+ All Categories
Home > Documents > Benjamin 74

Benjamin 74

Date post: 14-Apr-2018
Category:
Upload: luis-goncalves
View: 217 times
Download: 0 times
Share this document with a friend

of 34

Transcript
  • 7/30/2019 Benjamin 74

    1/34

    P s y c h o l o g i c a l R e v i e w1974, Vol. 81, No. S, 392-425S T R U C T U R A L A N A L Y S I S O F SOCIAL B E H A V I O R 1

    L O R N A SMITH B E N J A M I N *Department o f Psychiatry, University of Wisconsinand Wisconsin Psychiatric Institute

    A brief review of the literature on structural analysis of interpersonal be -havior is followed by a proposal which draws heavily f rom prior models,especially those of Schaefer and of Leary. The proposed model goes be -yond previous ones in that it has a highly explicit structure which definesbehavioral opposites, com plem ents, and antidotes. Built on two axes nam edaffiliation and interdependence, th e model describes dyadic social inter-actions in te rms of complementary proport ions of those under ly ing d im en-sions. Opposite behaviors appear at 180 angles whereas complementarybehaviors appear at topologically s imi la r positions on two separate planes.Antidotes are defined as opposites of complements . Using the questionnairemethod, the proposed structure has been tested by the responses of normalas well as psyc hiatric subjects. An alysis of these data by the techniques ofautocorrelation, circumplex analysis, and factor analysis supports the model.

    The assumption that behavior is orderlyand l awful is the basis of scientific psychol-ogy. If the assumption is val id , then itshould be possible to develop a model forpredicting which particular behaviors willtend to be associated with each other. Anal-ysis of the basic structure of social behavioris one possible approach which might be ex-pected to yield such predictions. The needfo r a structural model of social behavior hasbeen emphasized by Foa and Turner (1970):. . . there has been some reluctance to recognizethat specification of psychological components isl ikely to be as complex in construc tion and asrevolutionary in consequence as the notion of struc-ture has been in nuc lea r physics and in genetics[p. 246].

    Efforts to describe the structure of socialbehavior can be viewed in terms of twomajor categories: the multidimensional ap-proaches which include as many dimensionsas are needed to meet a given mathematicalcriterion, and the approaches which conf inethe number of dimensions to two or threeso that a model can be constructed in realspace. The multidimensional approach is1Special thanks fo r encouragement and helpwith th e development of this paper go to MarjorieH. Klein and James Greenley; the editors of Psy-chological Review also made helpful contributionsto the final version.2 Requests for reprints should be sent to LornaS. Benjamin, Department of Psychiatry, Univers i tyHospitals, Madison, Wisconsin 53706.

    exemplified by Cattell's 16 Personality Fac-tor analysis of personality, and the real-spacemodeling approach is exemplified by Leary's(1957) interpersonal circle.

    In defense of the multidimensonal ap-proach, Cattell notes:The busy psychometrist may sometimes feel thatsixteen sub-scores is a lot, but such is the realcomplexity of human nature, and if, as studiesshow, the major i ty of these personality character-istics are involved in most criterion predictions, am u c h better m ultiple correlation is to be obtainedby respecting th e complexity than by indulging ina fools paradise of over-simplification [Goldberg &Hase, 1967, p. 3].In response to the charge of oversimplifica-tion, the modeling approach can name theadvantages of parsimony and the manipula-tive possibilities fol lowing from having a pic-ture of the model in real space. This paperwill be concerned with the second alterna-tive, namely models which are simpleenough to be pictured in two or three di-mensions.

    The development of parsimonious struc-tural models has been pursued, sometimesindependently, by theorists from psychiatryand sociology as well as from psychology.At times there has been remarkable overlapin conceptualization suggesting independentconvergence on a common underlying struc-ture. Such overlap is illustrated by Chance(1966, p. 133) and Biermann (1969, p. 339)

    392

  • 7/30/2019 Benjamin 74

    2/34

    S T R U C T U R A L ANALYSIS O F SOCIAL BEHAVIOR 393w ho separately proposed identical modelsfor describing social behavior in general andth e process of psychotherapy in particular.Each of these theorists reduced Leary's in-terperson al circle to the four quadrants basedon axes representing the two dimensions :positive-negative and active-passive. Leary's(1957) interpersonal circle itself was firstdescribed in a paper by F re e dm a n , Leary,Ossorio, and Coffey (1951) and was bui l ton four nodal poin ts : Domina te and sub-m it were located opposite each other on thevertical axis, whereas love and hate wereoppositional nodal points on the horizontalaxis. Each category in the circle was de-fined in terms of these nodal points, so that,fo r example, boast was placed on the hateside of dominate , while teach appeared onthe love side of domina te .An interpersonal circle applying specifi-cally to parental behaviors has been proposedby Schaefer (1965). In his earliest m o de l ,th e vertical dimension w as defined by thepoints control-autonom y, whereas th e hori-zontal dimension w as defined by the pointslove-hate. This m o d e l w as supported byfactor analysis of ratings of interviews withmothers in the home, children's reports ofparent behavior, and teacher ratings of class-room behavior. A later version included athird dimension: lax versus firm control,and data in support of this model have beenobtained (Ranson, Schaefer, & Levy, 1968;Schaefer, 1971) in several cultures includ ingJapan, Canada, Czechoslovakia, Germany,Iran, and India. Schaefer com pared hismodel with several others and noted manypoints of convergence. A similar conclusionabout extensive overlap among models ofparent behavior resulted from a literaturesearch by Goldin (1969).Interpersonal models such as those pro-posed by Leary (1957) and Schaefer (1965)can also be related to classical psychiatrictheory. For example, Carson (1969) ef-fectively used four categories derived fromLeary (hostile-dominant; friendly-domi-nant ; host i le-submissive; friendly-submis-sive) and related his view of interpersonalprocess to the psychiatric theories of HarryStack Su llivan . Chance (1966, p. 132) re -

    lated her version of Leary's interpersonalcircle to Freud, Adler , Horney, Jung, andF r o m m .Rinn (1965) attempted to expand Leary'sinterpersonal circle so that it would applyto the intrapsychic domains of cognition andfeeling as well as to social interaction. Hesuggested that models in these domainscould be constructed such that specific cogni-t ive attitudes and specific feelings wouldparallel specific interpersonal behaviors. Forexample, the behavior "affectionate" wouldbe accompanied by the att i tude "sociable"and th e feeling "pleasant." Rinn 's mode ldoes not comprehensively cover cognitionand feeling, but the idea that cognition andfeeling might have a structure which paral-lels interpersonal behaviors deserves fur therdevelopment. The need for such efforts hasbeen detailed by Bergin and Strupp (1970) :T h ere is a r enewed appreciat ion that internal , in -t rapsychic or experient ial processes, whether theybe of a feel ing or of a cognitive nature, have con-siderable power to influence bodily processes, be-havior and the general state of the organi sm. . . .Massive denials of the problem since the t ime ofJ. B. Watson have not obviated its impor tance[p. 25].

    In addi t ion to developing an accuratemod e l for the s t ruc ture of interpersonal be-havior and its associated cognitive and feel-ing states, there is a need for a methodologywhich allows th e application of this modelto a single individual, to a person as a whole.Carlson (1971) noted that most studies ofpersonality involve between-subject ratherthan within-subject comparisons and con-cluded :Personality psychology would seem to be paying anexorbitant price in potential knowledge for thesecurity afforded by preserving the f o r m s of con-venience and m ethodological orthodoxy. M ustthese important , unanswered quest ions be left tothe literature and psychiatry? [p . 207-209].

