore, it may be

     useful

     to provide a

    more skeptical rendering of the chronology. In focusing on the

    several foundational studies of the FFA, I

     will

     suggest

     that,

     if

     the

    footings  of the FFA are less than secure, the structure subse-

    quently

     erected may not be a house all

     will

     wish to enter.

    Role   ofAllport

    The FFA approach in America may be said to have begun

    with Allport and

     Odbert's

      (1936) onerous compilation of all

    the

      terms

      in the

      1925 unabridged,  400,000 word  edition

     of

    Webster s

      Ne w

     International ictionary

     they judged  as usable

     to distinguish the behavior of one human being

      from

      that of

  • 8/18/2019 Five-factor Personality Description

    6/29

    192

    JACK BLOCK

    another (p.  24). They came  up  with  17,953  single-word  de-

    scriptor terms. T o this enormous collection, Allport and Odbert

    applied their definition  of

      trait

    as generalized and personal-

    ized determining

      tendencies—consistent  and

      stable modes

      of

    an  individual's adjustment to his environment (p. 26) and

    came up with a primary

     list

     of only

     (

    sic 4,504 nonjudgmental

      trait-names,

    still

     a

      very large set. They

     suggested that

      their

    alphabetical  listing  of

      trait-names might prove

      to be a  useful

    resource

      for

     psychologists developing rating scales

     and the

     like.

    However, Allport warned that  common speech is a poor guide

    to psychological subtleties (Allport,

      1961,

     p. 356).

    RoleofCattell

    Next  on the

      scene

      was

      Raymond Cattell

      (1943a,

      1943b,

    1947). Cattell subscribed

      to

      what subsequently

      has

      become

    known as the lexical hypothesis:  All aspects of hum an person-

    ality which

     are or

     have been

      of

      importance, interest,

      or

     utility

    have  already become recorded in the substance of language

    (1943b,  p. 483). He began with the Allport and Odbert trait-

    name listing but , noteworthy and

     often

      no t noted

     (but

     see John,

    Angleitner, &

     Ostendorf,

      1988), he

     deemed

      it insufficient.

    To

     make

     the list o f

     traits

     as

     complete

      as

     possible.

      . . in

     addition

      to

    all

      that could

      be

      obtained

      from  the dictionary,  the substance of

    all  syndromes  and  types which psychologists have  observed  and

    described

     in the

     past

     century or so [were added]. (Cattell,  1943a,

    p.

     491)

    Thus, he made sure that terms reflecting aspects of personality

    he

      deemed to be important—introversion/extraversion, emo-

    tional

      maturity,

      his

      construct

      of cyclothymia/schizothymia

    (the  essence of the agreeableness  factor,  according to French,

    1953,  p. 222), ascendance/submission,  Thurstone's  radical-

    ism/conservatism  variable,  McDougall's

      temper

    variables,

    and

     many

     more—were

     included

     in his

     starting list.

    Cattell's considered additions to the dictionary listing pro-

    vided

      by Allport and  Odbert  may

     well

      have been heuristically

    beneficial.

     B ut a consequence of his decision to introdu ce terms

    to represent a ccumulated psychological insights was to depart

    significantly from  the

     dictionary

      offerings  of

     Allport

      and Od-

    bert; he was no

      longer entitled

     to

     claim that

      his

     selected

     set of

    variables was a

     truly representative

     list.  . .

     derived

     from

      lan-

    guage (1945, p. 70).

    Cattell

     then proceeded through an a mbitious, retrospectively

    unspecifiable

     sequence  of semantic

     decisions

     to

     abbreviate this

    list. Applying  his  personal judgment  at  various stages  in the

    elimination  sequence, he first

     proffered

      17 1 terms

      (4%)

     of the

    Allport and Odbert adjectives as

     sufficiently

      representin g the

    personality  sphere. But because

      171

      traits were still too

    many at the time for his computational capabilities, the terms

    were further  clustered via correlational analysis aided by Cat-

    tell's sema ntic understanding of the terms, producing about 60

    clusters (194 3a, 1945). These were still too many, and so a fur-

    ther reduction was enforced to achieve a computationally man-

    ageable

      set of about 35 bipolar rating scale dimensions

      based

    on clusters

      of

     variables

      and

     termed

      by

     Cattell

      as the

      standard

    reduced personality sphere  (1945,  1947). Cattell went on to

    factor

      ana lyze some peer ratings based on these 35 variables

    and concluded that

      12

      primary factors underlay personality.

    These factors do not match

      well

     w ith the factors subsequently

    known

     as the Big Five. For a

      fuller

      description of the sequence

    by

      which Cattell evolved

      his

      lexical reductions, consult John,

    Angleitner,

     and Ostendorf (1988).

    More tha n 99% of the 4,504 trait-name adjectives of Allport

    and Odbert and those first added by Cattell had been eliminated

    along

     Cattell's

     way.

     This final set of 35 distilled variables

     proved

    to be u nusu ally consequential because of its acceptance and fre-

    quent

      use, in one

      form

      or another, by later researchers. Al-

    though Cattell's 35 variables certainly represented his own best

    judgment

      at the

      time

     o f

     what

      the

      most important variables

     of

    personality were, he also acknowledged that his reductions

    could

      well

      have eliminated personality features

      of

     relevance

    (Cattell, 1945, p. 7 1 ) .

    It

     is no denial of Cattell's brilliance, psychological acumen ,

    and  many  scientific  contributions

      to

      recognize, along with

    Joh n, Angleitner, and Ostendorf, that a n ultimate set of vari-

    ables for a scientific theory of personality should not be so de-

    pendent on any one person functi oning in a private, conve-

    nience-emphasizing,

     and

     subse quently unevaluated way.

     In

     par-

    ticular,

      one must wonder to what extent Cattell's prior

    theoretical notions regarding personality, however valid,

      influ-

    enced

     his

     sequential n arrowin g

     of

     possibilities

     and his

     construc-

    tion

     of variables by clustering terms. D id he introduce semantic

    structures th at subsequently would underlie and even entail the

    results

     afforded

      by subsequent empirical usage of his variable

    set? I believe so and illustrate this possibility in my discussion

    of

     the

     work

     of

     Tupes

     and

     Christal (196 1/199 2).

    Role of  Tupes  and  Christal

    Tupes

      and

      Christal, personnel selection psychologists

      em -

    ployed by the Air Force to im prove

     officer

      selection and promo-

    tion procedures,  reported

     a set of

     analyses subsequently hailed

    by FF A

     advocates

     as

      the discovery

    of the Big

     Five dimensions

    and as the  pivotal work tha t .  . . laid the foundations

     for.

      . .

    the five-factor model

    (McCrae, 1992,

     p.

      217). Their internal

    technical report has been

     often

     c ited over the years, but, because

    it was not

     published

     in an

     archival journal until recently,

     it was

    difficult

      to find.

      Because their  study

      has

      been

      so

      often

      and

    widely  hailed,

      it is

      useful

      to

      consider

      the

      Tupes

      and

      Christal

    contribution in closer detail than it usually has received.

    In  their  very  brief account (a bit more than eight pages of

    text in the

      original),

      Tupes and Christal described eight factor

    analyses of Cattell's rating variables. A ll the subjects rated were

    in their early 20s.

      Six of the

     analyses involved peer ratings,

      two

    were based on ratings of subjects provided by older, status-supe-

    rior raters. There were

     five

     male samples

      and one

     female sam-

    ple. Four of the analyses derived

     from

     Air Force

     data,

     two anal-

    yses

     were of data earlier analyzed by Cattell

      (1947, 1948),

     and

    two  analyses  employed  data previously  factored

      by

      Fiske

    (1949).

    The analyses of the Air Force samples initiated and were cen-

    tral

     to the

     Tupes

     and

     Christal

     report.

     Therefore, details

     of data

    gathering in the Air Force context assume special  importance.