    The present paper presents an extensiveelaboration of the models of interpersonalbehavior developed by Schaefer and byLeary. The proposed model has explicitlogical and mathematical properties and issupported by within-subject as well as be-tween-subject analyses of questionnaire data.There is no attempt here to develop exten-

  • 7/30/2019 Benjamin 74

    3/34

    394 L O R N A SMITH B E N J A M I N

    G o a w a y n o w 1 2 8E x c l u d e 1 2 7Isolate 126N e g l e c t , ignore 1 2 5Bluff, illogical 124A ba n d o n , reject 1 2 3Depr iv e 1 2 2S t arv e, po is o n 1 2 1Murderous attack 13 0Injure 131Frighten 1 32Restrain, o v e r p r o t e c t 1 3 3Shout, criticize 1 3 4Threaten 1 35Exploit 1 3 6Authoritarian 137Shame, guilt control 138Dominate 140

    F l e e , w i t h d r a w 2 2 8C o mpet e, try one-up 2 2 7N o Input, no response 2 2 6Stimulate self 2 2 5D e f y , s u s p e c t 2 2 4Distrust, grieve 2 2 3Temper tantrum 2 2 2D i s g u s t , refuse, spit 221Suffer, dlsaffillate 230D o not touch me 231Rigid, hide, gr imac e 2 3 2Cling, annoy 233C r i n g e , d e f e n d , whi ne 2 3 4Pr esent , yield, a p p e a s e 2 3 5R e se nt fu l c o mply 2 36Obey routines 2 37W a x y comply 238B e mounted, s u b m i t 240

    Need new me 328E x c l u d e self, one-down 32 7Isolate self 3 2 6Neglect, Ignore self 3 2 5Pu t self on 324R e c k l e s s 32 3Self deprivation 3 2 2S t a r v e , poison self 3 2 1Self hate, s u i c i d e 3 3 0Injure self 331Frighten self 3 3 2A pat h y 333Criticize self 3 3 4N o p r o d u c t s , nihilistic 335S e l f sacrifice 3 3 6Person = c a u s e 33 7A c c u s e self, guilt 3 3 8I am my ownm a s t e r 340

    120 Eman c i pa t e11 8 You can do it11 7 Encourage divergence11 6 Listen, e q u a l 1 tarian115 E x p l o r e , le t d i s c o v e r1 1 4 C o n fi rm , p r a i s e113 P l a y , a l l o w p e e r p l a y11 2 S m i l e , greet w a r m l y111 K i s s , heal, groom1 10 E m b r a c e , ten der , touch1 41 S uppo rt , cradle, nurse1 42 Indulge, T . L . C .143 P r o t e c t , keep company144 Reasoned p e r s u a s i o n1 45 S t i m u l a t e , t e a c h1 46 O v e r i n d u l g e147 Intrude14 8 P o s s e s s iv e

    220 B e e m a n c i p a t e d2 1 8 U n a s s a u l t i v e a s s e r t2 1 7 Individualistic2 1 6 C o o p e r a t i v e2 15 R e v e a l , d i s c o v e r214 D i s p l a y , court213 P l a y , peer p lay2 1 2 A p p r o a c h , s m i l e , understand21 1 A c c ept , gro o m, h ea l2 1 0 Hug, affiliate2 41 A c c e p t , n u r s e , s l e e p2 42 F lo w er child2 4 3 T r u s t244 Comply willingly2 45 A b s o r b , imitate246 Incompetent2 47 O v erc o n fo rm, d e f e r2 48 Satellite

    32 0 E m a n c i p a t e self31 8 Self-confident, reliant31 7 S e l f a c t u a l i z e31 6 Fair, just31 5 E x p l o r e self31 4 C o n f i rm self31 3 E n t e r t a i n self31 2 W e l c o m e self31 1 H e a l , groom self31 0 S e l f love34 1 S e l f s uppo rt34 2 S e l f c a r e343 P r o t e c t self344 Dignified, r e s p e c t self345 S e l f taught, ac c o mpl i s h ed34 6 O v e r i n d u l g e self, a d d i c t34 7 S e l f preo c c upied348 S e l f po s s e s s e d

    F I G U R E 1. The char t of social behavior. (The first surface describes parentlike behav iors ;th e second, chi ld l ike; and the t h i rd , in t ro jected a t t i t udes f rom significant others. From "ABiological Model fo r Unders t and ing th e Behavior of Ind iv idua l s " by L. S. B en j am i n . In JackW e s t m a n (Ed.), Individual differences in children. N ew Y o r k : Wiley, 1973, p. 220. Copy-r igh t 1973 by Wiley. Reprinted by permission.)

  • 7/30/2019 Benjamin 74

    4/34

    S T R U C T U R A L A N A L Y S I S O F S O C I A L B E H A V I O R 395sions of the model to the domains of cogni-tion or feeling.

    A M O D E L OF S O C I A L B E H A V I O RT he proposed model appears in a d ia m ondshape on three planes or surfaces and is pre-sented in Figure 1. The three surfaces arerespectively named parentl ike ( t o p ) , child-l ike (m idd l e ) , a nd int ro je c t (bo t tom ) . Onth e parentlike surface, behaviors which areprototypically charac teristic of pare nts areentered, and, in general, these are active innature and concerned with what is going

    to be d o n e to or for the other person.On the second surface, th e childlike plane,behaviors which are prototypically charac-teristic of children are listed; these are typi-cally reactive and concerned with what isgoing to be d o n e to or for the s e l f . Thusth e active-passive dimension of concern toBi e rman n (1969), Chance (1966), and Os-good (1957) is implicitly included by thepresentation of the model in two planes: thefirst plane representing the active, concern-with-the-other domain, and the second planerepresenting the reactive, concern-with-the-self domain .The horizontal axis in Figure 1 is na m e daffiliation and compares to the horizontalaxis in the models of Schaefer (1965) andLeary (1957) w ho were in agreement thatone dim ension should be defined by the polesof hate an d love.Leary named domina te and submi t as op-posites on the vertical axis whereas Schaeferindicated that in this dimension autonomy isth e opposite of control or dom ina te . Eachdefinition seems reasonable, but the resul t ingclassifications are quite different. The m ode lpresented in Figure 1 resolves this di lemmaby defining subm i t as the complement ofdo m inate while calling em ancipate (allowau to n o my) th e opposi te of domina te or con-trol . Thus th e ver t ica l d imension in theparentl ike plane of Figure 1 ranges fromdominate to emancipate , while the verticaldimens ion in the complementa ry chi ld l ikeplane ranges f rom submit to be emancipated.This vertical dimension is n a m e d interde-pendence.

    In general, opposite behaviors appear inFigure 1 at 180 angles on each plane, andcomplementary behaviors appear at topo-logically similar positions in the parentlikeand childlike planes of Figure 1. For ex-ample , submi t is the complement of dom i -nate, and be emancipated is the complementof em ancipate. Each successive point on thefirst (pa ren t l i k e ) surface of Figure 1 ismatched by one in the same topological loca-tion on the second (childlike) surface, th etw o being com plem entary. Beginning withthe point at 12 o'clock and moving clock-wise, complementa ry pairs respectivelyshown on the parentlike and childlike sur-faces of Figure 1 are emancipate - be eman-cipated; you can d o it-unassaultive assert;encourage divergence-individualism; equa-litarian-cooperate; explore, let discoverreveal, discover. Moving clockwise to adifferent quadrant s tart ing at the point domi-nate (270) th e complementary pairs aredominat e - submit; shame, guilt control -waxy comply; authoritarian-obey routines;exploit - resentful comply; threaten - pre-sent, yield, appease; shout, criticise-cringe,defend, whine, and so on around th e m ode lfor a total of 36 complementary pairs .The idea of specifying complements hasalso been put forward by Parsons (in Bald-w i n , 1967), Feffer (1970), Mueller (1969),Mueller and Billing (1968), Schaefer(1971), Foa (1966), and Carson (1969).Although they m ention the im portance of theconcept of complementa ry and reciprocity,these theorists have not yet developed th eidea in depth.The most explicit development of theidea of complementari ty has been offered byCarson (1969) who states:When a person "offers" behavior fal l ing wi thinany of the quad rants of the interpersonal circle,he is, in effect, "inviting" th e other person toadopt a com plementary s tance in respect to both ofthe principle d imensions wi thin the circle [p . 147].In Carson's terms, this means, for example,that the hostile-submissive person invitesrelations with a hosti le-dominant person;and the friendly-submissive person invitesresponse from a friendly-dominant person.Turning to developmental data, Carson fur-