    For example, the individuals in one sample knew each other as

    little

      as 3

     days;

     in

      three other

      Air

     Force

     samples, the

     length

      of

    acquaintance was 2, 3, 8, or 26 weeks.

    Th e

     raters

      were untrained, psychologically naive subjects,  and  little

  • 8/18/2019 Five-factor Personality Description

    7/29

    A  CONTRARIAN VIEW OF THE FIVE-FACTOR APPROACH

    193

    time could b e made available fo r training them a s raters.  Therefore

    the

      rating procedure must

      be simple,

      requiring

      a

      minimum

      of

    trained judgment. Only

     about

     two hours [in one  sample, only one

    hour] were available

     for

     the rating. . . [of] 2 5 to

     30

     subjects on 30

    or 35 traits.  (Tupes, 1957, p. 4)

    An issue must

     b e

     raised here that  will

      be

     more  fully  consid-

    ered

     later

     in

     this article.

     Can the

     psychological perceptiveness

     of

    Air  Force  officers  and

      officer  candidates,

     as quickly expressed

    by  3-point ratings

      on 30 or so

     scales

      in an

      officially

      required

    research

     p rogram regarding 12

      to 30 of

     their peers know n

     fo r

    such

      short periods, provide

     a

      fundam ental data basis

     for

     dis-

    cerning

     the  essential dimensions for the scientifically

      sufficient

    description

     of

     personality?

    In

     six of the analyses, 8 orthogonal factors were extracted; 5

    factors

      were  extracted  in  another analysis  and

      12

      factors were

    extracted  in the  analysis of the

      female

      sample. Summarizing

    their

     findings, Tupes and Christal concluded that  in each of the

    eight analyses  based  upon the Cattell  variables,  five fairly

    strong and

      recurren t factors emerged. They were impressed,

    and   later other psychologists were impressed,  by the  marked

    congruence among the five-factor patterns derived  in so many

    analyses.

      suggest tha t, fo r technical reasons, the degree of recurrence

    of

     this

     five-factor

     structure over their eight analyses

     may not be

    so striking as has been assumed. Although the analysis of their

    first Air Force samp le used the centroid method of

     factor

     anal-

    ysis prevalent

     at the

     time, followed

      by

     subjective rotation,

      the

    remaining  seven

      factor

      analyses

      all

      employed

      the

      multiple-

    group

     method

     of

     factor  analysis.

     The

     multiple-group method

     of

    factor

      analysis (see, e.g., Harm an, 1967, chap.

      11),

     now long-

    abandoned  and  mostly forgotten,  was frequently  used  in the

    precomputer era to lessen the laborious calculation centroid

    factor

      analysis

     o therwise required.

     The

     multiple-group method

    permitted the  extraction of a  number of factors  in one labor-

    saving

     operation.  The

     centroid method,

     on the  other hand, ex-

    tracted factors

     serially  and

      involved protracted computations

    of successive residual matrices. To lessen the  subsequent com-

    putations,  the  multiple-group factor analyst first had to  parti-

    tion the set of

     variables into

      a

      number

      of

     groups (i.e., antici-

    pated factors) according  to preconceived hypotheses. All these

    preconceived  factors then were extracted simultaneously, fol-

    lowed by the calculation of a residual matr ix. This residual ma-

    trix,

     if its elements were sizable, might then be subjected to  fur-

    ther analysis, usually by the centroid method.

    In

      the first of the Tupes and Christal  factor analyses, eight

    factors  were arduously extracted

      via the

      centroid method.

    These factors were then subjectively rotated toward orthogonal

    simple structure so as to residualize three o f them (i.e., via ro-

    tation,  the  loadings of variables on three factors were deliber-

    ately m ade generally low enough so that these factors could be

    said to be

     unimportant, thus justifying  their elimination

      from

    subsequent  consideration).  However,

      in the

     subsequent  seven

    factor  analyses, wherein

     the

     multiple-group method

      was

     used,

    the variables were grouped into five subsets p restructured  so as

    to correspond to the five rotated factors decided upon in the first

    study.

     That

     is, the

     orthogonal factor structures

     of the

     last seven

    analyses were created  to  conform

     insofar

      as  possible with the

    factor  structure solution settled upon  in the first  analysis. In

    effect, Tupes and Christal used their first factor solution as the

    target matrix  for all  their subsequent analyses. Factors repre-

    senting other than these five pregrouped sets of variables were

    residualized, except in the female sample in which the fifth fac-

    tor was split into two

     subfactors.

     A s Horn  (1967) has compel-

    lingly  observed, Procrustean  rotations to fit a target matrix can

    show  seemingly impressive congruence with  the  target even

    when random variables are involved. Inevitably, then, the pre-

    structured  solutions used in the  last seven Tupes and  Christal

    multiple-group factor analyses fitted the target matrix at least in

    part because of fitting

     error.

    The extent to which capitaliza-

    tion  on  chance w as  involved h as never been evaluated,  but it

    certainly  underlies a portion of the factor  recurrence

    observed.

    A  more im portan t concern to register, however, is that the

    semantic structure  underlying the Cattellian variables used  in

    all

      the

      analyses

      by

     Tupes

     and

      Christal

      may

     have intrinsically

    predestined  the facto r solutions subsequen tly observed.  For ex-

    ample,

     their first factor was defined by variables

      labeled

     as  se-

    cretive, silent, self-contained

      or

      reclusive

    as

     opposed

      to

     sociable, and

      talkative. Referring

     to

      Soule s D ictionary of

    English Synonyms

      (Sheffield,

      1959),  under  the  term

      secre-

    tive, there is the

     following

     entry (cited completely):  reserved,

    reticent, close, uncomm unicative, cautious, wary, taciturn (p.

    470).

      As  these synonyms attest,  the  variables  identifying  the

    first

     factor

     of

     Tupes

     and

     Christal were highly

     redundant. There-

    fore, if the subjects providing the basic data were not responding

    incoherently,

      there  is no problem  in  understanding  how such

    redund ant rating scales will, in a facto r analytic context, gener-

    ate a  factor  dimension.  A  second factor was defined  by  such

    variables  labeled

      as

      composed, calm, placid,

    and

     poised. Again referring  to  Soule,  under the heading com-

    posed is the  following:  calm, quiet,

      unruffled,

      undisturbed,

    unmoved,

     tra nq uil , placid, sedate, collected, self-possessed, im -

    perturbable,  cool

    (p .

      109).  Although

      no t

      included under

     composed,

    the

      term

      poised

    when referenced identifies

      composure as a synonym. It is obvious how such virtually

    synonymous

     rating scales will generate

     a

     mathematical factor.

    A  third factor

     was

     based

      on

      such ratings

      as

      good-natured,

    cooperative, and

      mild;

    a fou rth factor stemmed

      from

      rat-

    ings

     of  artistic, imaginative, and   intellectual; a fifth fac-

    tor emerged from ratings of the eq uivalent scales  responsible,

    scrupulous,

    and

      seeing

      a job

      through

     in

     spite

      of  difficulties

    or temptations.

    It is not

      surprising when semantically related

    variable sets prove to load on the same factor; as these terms are

    used

      by

     often  inarticulate

     or

     language-insensitive  raters, their

    redundancies are great. Consequently, their factorial equivalen-

    cies

     may

     only

     testify  to the

     reliability

     and

     coherence

     of the

      rat-

    ings

     made of the subjects. Rather than representing trul y sub-

    stantive findings, the  Tupes  and  Christal

      factor

      findings may

    simply  reflect  the prestructuring that somehow crept into and

    characterizes  the Cattell-specified  variables. Therefore,  al-

    though

     five

     factors

      may

     well characterize

      the  data sets

     Tupes

    and Christal created or used, this finding may not be of great

    moment.

    2

    2

     The preceding discussion does not reintroduce  the Shweder

     (1975)

    argument that personality judgments  are no more  than  statements

    about

     how

     respondents classify things

     as

     alike

     in

     meaning (p. 482).