  • 7/30/2019 Benjamin 74

    5/34

    396 L O R N A SMITH B E N J A M I NDefy, s u s p e c t NonconformityEmotional instability (B )Hostility (S)

    B e emancipated

    D i s a f f il i ate

    Rigid, hjde

    B e individualist ic

    ExtroversionStable< S , B )

    H u g , a f f i l i a t e

    Sub m tLove (S )Emotional stability (B )Conform ity Comp ly w i l l i ng l y

    F I G U R E 2. Maccoby an d Masters' (1970) summary of two-factor child behavior models com-pared to the childlike surface of Figure 1. (The inner ring is Schaefer's modelaxes indicatedby S; th e m i d d l e ring is f rom Becker and Krug's modelaxes indicated by B; and the outerring is f rom Baumrind and Black. Ad ded points outside th e three rings are from Figure 1.Adapted f rom "Attachment and Dependency" by E. Maccoby and J . C. Masters. In P. Mussen(Ed.), Carmichael 's manual o f child psychology. New York: Wiley, 1970, p. 119. Copyright1970 by Wiley. Reprin ted by permission.)

    triers th e point about complementar i ty : "Al-most uni formly , studies . . . have come upwith th e same conclusion: hateful behaviorin the parents begets hateful behavior in thechild, and loving in the parents begets lovingbehavior in the child, [p . 151]."The naming of the surfaces began with thefact that the first was based on Schaefer's(1965) factor analyses of parent behavior.Since th e second surface w as intended tocomplement the first, it seemed logical toassign it the name chi ldl ike . The na m ingof the second surface as childlike has vali-dation in its resemblance to empirical lybased m odels of chi ld behavior. For exam -p l e , Maccoby and Masters' (1970) compari -

    son of three different models of child be -havior are reproduced in Figure 2, and rep-resentative points from the childlike surfaceof Figure 1 have been added. Figure 2shows that the childlike surface of Figure 1is quite consistent with existent models ofchildren's behavior.

    Behaviors charted on the first surface asparentl ike are not necessarily more responsi-ble or blameworthy than those classified onthe second surface as childlike. The theoryof Figure 1 suggests that persons occupyingtopologically similar positions on the firsttwo surfaces are in complementary relation,but , as Carson suggests (1969), one m e m -ber of a complementary pair is not m o r e"responsible" for the combination than th eother. Bell (1968, 1971) ha s recentlyurged the rejection of the popular frame-of-reference which assigns primary responsi-bility to the parentl ike member of a dyad.T he usefulness of regarding the parent andth e child as equally responsible has recentlybeen validated by stochastic analysis ofmother-child interactions (Harper, 1971).Nor do the complementary re la tions be-tween th e parentlike and the childl ike sur-faces describe permanent, fixed charactertraits or roles. A n individual (parent,child, therapist, patient, boss, or employee)

  • 7/30/2019 Benjamin 74

    6/34

    S T R U C T U R A L ANALYSIS OF SOCIAL BEHAVIOR 397could , for exam ple, be characteristically andrigidly submissive but need not be. Hemight be dominant in some si tuations andsubmissive in others. Figure 1 is intendedto describe the structure of dyadic interac-tions at any given moment , but the proto-typic names parentl ike and childl ike are notintended to imply rigidity of role. How -ever, it is possible to use Figure 1 to under-stand any consistent trends in behaviorwhich d o characterize a given adult . Forexample, an individual characteristically re -la t ing to those in authori ty with markeddeference m a y b e continuing a role begunwith parents during early chi ldhood. Or aperson who is characterologically dominatingm ay be identifying with a chronically con-trolling parent. Variations on this theme ofth e re la tion between adult behavior and ex-perience with significant others during child-hood are explored e lsewhere (Benjam in,1973, p. 237). The idea of re la ting adultbehavior to childhood experience with par-ents is, of course, a basic principle of psycho-analysis, and it has recently been convinc-ingly extended to include siblings as an in-fluential early cause of consistencies in adultbehavior ( T o m a n , 1971).A ll points in Figure 1 are represented inte rms of complementa ry proport ions of thebasic d im e ns ions : affiliation and interde-pendence. The points can be described bythe equation |X| + |Y| = 1, where X re fe rsto points on the abscissa, and Y to those onth e ordinate. For example, on the parent-like surface, th e point stimulate, teach has+4/9 affiliation and 5 / 9 independence.The absolute values of these coordinatessu m to 1. If the square of X and Y hadbeen used, then Figure 1 would appear asa circle rather than as a d i a m o n d . Use ofabsolute values of X and Y instead of thesquares is conceptual ly more parsimoniousand al lows the poles of the axes to be m o r esalient than they would be if the surfaceswere circles. The poles of the axes repre-sent primitive , "basic" behaviors whichcould , loosely speaking, be named sexuality,pow e r , m urde r , and separate territory.Points located progressively fur ther andfur ther from these poles are less pr imit ive;

    those midway between th e poles are themost "civilized." For example, on the topsurface of Figure 1, at abou t 45 , the pointsconfirm, praise and explore, let discover ap-pear; at about 315, the points stimulate,teach and reasoned persuasion appear. Theseare quite "balanced" and genteel by compari-son w ith sheer power, m urd er, sexual ity , andautonomy located at the poles.Debating the relative merits of using dia-m o n d or circular shapes for the planes inFigure 1 is less important than testing th ebasic logic of arranging behaviors on aclosed continuum which allows definitionsof opposites and com plements. The mostcritical tests of the basic logic of Figure 1are the within-subject autocorrelations andth e between-subject factor analyses appear-in g later in this paper.Whereas th e parentl ike and the childlikesurfaces describe interpersonal behaviors,th e third or introject surface describes intra-personal behaviors or attitudes, that is, be-haviors directed toward th e self rather thantowa rd others. Points on the introject sur-face were named by deducing what wouldhappen if parentl ike b ehaviors charted on thefirst surface were directed toward the self.For example, s tart ing at 270 on the firstand third surfaces, dominating behaviorturne d inw ard resul ts in the point b e m y o w nmaster. Shame, guilt control turned on theself results in accuse self, guilt; shout, criti-cise turned inward resul ts in criticize self.This logic for naming third-surface pointsapplies for every topologically similar set ofpoints on the first and third surfaces, re-spectively.T he idea that attitudes toward th e selfrepresent introjection of the way one hasbeen treated by significant others has prece-dent in both th e clinical and research litera-ture (e.g., Herbert, Gelfand, & Hartman,1969). Sull ivan (1953, p. 16) was a m a j o rclinical exponent of the point of view thatfrom early infancy, a child's self-concept re-flects the way others thought of him andtreated him. The psychoanalytic idea of in-trojection has also been formulated in socio-logical terms (Cottrell, 1971): "The selfemerges and is perceived by the individual

  • 7/30/2019 Benjamin 74

    7/34

    398 L O R N A SMITH B E N J A M I Nonly through th e responses of referenceothers whose role he takes toward his ownacts [p. 552]." Cottrel l credi ts this form ula-tion to George Herbert Mead who first de-scribed "taking th e role of the other." Fo a(1961, 1966) and Parsons (in Baldwin ,1967) are two other theorists w ho also en-dorse th e idea that self-concept reflects ex-perience with significant others. Recent sur-veys (Coopersmith, 1967) have conf irmedth e relation between parent-child interac-tions and children's self-concepts and havedemonstrated a relation between self-conceptand behavior (Felker & Th o mas , 1971).

    The points on all three surfaces of Figure1 are assigned a code number which reflectsth e s t ructure of the chart. A ll points onthe first surface begin with a 100s digit;those on the second surface, wi th a 200sd ig i t ; and those on the th i rd sur face , w i th a300s digit. The 10s digi t of the respectivecode numbers is assigned according to theconventions of g e o m e t r y : 1, 2, 3, and 4 forthe respective Cartes ian quadrants . The I sdigits range from 0 (for points on the polesof th e axes) to 9, descr ib ing the 9 succes-sive subd ivisions of each qua d rant. Themain advantage of the coding system is thatit facilitates finding points on the m o d e l .For example, th e point neglect, ignore isidentified by the co de n umber 125 as beingon the parentl ike surface, second quadrant ,fifth subdivision. A discussion of s o me im -plications of the code numbering system andi llustrat ions w ith everyd ay exam ples appea rse l sewhere (Benjamin , 1973).Antido tes are specified in t e r m s of thecomplement of the opposite. In o th e r w o rds ,th e antidote for a given behavior is foundat the poin t complementary to its opposi te.For example, th e ant ido te to defy, suspect(224) is found by first noting the oppositepoint, comply willingly (244) , and then find-ing i ts complement, reasoned persuasion(144). Figure 1 specifies that if a personis in the interpersonal posture described byth e point defy, suspect (2 2 4 ) , he is perceiv-ing the other member of the dyad as behav-in g according to point b l u f f , i llogic (1 24) .Thus, th e m e m b e r who is being defied and

    suspected must change from 5/9 affiliationto +5/9 affiliation (i.e., be more fr iendly) ;and f rom 4/9 emancipation to 4/9 do mi n a-t ion (i.e., assume more interpersonal power).If he can do this, then his reasoned persua-sion (144) should elicit comply willingly(244). In practice, the switch from disaffili-ation to genuine affiliation and from hostileemancipation (negl igence) to mo dera tepower , is not always easily d o n e ; the figurespecifies wh at is needed as an antidote todefy, suspect (224) but does not indicatehow to do it . The first two surface s ofFigure 1 describe 36 such sets of behaviorsand antidotes.The m o d e l in Figure 1 could be elaboratedupon in term s of its relation to other l i tera-ture (clinical , infrah um an prim ates) as wel las in t e r m s of clinical within-subject appli-cat ions. Available examples of the lat ter in -clude (a ) meas u remen t s of patient change,( Z > ) m e a s u r e m e n t s of changes in the in ter -personal style of therapists in t ra in ing , (r)descriptions of developmental changes inparent-child interactions, and (d ) co mpar i -sons of m e m o r y of chi ldhood relat ions withparents to adult relations with significantothers. These possibil i ties will be presen tedelsewhere, and the present paper will be de-voted to the available formal statistical testsof the val idi ty of the m o d e l .

    ANALYSIS OF THE M O D E LData have been obtained through a seriesof que stionnaires labeled A, B, and C. TheSeries A questionnaire al lowed rat ing of in-d iv idua l s in t e r m s of the po i n t s on the firsttw o surfaces of Figure 1. For example , theitem describing the point dominate (140)

    read:M y is th e boss of our relat ionship, always"o n top," in control of, in charge of how we usethe available time, space and supplies. He/sheinsists I comply wi th h im /her qu ick ly and quiet ly"just because he/she said so."T he rater w as asked to read into the blankth e n a m e of significant others such as spouse,parent, boss, or therapis t . For the m o s t

  • 7/30/2019 Benjamin 74

    8/34

    S T R U C T U R A L ANALYSIS O F S O C I A L B E H A V I O R 399par t , i tems were worded 3 in concrete be-havioral term s and requ ired no sophisticatedreasoning fo r interpretation. Years of clini-cal experience are not required to under-stand th e relation between th e above itemand the concept "dominate." Such directrating of behaviors has recently been shownto be m o r e effective in predicting behavior(Goldfried & Kent, 1972) than the proce-d u r e of predicting behavior from indirectmeasures of hypothetical personality con-structs.Each item was rated on a scale ranging0-100 w ith 10-point interv als m ark ed andanchor points labeled NOT AT ALL (descrip-t ive of the person being rated) at 0; M O D -E R A T E L Y at 50; an d P E R F E C T L Y at 100. Useof this simple single-stimulus procedure fo rm e a s u r e m e n t rather than the m o r e prestigi-ous forced-choice format (Cronbach, 1960)is supported by Scott's (1968) investigationof major personal i ty tests given under dif-ferent formats with comparable results. T herationale for forced-choice format has usu-ally been that it reduces the possibility ofdefensive distortion, but Scott's results chal-lenge this assumption.The ratings are interpreted as measur ingthe subject's view of his relations with sig-nificant others in both his present and hispast. It is assumed that these ratings ofperceptions (e.g., m em ory of how m othe rbehaved) re la te more importantly to therater than do the actual behaviors of thepeople being rated (e.g., how mother actu-ally behaved). Stated another way, the as-sumpt ion is that one is m ove d by how onesees the world more than by how the worldreally is .A n alternate series of questions, Series B,reworded th e Series A items so that th erater was rating himself in relation to theother person rather than simply rating theother. For example, the self-rating SeriesB item for the point dominat e (140) wasI am the boss of my relationship with myI am always "on top," in control of, in charge of

    3The wording of the i t ems was refined duringm a n y meetings with interested psychiatric resi-dents. Special thank s go to James Guerro , GlenSchurette, Nancy Caine, Russell Caine, and BruceHoltzman.

    ho w we use the available time, space and supplies.I insist he/she comply wi th m e quickly and quiet ly"just because I said so."Self-reports have been criticized on theground s that d efensiveness precludes accu-racy, and the inference usually is (e.g., M c-Clelland, 1972) that indirect methods suchas the Thematic Apperception Test or theRorschach are needed to get at the truth.However, at least one recent study (Scott& Johnson, 1972) has shown that self-reports can correlate better with peer ratingsthan do the ind irect, m ore "sophisticated"personali ty m easurem ents. There is, then,no pretense of outsmart ing th e subject bypenetrating defenses with subtle items andcom plicated expert inferences. The approachis simply consistent with the old medicalaxiom : " If you want to know what is thematter with the patient, ask him." It isunderstood that th e subject's answer is notth e whole story, but it appears to provide anexcellent starting point for inquiry.A third questionnaire, Series C, allowedraters to score themselves in general termson each of the 108 points in Figure 1. Thewording was so general that results of SeriesC ratings might be interpreted as measuringtraits, or habitual ways of relating. For ex-ample, the wording for the point dominat e(140) was "I control, am responsible forwh at happens to others. Both the good andth e bad which happens to others is relatedto something I have done, or have notdone." The wording for the point / am myown master (340), representing dominationtu rned inward, was "I am responsible for,in control of, what happens to me. I planahead, look back, set my bearings and setsail . I am the captain of m y ship, the m asterof m y fate."Statistical tests reported in the remainderof this paper are based on the followingsamples.4 (a) Maternal ra tings of their nor-

    41 would l ike to acknowledge the generous co-operation of Thomas V. Geppert and his colleaguesat the Dean Clinic of Madison, of Marc Hansenand his colleagues at the University of WisconsinFamily Health Service, of Mourad Arganian of theUniversity of Wisconsin Child Psychiatry Section,an d of innumerable psychiatric residents at theUniversity of Wisconsin Department of Psychiatry.

  • 7/30/2019 Benjamin 74

    9/34

    400 L O R N A SMITH B E N J A M I NOO ANA1BXCU

    n c0 '0 '~0 "'"0 , 0 ,c

    D C

    r ijnK

    )

    C

    > M 3

    01t cJf3 C

    ,10 J1 C[NATE

    30

    ,10i

    '"t70 J

    10

    (

    t

    )

    ,"i

    70' t ENCOt}0 "90 i

    J Jio7]PCS

    e$ 0c[N1> S E

    3UK "1

    e> o

  • 7/30/2019 Benjamin 74

    10/34

    S T R U C T U R A L ANALYSIS OF S O C I A L B E H A V I O R 401m al children (Series A) and of themselvesin relation to these children (Series B) wereobtained through pediatrics outpatient clin-ic s (w=171). ( b ) Maternal rat ings ofchi ldren brought to a child psychiatry clinic(Series A) and of themselves in relation tothese children (Series B) were obtained (n= 51). (c ) Undergraduate s tudents (mos t lyfemale ) in a class on family life rated them-selves on Series C (w = 200). (rf) Therewas a sample of persons will ing to rate anenti re battery of questionnaires includingratings of themselves in general (Series C),of a significant other person (Series A),and of themselves in relation to that signifi-cant other person (Series B); the i r memoryof their mother in early and middle ch i ld -hood (Series A) and of their relation toher at that time (Series B); the i r memoryof their father in early and middle chi ldhood(Series A) and of their relation to him atthat t ime (Series B). (There were 110 sub-jects, including 60 n o rma l and SO psychiatricsubjects.) (e} The battery of question-naires mentioned in d usual ly took 4-8hour s for com plet ion and has recently beenreplaced by a short- form series. In theshort-form series, each chart point is repre-sented by a brief phrase presented in ran-domly selected order. For example, thephrase describing chart point dominat e (140)is "controls, is in charg e of m e." In theshort-form series, ratings are m a d e on a5-point scale labeled 0, 25, 50, 75, and 100;only odd-numbered char t po in ts are sam-pled in one form (Short Form 1), whereaseven-numbered chart points are measured inthe other (Short Form 2). The short-formbattery can be completed in less than anh o ur , and early returns suggest results arecomparable to the long-form procedure.(The sub jec ts were 36 medical students, 6psychiatric patients, and 50 psychiatric res-idents and their supervisors rating eachother.)