     As

    Block, Weiss, and Thorne (1979) observed,  in no way whatsoever  is it

  • 8/18/2019 Five-factor Personality Description

    8/29

    194

    JACK BLOCK

    Factor interpretation

     and

      factor naming

     is

     always

     a  difficult

    problem and usually cannot be done quickly or

     satisfyingly  (but

    see Meehl, Lyk ken,

     Schofield,

     & Tellegen,

      1971).

     As personnel

    selection psychologists w orkin g in the

      unfamiliar

      field of per-

    sonality,

      Tupes and Christal did not elaborate conceptual un-

    derstandings of their five factors. Instead, guided by earlier work

    of French (1953), they sim ply entitled their five factors, without

    any elaboration

      of

      their

      meaning

      whatsoever,

      as

      Surgency

      or

    Extroversion, Agreeableness, Dependability, Emotiona l Stabil-

    ity or the

     Opposite

      of Emotionality, an d

     Culture.

    These terse

     Tupes

     and

     Christal

     labels have been extrao rdinar-

    ily

     durab le over

     the

     years

     and

     have shaped

     thinking and

      research

    interpretations.

      It is of

     historical

     and

      evidential interest

      to ex-

    amine more closely the original basis  in  French's work for one

    of these perduring and portentous factor labels.

    French, a psychometrician at the Educational Testing Ser-

    vice, had  reviewed 6 8  factor analyses  of  personality variables

    and   issued a laconic, cau tious report presented as a reference

    source on

     factorial

     studies

     rather than as a theoretical

      exposi-

    tion  . .

      .

    (p.

      8). Abstracting

      from

      these

      68

      studies, French

    offered

      a

      total

      of 49

      factors, presented alphabetically from

      Agreeableness

    to  Will

     Control,

    as

     having been observed

      in

    at least two studies. The French review was a careful evaluation

    of a

     body

     of

     literature

     he

     recognized

     as deficient  in

     ma ny ways.

    Often,

      the

     factor analyses available

     to

      French were based

      on

    data conceptually or methodologically unacceptable even by

    the  prevailing standards

     of the

     time. Consequently,

      the

      resulting

    factors were of ambiguous meaning or implication and French's

    effort  to  identify  or

      summarize reproducible factors

      was se-

    verely limited, as he was

     fully

      aware.

    The

     very

     first

     factor presented

      by

     French

      as

     perhaps replica-

    ble,  Agreeableness (on e of the Big Five ), was predicated on

    10

     factor analyses,

     as

     follows: Lovell (1945) factor analyzed

      13

    Guilford

     qu estionnaire variables. Thurstone

     (1951)

     reanalyzed

    Lovell's

     matrix. Baehr

     (1952)

     factored

     a

     variation

     on the

     Guil-

    ford

      variables. Bolanovich (1946)  factored ratings b y supervi-

    sors of field engineers on such variab les as personality,

    sales

    ability, and   desire for self-improvement. Brogden's  (1944)

    paper, A Multiple-Factor Analysis of the Character Trait

      In -

    tercorrelations Published

      by

      Sister Mary  McDonough

    was

    based on religiosity ratings of early adolescents  attending paro-

    chial

      school during

     the

      1920s (McDonough,

      1929).

      Tschech-

    telin  (1944)

      factored ratings

      of

      children

      by

      their classmates

    with

      regard to such variables as

      good sport,

    punctuality,

    and  sense of humor. Rey burn and Taylor (1939) factored rat-

    ing

     data

     previously published by Webb

     (1915).

     And three fac-

    tor  analyses by Cattell

     (1945,

      1947,  1948) were scrutinized as

    well.

    I suggest that, although these several studies were respectable

    products  of their

      times

      and although French was diligent in

    seeking common denominators  across analyses, an everlasting,

    overarching, theoretically useful psychological

      concept cannot

    be  extracted

      from

      this empirical melange. Instead, we have a

    broad, bland, impressively unincisive umbrella of a label—

    possible to proceed from specific semantic similarity judgments.  .  .to

    a  specification  of just  which individuals are rated high or low on a  par-

    ticular rating

     dimension

    (p.  1057).