    Data have been gathered over a period of

    four years, and the analyses presented arerepresentative but by no means exhaustive.Analyses presented include (a) autocorrela-tions among items corresponding to pointson Figure 1 conf irming its structure; ( & ) acircumplex table of correlations confirmingthe structure of Figure 1; (c) factor anal-ysis yielding th e proposed underlying dimen-sions and generat ing a reasonable facsimileof Figure 1; ( d ~ ) the principle of comple-ments being confirmed by correlations be-tween sur faces ; (e) reliability being high innormal samples and being used to charac-te r i ze sub jec ts ; and (/) rating in t e rms ofsocial desirabil i ty being shown to charac-ter ize norm al subjects and endorsements ofsocially undesirable i tems being more l ikelyin psychiatric subjects.Autocorrelat ions Among Items: Within-Subjects Analysis

    Autocorrelat ions among i tems correspond-in g to points o n Figure 1 confirm it s struc-ture. Individual subjects, their spouses,therapists, or other relevant persons can beshown th e computer analysis of the ratingsin the form shown in Figure 3. This is ananalysis of a single subject, and it al lows aformalized examinat ion of relat ions amongmemories of early childhood experience andadul t behavior . The f o r m a t of Figure 3, tobe discussed in detail below, allows individ-uals to compare their perceptions among re -lations with their parents, spouses, andchildren. There have been many dramat icmoments when subjects have viewed th ecomputer analysis of their ratings and con-vincingly made or accepted observationssuch as "I'm treating my son just l ike m yfather treated me"; "I choose boyfr iendswho are mean to me in the same way mymo th er was"; "I'm just exactly the oppositeof everyth ing m y father was"; "M y husbandtreats me just l ike my mother did , and Ireact to him j u s t as I did to her." The ex-position of the clinical implications of this

    F I G U R E 3. M ap s and autocorrelations. (Data ar e from a mother's ratings of her own parent-l ike behavior in relation to her 18-month-old son. The map at the top of the figure showschart points receiving above-median endorsement and presents the raw data fo r each respectivei tem. T he m i d d l e figure presents 18-point autocorrelat ions performed on these ratings, and thebottom figure, the 36-point autocorrelations.)

  • 7/30/2019 Benjamin 74

    11/34

    40 2 L O R N A SMITH B E N J A M I Nprocedure is beyond the scope of the presentpaper which is to be confined to fo rma l test-ing of the structure of Figure 1.Data for Figure 3 w e re from a mother 'sratings of her own parentlike behavior in re-lation to her 18-month-old son. The toppart of the figure presents a m ap of thispart of her responses to the Series B ques-tionnaire . In constructing a m a p , th e com-puter program f inds the subject 's medianrating of all i tems of the questionnaire (i.e.,not jus t the parentl ike i tems) and prints ou tthe nam e of the chart points receiving above-median endorsement a long with th e scoreactually assigned to the corresponding i tem.Thus the map presents the chart pointswhose items were judged to be relativelymore characteristic of the person being ratedand gives a phenomenological impression ofth e na ture of the relationship. The map a tthe top of Figure 3, for example, has mostabove-median endorsements falling on theaffiliative side of the chart, suggesting abasically fr iendly mother-son relationship.Such friendliness is typically obtained innormal populations and from subjects askedto rate th e questionnaires in t e r m s of theirideal of what a good relationship should be.The map in Figure 3 deviates slightly fromth e norm a l and the ideal in that there ismore than usual endorsement of fr iendlypower (overindulge = 70, in t rude = 60, pos-sessive = 70) and of unfr iendly al lowing ofautonomy (go away now = 30, exclude =30). It m igh t be noted in passing, however,that excessive endorsements of fr iendlypower (intimacy, symbiosis) a l ternatingwith the opposite tendency to exclusion isqui te common between mother and chi ldwh en the child is in the age range 18-24months; the data on this theme are c om -pletely consistent with the clinical observa-tions of Mahler (1968).T he bottom third of Figure 3 presents aset of 36-pair autocorrelations ( d f fo r eachr = 34) computed for the data appearing inthe map in the top third of Figure 3. Forexample, the r at Lag 1 was obtained bypairing the scores for adjacent points onFigure 1. In other words, th e score fo rthe item describing the point encourage d i-

    vergence (70) w as paired with th e score forlisten, equalitarian (80); th e score for listen,equalitarian was paired with the score forexplore, let discover (90). At Lag 2, rswere among points two steps apart. Forexample, the score for encourage divergence(70) was paired with the score for explore,le t discover (90); the score for listen, equali-tarian (80), with that for confirm, praise(80), and so on. At successive lags, pair-ings were among points hypothesized to befurther and fu r the r apart unt i l at Lag 17,near opposites were p a i r e d : encourage d i-vergence (70) w as paired with shame, gui l tcontrol (10); listen equalitarian (80) waspaired with authoritarian (10), and so on.T he bottom third of Figure 3 shows thatwh en adjacent points were paired, rs w e rehigh and positive; when orthogonal pointswere paired (Lag 9), rs were near ze ro;wh en opposi te points were paired, rs w e relarge and negative. The smooth, nearly con-tinuous transi t ion from lag to lag shownfo r the single subject in Figure 3 w as highlycharacteristic of individua ls from n o r m a lpopu lations. Ty pically, the auto corre lationsappeared in the inverted-normal curve formshown at the bottom of Figure 3. To docu-m e n t th e generality of that finding, i t wasconvenient to select a single number repre-senting the degree to which each subject'sautocorrelation approximated the inverted-norm al curve shape. Such a num ber wasfound in the product-moment correlation co-efficient (r) between each autocorrelationcurve and an inver ted normal Z c u r v e(Grant , 1962). T he r ight-hand s ide of theautocorrelat ion curve shown in the bottomof Figure 3 appears in m irror im age to theleft-hand side because of the statistical re-dundancies which wil l become apparent tothe reader who computes a few sample rsusing the raw data from the top of Figure 3.The 36 pairs involved in computing the rat Lag 1 were identical to those involved inth e r at Lag 35; those for Lag 17 were iden-tical to those for Lag 19, and so on. Be-cause th e 36-point autocorrelations appearedin m irror image, only half of the points inth e autocorre la tion c urve (Lags 1-17) andhalf of the points in the norm a l Z curve