    Agreeableness—under which

     any

      number

     of

     importantly dis-

    tinguishable personality qualities

     may

     escape analysis.

     I

      further

    suggest, on similar grounds, that the other

      four  far-flung,

      allu-

    sive

      labels taken over  by  Tupes  and  Christal should  not  have

    entered the personality firmament on the basis of this early fac-

    tor an alytic review.

    Role

     of

      orman

    The

      unpublished

      Air

      Force technical report

      by

      Tupes

      and

    Christal might well have languished, unattended and without

    consequence, had not Norman (1963) picked up the baton.

    Norman

      had

     taxonomic  concerns.

      He

     noted that the con-

    struction of more

      effective

      theories of the development, struc-

    ture,

     and

     functioning

     of

     personality will

     be

     facilitated

      by

     having

    available

     an

     extensive

     and

     well-organized vocabulary

     by

     means

    of

      which

      to

      denote

      the

      phenotypic

     attributes  of persons

    (p .

    574). And he accepted the

     Allport

      and Odbert

     argument

      that

      perceptible  differences  between persons in their characteristic

    appearance or mann er of behaving or changes over time and

    situations

     of

     single individu als

     in

     these regards become codified

    as a

     subset

     of the

     descriptive predicates

     of the

     natura l language

    in

     the course of its development

    (p . 574).

     Norm an's discussion

    of the issues involved is sober and sophisticated:  It is explicitly

    not  assumed that complete theories

      of

     personality  will simply

    emerge automatically from such taxonomic efforts..  . . There

    is  a  good deal more  to  theory construction  . . .  than  the

    development

      of an

      observation language—even

      a

      good one

    (p.

      574).

    However,

     his

     empirical

     offering

     consisted only

     of a

     replication

    of

      the study of Tupes and Christal. On the basis of the factors

    identified

      by Tupes and Christal, Norman selected 20 of their

    Cattellian variables

     (the four

     variab les best represen ting each of

    the five factor s). B y his variable selection procedure, he limited

    himself to the variables most likely to demonstrate simple struc-

    ture.  He then had several groups of undergraduates  offer  peer

    ratings using this

     restricted set of

     variables.

     By

     this  narrowed

    selection

     from

      among wha t has already been suggested m ay be a

    prestructured set of variables, it

     follows

     that the factor structure

    subsequently observed may have been a foregone conclusion.

    Norman's findings, however, were subsequently viewed as fur-

    ther empirical support for the existence, primacy, and perhaps

    sufficiency  of the five

      orthogonal personality factors reported

    by

     Tupes and Christal (No rma n renamed the Dependability

    factor  as Conscientiousness, a replacement label that subse-

    quently

     achieved p reference).

    It

     should be noted that in collecting peer ratings by laypersons

    on a

     restricted

     set of 20 variables and in emphasizing the

     basic

    importance of the five personality factors subsequently ex-

    tracted

      from

      this kind

      of

      data, Norman would seem

      to

      have

    shifted from

     h is

     announced focus

     on the development of a

      sci-

    entifically

     oriented language for the description of personalities

    to a study of the way laypersons use a  constrained language to

    characterize other laypeople. There

      are connections—impor-

    tant,

     useful, even crucial

     connections—between

     these two em-

    phases,

      but

      they

      are not the

      same. Psychologists certainly

     can

    learn a great deal for their science by studying the nature of lay

    observations. But it does not

      follow  that

      lay usages should be

  • 8/18/2019 Five-factor Personality Description

    9/29

    A  CONTRARIAN VIEW  OF THE  FIVE-FACTOR APPROACH

    195

    taken over to provide the basic concepts of the field of personal-

    ity psychology.