  • 7/30/2019 Benjamin 74

    12/34

    S T R U C T U R A L ANALYSIS OF SOCIAL B E H A V I O R 403(Z = 3.70 to Z = 0) were used , yielding d f= 15 for each test of goodness of fit.T he r (15) between th e inverted-normalcurve and the autocorrelation curves can beregarded as a coefficient of internal con-sistency since it reflects th e degree to whichraters gave similar ratings to i tems samplingchart points hypothesized to be a d ja c e n t ;th e degree to which they gave opposite rat-ings to i tems sampling points hypothesizedto be opposites; and the degree to whichthey showed no relation among i tems hy-pothesized to sample orthogonal points.Typically, the r (15) measuring internalconsistency w as near .90 in the long-formquestionnaires. For example, for 15 dif-ferent analyses, th e range of average rs ina sample (sample 4) of 60 normal subjectsw as .83-.9S, and the average of the averagew as .92. Such internal consistency was ob-tained to a slightly lesser degree in the short-form questionnaires. A group of 36 m e di -cal students showed an average coefficientof internal consistency of .89, when averag-ing across ratings of self and of relationswith significant others (range .79-.93).Autocorrelations such as those shown atthe bottom of Figure 3 were invulnerable tochanges of the order of the i tems in thequestionnaire. Results suggest that at leastth e language of the questionnaires (if notalso the actual behaviors described by thatlanguage) conforms to the structure pro-posed in Figure 1.To avoid the mirror imaging involved inthe 36-point autocorrelations, an evenharsher test of the structure of Figure 1w as constructed using tw o series of 18-paircorrelations rather than one series of 36-paircorrelations; the results appear in the mid-d le of Figure 3. Here, there were no mirrorimages. The first 18 lags of the m i d d l e partof Figure 3 were obtained by taking the 18points on the disaffiliative side of the chart(i .e. , Chart Points 140, 138, 137, 136, . . .128) and computing lags by "spilling over"into th e affiliative side of the chart as needed.For example, Lag 1 paired data for Points140-138, 138-137, 137-136, . . . 128-120;Lag 2 paired Points 140-137, 138-136, . . .128-118, . . . ; Lag 17 paired Points 140-

    128, 138-120, 137-118, . . . 128-147. Thefinal 17 lags were obtained by taking the 18points on the affiliative side of the chart(120, 118, 117, ... 148) and "spillingover" into the disaffiliative side, as needed.For example, Lag 19 paired Points 120-138, 118-137, 117-136, . . . 148-120. Lag35 paired Points 120-118, 118-117, 117-116, ... 148-140. Thus , the 18-pair auto-correlations shown in the m i d d l e of Figure3 involved 35 rather than 17 different auto-correlations. Because they had less thanhalf the degrees of freedom (d j - 16 ra therthan c f f = 34), these 18-point autocorrela-tions were much less orderly than the 36-point autocorrelations. Lags 1-17 and 18-35 of the 18-point autocorrelations did notinvolve mirror images and therefore wereeach independently compared to an appro-pria te part of the normal curve. T he aver-age of the separate halves of these 18-pointautocorrelations was the final number usedto represent the degree to which successive18-point au tocorrelations approx im ated th einverted-normal curve shape.

    Developmental n o r m s fo r 18- and 36-pointautocorrelations are presented in Figure 4.Data are from sample (sample 1) maternalratings of ped iatric outpa tients. Inspectionof the figure reveals at least four features ofautocorrelation curves as described by theircorrelations with an inverted-normal Zc u r v e : (a ) The 36-point autocorrelationsconsistently appeared in close approximationto the inverted-norm al curve. From A geGroup 7-9 through Age Group 14-21, theaverage rs were near .90 for both parentlikeand childl ike behaviors. The practice ofsampling at points closer in real time duringthe first five years of life (shown on theabscissa of Figure 4) is based on traditionwithin developmental pediatrics; this, inturn , is based on the fact that most of thephysical development of the bra in and headoccurs within the first five years of life (seeTanner, 1970, p. 85). (& ) The 36-pointcorrelations for childlike behaviors corre-lated about .9 0 with an inver ted-normal Zcurve beginning with the first five monthsof life and continuing through th e entireage range studied. It appears that child-

  • 7/30/2019 Benjamin 74

    13/34

    404 L O R N A SMITH BENJAMIN

    O - - - Og BD a

    P A R E N T - L I K E 1C H I L D - L I K E )PARENT-LIKE )CHILD-LIKE J

    36 POINTS

    18 POINTS

    1 2 3A G E GROUP

    FIGURE 4. Comparison of developmental trends in 36-point and 18-pointautocorrelations. (Each individu al 's respective autoco rrelation curvesseeFigure 3were correlated wi th an inverted-normal-Z curve to measuregoodness of fit, d f = 15 . Sample was of 171 normal ch i l d ren . Group 1 = age0-5 m o n t h s ; Group 2 = 6-11 m o n t h s ; Group 3 = 12-17 m o n t h s ; Group 4 =18-23 m o n t h s ; G ro u p 5 = age 2; Group 6 = age 3; Group 7=age 4; G ro u p8 = age 5-6; Group 9=age 7-9; Group 10 =age 10-13; Group 11= age14-21.)

    like behavior conforms to the s t ructure ofFigure 1 from the earliest years, (c ) A u-tocorrelations of ratings of children's par-entlike behavior becam e progressively m oreorder ly w ith increases in age. For A geGroup 0-5 months , the r between 36-pointautocorrelations and the inver ted-normal Zcurve w as below .40; it showed nearly pro-gressive increases with each successive agegroup unti l parentl ike behavior reached anadult level in the age range 7-9. It ap-pears that parentl ike behavior requires de-velopmenta l t i me and/or experience to ap-proximate th e structure proposed in Figure1. (d ) The 36-point a utoco rrelation s al-w a y s appeared in closer approximation toth e inverted-normal curve than did the 18-

    point autocorrelat ions perform ed on thesame data . In Figure 4, the average rs be-tween the 36-point autocorrelation curvesfor childlike behavior and the inverted-nor-m al curve rarely dropped below .90 whereasthose for the 18-point autocorrelations didnot exceed .55. N ot shown in Figure 4were th e rs for mothers rat ing their ow nparent l ike behavior in relation to their chil-d r e n ; th e average of these 36-point rs forany of the age groups shown in Figure 4never d ropp ed below .90. Such high-aver-age rs were also obtained in the sample of110 ind ivid uals rat ing the en ti re battery .Because the 36-point autocorrelations typi-cally correlated so highly with the inverted-normal curve , fai lure to obtain r in the .90

  • 7/30/2019 Benjamin 74

    14/34

    S T R U C T U R A L ANALYSIS OF SOCIAL B E H A V I O R 40 5range is considered remarkable whereas itwould not be so noteworthy in the case of18-point rs . Clinical applications of failuresto obtain rs in the .90 range fo r 36-pointautocorrelations curves are discussed subse-quently in High Reliability in Normal Sam-ples, and subsequent references to autocor-relation curves will refer exclusively to the36-point type.Circumflex Table o f Correlations: Belween-Subjects Analysis

    A circumplex table o f correlat ions con-firms the structure of Figure 1. Most per-sonality tests consist of one or more d imen-sions, with many items on the test measu r -in g each dim ension represented. Thus th eMinnesota M ultiphasic Personality Inven-tory ( M M P I ) has m any item s on the schiz-ophrenia scale, many on the depression scale,and some belonging to both scales. Theusual procedure in personali ty m easurementis to have many i tems repeatedly samplingwithin one or m ore dim ensions. However ,the questionnaires testing Figure 1 sampledtw o underlying dimensions but did not at-t emp t to develop tw o sets of homogeneousscales, one represent ing pure affiliation, andthe other, pur e interdependence. Rather ,the questionnaires used items hypothesizedto simultaneously measure both dimensionsto systematically varied degrees. Thus, th eitem for stimulate, teach (145) theoreti-cally sampled relationship in the a m o u n tof 5/9 affiliation and 4/9 dominat ion andwas not intended to belong either to a setof homogeneous i tems measur ing the be-havior trait called dominance or to a di-mension called affiliation. There is littleprecedent in the literature for at temptingmeasurement of behaviors no t clearly as -signed to specific scales, but the possibilityand need for such an approach has been rec-ognized (Horst, 1968).I t may be desired to evaluate th e abil i ty to reactappropriately to interrelated st imulus elements. Inthat case, a st imulus si tuat ion must be defined asa set of interrelated elements to which the appro-priate response is one which recognizes these in -terrelationships. Such types of stim ulus patternsi m p l y a highly sophisticated type of measurementand very little research has been done with them.