    In  contemplating

      his

     1963 study,

     and

      prompted

      by the

     sug-

    gestion of Tupes and Christal that the C attellian variables might

    have

      omitted some

     of the

      personality concepts residing

      in the

    Allport-Odbert adjectives, Norman decided that

      it was

     time

     to

    return

     to the

     total

     pool

     of

     trait names

     in the

     natural language—

    there  to  search  for  additional personality indicators  not  easily

    subsumed under one or another of these five  recurrent factors

    (1963, p. 582).

    To locate personality terms

     that

     were new o r previously omit-

    ted, Norman  (1967)  searched  a  later unabridged edition  of

    Webster and found  175

     single-word

     descriptors to add to the list

    compiled by Allport an d  Odbert. Then, applying his own set

    of decision rules for sorting through this descriptor lexicon, he

    excluded

     terms judged

     to be

      purely evaluative

     and

     mere quan-

    tifiers, ambiguous, vague,  and . . .  metaphorical, very

    difficult,  obscure,

     and

      little-known,

    and

      anatomical, physical

    and grooming characteristics.

    The

     remaining terms

     he

     further

    sorted into

     the

     categories

     of  stable

     traits, temporary

     states,

    and   social roles.

    Excluding

     the latter two categories as inap-

    propriate for the  purpose  of personality description, Norman

    was

     left with 2,800 single-word descriptors deemed to represent

     stable traits. These remain ing terms were then administered

    to undergrad uates to empirically assess their understandability

    by  laypersons, thei r social desirability, and the degree to which

    undergraduates believed these terms were descriptive

      of

     them-

    selves

     an d

     their peers. After  Norm an removed terms

     he

     judged

    to be ambiguous or

     unfamiliar

     to the typical college student or

    redundant,  1,431  terms remained. With this last culling, Nor-

    man  had a pool of descriptors he believed approached  suitabil-

    ity fo r

      the development

     of a

     structured taxonomy.

    Norman then proceeded

      to a

     further,  semantic sorting

     of his

     stable trait terms that, although never formally  reported  an d

    published by him (but

     described

      by Briggs, 1992; Goldberg,

    1981,  1990; John  et  al.,  1988; John, 1990),  may have

      influ-

    enced later research findings. Impressed by the factor  structure

    that

     he,

     Tupes,

     and

     Christal

     had

     identified  earlier,

     and

     using

     his

    understanding of the psychological m eaning of the factors, Nor-

    man personally sorted h is 1,431 term s according to their judged

    fit into his five dimensions an d assigned the terms to the positive

    or

     negative pole

     of

     each dimension.

     As a final

     step,

     he

     examined

    the

     terms

     at

      each pole

     a nd

     further formulated what

      he

     judged

    to be

      semantic clusters within  each pole.

      In

      all,

     he

     sorted

      his

    1,431 terms into 75 sem antic clusters specifying one or the other

    end   of the five-factor  dimensions  he  viewed  as  paramount.

    Fewer

      than 25 terms  were

      left

      unclassified by Norman. The

    John,  Angleitner,  and

      Ostendorf

      (1988)

      historical review

     may

    be

      consulted

      for

      closer information regarding this Norman

    effort.

      Although Norman did not proceed

     further

     with his terms

    and trait clusters, his delineation and structuring of the trait

    lexicon

      provided

     an

      important starting point

      for

      subsequent

    efforts

      to

     advance

     a

     lexically based tra it taxonomy.

    By  the mid-1960s, the initial phase of the FFA may be  said

    to have ended. A lthough a number of

     subsequent

     articles (e.g.,

    Borgatta,  1964; Digman

      &  Takemoto-Chock,

      1981; Smith,

    1967) also reported sim ilar five-factor solutions of rating data,

    these studies all were based on v arious versions of the Cattellian

    or Norman variable

     sets

     and so these later studies can be classi-

    fied as

      further

      manifestations  of  this early stage  of the  FFA.

    Given that prestructured sets of variables may have been used,

    it  follows that

     the fiveness of the

      factors emerging

      in

     these

    later studies  and  their appreciable

      (but

      no t

      full)

      similarity  of

    factor interpretation need not be viewed as persuasive.