    We shall not go furth er into this apparent con-tradiction of the rule that stimulus elements shouldbe independent [p . 6].T he existing exception for the philosophy

    of having sets of homogeneous i tems mea-suring independent dimensions is the cir-cumplex method which has been used inpersonali ty m easurem ent research to system-atically sample points thought to lie in vary-ing degrees between two underlying dimen-sions. The circumplex method is an out-growth of Guttman's (1966) facet theorywhich defines variables systematically interms of their component facets. Foa (1961)explains his application of Guttman's facettheo ry :It seems indeed that a ci rcular arrangement canalways be described on two dimensions. On theother hand, not every two-factor structure willnecessari ly produce a circumplex. The circumplexrequires th e existence of an interrelationship be-tween the factors. A sufficient condition for acircumplex is that the factor loadings of everyVariable i, belonging to the set, stand in the rela-t ionship : cVt + k'b\ = h " ,where c, k, and h are arbitrary constants, and oan d 61 are the loadings of Variable i on the firstand second factors, respectively.This is the well known equation of the ellipse,When this relationship between factor loadings ex-ists, the predicted correlation coefficients, r^ t =atai + bibi can be ordered in a ci rcumplex pat-tern [p. 346].

    In application, (G uttm an, 1966) a ci rcum-plex will yield a matrix wherein ". . . thehigher correlations are found near th e maindiagonal ; moving away from th e diagonalcell the coefficients decrease and then in-crease again [p . 455]." Although Figure 1is based on absolu te values rather thansquares of the components, the circumplexrationale can be applied because the sequenc-ing of points is the same in a d iamond as itwould be in a circle.Table 1 presents th e intercorrelationsa m ong maternal ratings of 221 children(171 normal , 50 psychiatr ic; Samples 1 and2 combined ) in t e rms of the 36 points onthe childlike surface of Figure 1. The cir-cumplex pattern is c onf i r me d ; loadings nearth e diagonal are high and positive (in the.40-.50 ranges; d j = 2\9, r = .14, signifi-

  • 7/30/2019 Benjamin 74

    15/34

    406 L O R N A SMITH BENJAMINTABLE 1CORRELATIONS A M O N G CHART

    Chartp o i n t s2102412422432442452462472482402382372362352342332322 3 12302 2 12222232242252262272282202182 1 72162152142 1 32 1 22 1 1

    210 241

    28

    242

    1349

    243

    265 142

    244

    06332929

    245

    1438424436

    246

    100-15

    08-0205

    247

    0100816283220

    248

    0610072 12 3292 140

    240

    122 12423341 3042324

    238

    009241 5261 707363036

    237

    -04051 020352 223392 11 709

    236

    -14-24-21-06-01-09432 21 3092 124

    235

    1 2-031 1

    1 53 110253426233 1243 5

    234

    -10-20-23-11-03-02452 11 402

    -022 73720

    233

    0804011 4

    -052734302 50

    09141 91 92 5

    232

    -23-37-39-30-14-16260 1

    -01-16-05

    1 640064005

    231

    -22-27-3 7-19-11-1428040

    -14-09

    1 73910450573

    Note. Total subjects (N = 221) Include 50 child psychiat ry patients and 171 normal children.

    cant at the .05 level), and those in the mid-dle range are distinctly negative (quite afe w are significantly negat ive) . Proceedingaway from the diagonal, there is a gradualre turn of the rs to the original high-positiverange. This same circumplex pattern w asobtained in other samples for both the par-entlike and the childlike surfaces. In noanalysis w as there any shuffling of the pointsin Figure 1 to obtain a better c ircumplex.The arrangement of the Figure was on ana priori basis, and Table 1 followed directly.Ultimately Figure 1, as w ell as the qu estion-naire , may be refined on the basis of suchpost hoc rearrangements, but the order in

    Table 1 was not due to any such "trial anderror" manipulating.The circular ordering shown in Table 1was consistent with the circular orderingsuggested by the autocorrelations of Figures3 and 4. However, the data in Table 1were based on between-individual compari-sons. Thus, the structure hypothesized inFigure 1 is confi rmed by both within-indi-vidual , and between-individual comparisons.

    Factor Analysis: Between-Subjects AnalysisFactor analysis yields the proposed under-lying dimensions an d generates a > reasonablefacsimile o] Figure 1. Factor analysis is

  • 7/30/2019 Benjamin 74

    16/34

    S T R U C T U R A L ANALYSIS OF SOCIAL B E H A V I O RP O I N T S A R R A N G E D I N C I R C U M F L E X O R D E R

    407

    230

    -12-21-26-21-06-07280907

    -10031 23407431 25 16 1

    221-19-36-31-20-12-163009

    -02-14-011 9431 54609666753

    222

    -10-12-34-07-1907241 01 0

    -24-12

    1 428

    -11392639383 141

    223

    -18-28-24-1504

    -10291 51 7040823371 53 71 24134444025

    224

    -14-22-27-11-18-01300803

    -11-0920421 2391 6484736534630

    225

    02-04-050

    040

    260

    04-15-0420201 524162 73 12 12 7332 226

    226

    -19-28-28-17-04-12350

    05-04061 6472 14604565 7465430534629

    227

    -11-13-16-0814

    -0424162007102 1281 3251 63336303423492 11 64 1

    228

    -17-21-22-20-11-11280502

    -08-02

    16411 14016535 75 243334144385330

    220

    -24-22-27-20-09-12

    1 60402

    -15-06

    1 132072 70 15249325329343027432 342

    218

    142525162519

    -151 809060119

    -03090

    -10-09-070

    -030

    -04-02-08-0610

    -020 1

    217

    1 6392 1242 724

    -10030107

    -021 7

    -14-02-04-10-14-05-09-020

    -06-06-05-0403

    -12065 3

    216

    2034382 54827

    -172 31 320203 1

    -0424

    -0301-14-10-08-08-18-10-11-08-14

    0 1-16-084346

    215

    083930293328

    -061 109051 31 5

    -09040 1

    -07-10-04-03-10-02-05-07-01-0804

    -08-0455 554

    214

    14342840284607271 6030630

    -0305011 3

    -14-07-11-0405

    -1002

    -02-1404

    -13-0336394654

    213

    1 53938341640

    -120902160

    1 2-15-04-0803

    -28-21-16-14

    0 1-19-09-15-18-10-20-154 13641395 1

    212

    344340381 538

    -021 41 01 8021 3

    -151 0021 8

    -27-21-17-17-11-15-04-07-12-08-12-23323344334557

    211

    224450431 548

    -03091 1061 603

    -2404

    -101 8-31-29-13-24-04-30-11

    01-26-10-20-291 4141 719273 144

    commonly used to test and/or constructpersonality theory on an em pirical basis, andwh en used in this way usually results inm a n y dimensions (e.g., 16 personality fac-tors). However, some critics (e.g., A r m -strong, 1967) feel that such empirical useof factor analysis "... m a y b e misleadingas far as the development of theory is con-cerned. The use of com prehensive and ex-plicit a priori analysis is proposed so thatthere will be independent cri teria for evalua-tion of factor analytic results [p, 17]."I t may not be necessary to choose betwe enthe extremes of constructing a model on apurely rational basis and testing it empiri-

    cally or having no model at all and reportinga complex of empirically derived factors.Such a compromise in approach is i l lustratedby Schaefer (1965) who em pirical ly d erivedfactors for describing parent behavior, con-structed a rational model (circumplex) usingthese factors as a basis, and then tested th em o d e l fu r ther with factor analyses of differ-ent sam ples. His factor analytically basedmo de l was, as indicated above, the basis forthe parentlike surface of Figure 1. Schae-fer 's preference was to use the principalcomponents analysis with varimax rotationof three factors (acceptance versus rejec-tion ; autonomy versus control ; lax versus

  • 7/30/2019 Benjamin 74

    17/34

    408 L O R N A S M I T H B E N J A M I NTAB LE 2

    F A C T O R L O A D I N G S F R O M P R I N C I P A L C O M P O N E N T S A N A L Y S I S F O L L O W E D B Y V A R I M A X R O T A T I O N