    The curren t FFA frequently memorializes and cites as foun-

    dational the empiricism of

     this

     first

     phase

     of the

     FFA.

     There-

    fore, to the extent the concerns expressed previously regarding

    this

     first

     phase

      are

      cogent,

      a

     muting

     of

      these observances

      may

    be in order. However, it is on the basis of contemporary accom-

    plishments rather than  its  early history  that  the FFA  must

    achieve  its claimed deservedness. And so, the later work that

    has

      zoomed the FFA into its present prominence must be

    considered.

    The second, and

     current,

     phase of the FFA began with  the

    work

      of

      Goldberg (e.g., 1977, 1981, 1982, 1990,  1992),

      who

    sought to go beyond the constraints set by

     Cattell's

     choice of

    variables. Working

     with

      larger numbers  of adjectival

     descrip-

    tors, he has

     presented

     a refinding and refining of the five factors

    reported by earlier investigators. The consequent claims for this

    lexical approach merit alternative analysis. Further, the team of

    Costa

     and McCrae

      (e.g., 1985,

     1992c;

     McCrae

     &

     Costa, 1985,

    1987) has

     brought

      the

     FFA, which

     had

     been focused exclusively

    on  adjective-based ratings of self or  others, into  the  question-

    naire

     realm. Th eir adopting

     and

     adapting

     of the Big

     Five frame-

    work also warran ts close scrutiny.

    Five-Factor Lexical Approach

    Goldberg  (1971) had long been concerned w ith the question,

      Why measure

     th t

     trait? With close knowledge of Norman's

    (1963)

      work

      on an  adequate taxonomy o f  personality attri-

    butes (p. 57 4) and heuristically using the lexical hypothesis,

    during

     the

      1970s Goldberg began

     and has

     sustained

      a

      meticu-

    lous lexical and taxonomic

      effort.

      Among his many

     studies,

     he

    evaluated the role of the evaluation component  in adjective us-

    age

     (Peabody

      &

     Goldberg, 198 9), their

      frequency  in

      various

    categories (Ha mps on, Joh n, & Goldberg, 1986), the consis-

    tency w ith w hich adjectives

     a re

     used

     by

     laypersons (Goldberg

     &

    Kilkowski,

     1985), the

     level

     of

     abstractness

      of

     adjectives (J ohn ,

    Hampson,

      &

     Goldberg,

      1991),  the

      influence

      of

      unipolar

      and

    bipolar contexts for the way adjectives are employed (Goldberg,

    1992),

     and—not

     least—the factor structure underlying lay us-

    age

     of

     adjective-descriptors.

    For

     ma ny years, Goldberg's taxon omic work

     was

     communi-

    cated only

     informally

      to  members of an  invis ible college of

    assessment psychologists

     at

     conferences

     or via

     technical reports

    (e.g., Goldberg,

     1977,1980, 1983).

     Increasingly,

     over

     the years,

    his

      views

     an d arguments became persuasive.  Two pub lications

    summarizing  his  lexical work appeared relatively early  and

    proved  especially

      influential

      (Goldberg, 1981, wherein

      he

    coined the phrase the Big-Five ; 1982). More recently, he has

    presented a major,  integrat ive report intended  to alleviate any

    qualms about the generality of the Big-Five structure of per-

    sonality descriptors (Goldberg,  1990, p. 1223). Along the way,

    he  has

     offered

      a variety of sets  of adjectives  as  marker vari-

    ables so that indiv iduals can be assessed in terms of the lexical

    Big Five factors (Goldberg, 1980, 1983, 1992).

    Goldberg's program  of thinking and research has been pred-

  • 8/18/2019 Five-factor Personality Description

    10/29

    196

    JACK  BLOCK

    icated on the assum ption, questioned earlier, that the method of

    factor  analysis is  suitable an d

      sufficient

      for  achieving a  scien-

    tifically

     c ompelling personality taxonomy. In addition, the lexi-

    cal  approach involves premises, procedures, and findings that

    often

      adm it alte