    DisaffUiat ive char t p o in t

    Go away nowEx c l u d eIsolateNeglect, ignoreB l u f f , illogicalAb a nd on , rejectDepriveStarve, poisonMu r d e r ou s at tackI n j u r eFr ightenRestrain, overprotectShout, crit icizeThreatenExploitAuthor i t a r ianSham e, gu i l t controlDo min a t e

    L o a d i n gCode128127126125124123122121130131132133134135136137138140

    l403547607735566168456969667578113308

    2-31-11-11-23-17-53-28-55-12-32-28-08-35-26-20-07-35-22

    3452640213018431312442723263023805776

    43952273522483213021215

    -20-06261305-20-11

    Affiliat ive c h a r t point

    E ma nc i p a t eY ou can do i tEncourage divergenceListen, equal i t a r i anExplore, le t discoverConf i rm , praisePlay , al low peer playSmile, gree t wa r m lyKiss, heal, groomEm brace , t ender touchSuppor t , cradle, nurseIndulge , T L CProtec t, keep com panyReasoned persuasionSt imula te , teachOverindulgeI n t r u d ePossessive

    L oad ingCode120118117116115114115112111110141142143144145146147148

    l-06-34-40-57-54-67-20-12-32-13-46-44-09-68-66-046331

    214425146625773698583737168323734-1206

    3-06-37-39-27-15-05-09-39-06-03-20-04-190605335066

    4613139152910-05-05-02-200417-3802-02-51-1110

    Note , The 110 adu l t s rated their m em ory of their m other ' s parent l ike behavior . The f irs t factor is nam ed disa f f i l ia t ion; th esecond, affi l iation; the third, power ; and the f o u r th , emanc ip a t ion .

    firm control) and a typical analysis (e.g.,Schaefer, 1965) accounted fo r 66% of thevariance.Applications of principal com ponents anal-ysis with varimax rotation to the di fferentsamples testing Figure 1 generally yieldedresults consistent with the figure. An ex-ample is presented in Table 2; data werefrom th e 110 individuals who rated the en-tire battery of questionnaires (Samp le 4),th e part reported in the table being f romrecall of mother 's parentl ike behavior whenth e raters had been in age ran ge 5-10 yea rs.The table presents items in a f o r m a t iso-metr ic wi th Figure 1; affiliative i tems ap-pear on the right-hand side of the t ab le ;disaffiliative i tems appear on the left-handside. The top of the columns representi tems close to the independence pole, and thebottom of the columns represent th e domina-tion pole. The balance of the item s arearranged between these poles in order ofth e degrees to which they represent th erespective poles. Factor 1 has high-posi t iveloadings on the disaffiliative side of the table,and distinctly negative loadings on the affili-ative side; this factor is named "disaffilia-

    tion." The second factor is j u s t th e opposite:negative loadings appear on the disaffiliativeside, and positive loadings appear on theaffiliative s ide ; this factor is na m e d "affilia-tion." On the right-hand side of the table,the third factor has negative loadings fori tems representing the emancipation pole,and loadings shift to positive at the dom i -nance pole. This trend is not so clear forth e third factor on the left-hand side of thetable, but the factor is nevertheless named"power." On both th e right- and left-handsides of the table, th e fourth factor showshigh-positive loadings at the emancipationpole and negative loadings at the dominancepole. It is named "emancipation." Thusfour factors emerged which correspond toth e four poles of Figure 1, and these fourfactors accounted fo r 64% of the variance.These same four factors a lmost a lwaysemerged in the different samples studied.The degree of factor similari ty among tw osamples as measured by the Wrigley andN e w h a u s .coefficient of congruence (Har-m an, 1960) was .95 for the disaffiliationfactor, .89 for the emancipation factor, .90for th e affiliation factor, and .85 for the

  • 7/30/2019 Benjamin 74

    18/34

    S T R U C T U R A L A N A L Y S I S O F S O C I A L B E H A V I O R 409power factor. This comparison of factorsimilarity was between an analysis of adultsrat ing their m em ory of their m other 's par-ent l ike behavior (Sample 4, nHQ) andan analysis of mother 's rat ing their ow nparentl ike behavior in relation to their chil-d ren (Sample 1, w=171).Disagreements about how m a n y factors toextract from a factor analysis are wide-spread, and the arguments appear to be ofimportance because, in principal componentsanalysis (Cooley & Lohnes, 1962, pp. 151-153), th e n u m b e r of factors extracted is sup-posed to be a measure of the true dimension-ality of the dom ain und er study. Forexample, in factor-analyzing results of in-telligence tests, the number of factors ex-tracted is often interpreted as an indicationof how m a n y di fferent abili ties comprise in -tell igence. Thus, w hether the analysis yield s2 or 12 factors can m a k e a difference in theresulting description of intell igence.If the number of factors to be extractedfrom th e data testing Figure 1 was to bedetermined by the rule that each be asso-ciated with a latent root greater than orequal to 1, slightly less than 20 factors usu-ally emerged , and they accounted forslightly more than 70% of the variance aftervarim ax rotation. The reduction of thenumber of factors to 4 was done on the basisof the observation that the first 4 factorsemerging related logically to Figure 1 andaccounted for almost as much variance ascould be accomplished by rotating a muchgreater n umber of factors.Convention would suggest that th e emer-gence of four factors associated with th e fourpoles on the surfaces in Figure 1 necessarilymean s that there are four orthogonal dimen-sions, not two. However, th e view that fac-tor analysis uncovers "true" dimensional i tyhas been challenged by G ut tman (1966)w ho reviewed th e multiple factor analyticefforts of many invest igators and r e m a r k e d :In these algebraic approaches the notion of orderamong va r iab les is absent. More ser iously, th eapprox imate computa t iona l p rocedures used ac tu -al ly have bl inded researchers f rom seeing simpleorder pat terns in their own data which may haveimportant psychological implications [p. 444].Guttman proceeded to m a k e his point by

    reanalyzing th e data from a published studyof number abi l i ty which had yielded 10 com-m on factors by the centroid m ethod (a s im -plified approximat ion to the principal com-ponents solut ion according to Cooley &Lohnes, 1962, p. 153) and showed how 5out of the 10 factor s could be p lotted in asimple two-dimensional scheme. He de-scribed his method of finding order amongmany factors as "a simple trial-and-error-graphic method" which plots the correlationsin t e r m s of the approximate rank order ofth e sizes of the r. G ut tman (1966) com-ments on the two-dimensional figure he gen-erated :Perhaps the most s t r ik ing feature . . . is that itsucceeds in por t ray ing in two dimensions th e struc-tu re of the inte rrel atio ns of seventeen observedtests despite that fact that the conventional factoranalys is or iginal ly m a d e of the data prescr ibesfive dimens ions (or common factors) for thesesame tests. If the n u m b er of dimensions is re-garded as a criterion for parsimonious analysisthen surely the two dimensional port rayal is m o r eparsimonious than the five-dimensional [p. 450].Later, Guttm an (Schlesinger & Guttm an,1969) published a computer program d e-signed to plot such a two-space for severalfactors to "show how a certain definitionalstructure of the test variables is reflected ina two-space [p . 95]" and called his approach"smallest space analysis."Guttman approaches factor analysis withconsiderably less reverence and more mat ter -of-factness than wri ters who view i t as thekey to underlying dimensionality. He notes(Gu ttm an, 1966) : "As has been pointed outelsewhere a 'factor' in the sense of conven-tional factor analysis is essentially but aweighted average of the observed tests andcan be regarded s imply as an addi t ionaltest [p. 447]."

    With a similar approach to factor analysis,it is possible to take the factors of Table 2,and on any reasonable basis (i t need not beGuttman's s imple t r ial-and-error-graphicm e t h o d ) reduce them to a simpler two-spaceto see wh eth e r the final result will corre-spond to Figure 1. This is easy to accom-plish by a s ingular t ransformation, T, whichsubtracts th e vector of factor loadings fo rdisaffiliation from th e vector of factor load-

  • 7/30/2019 Benjamin 74

    19/34

    410 L O R N A S M I T H B E N J A M I N

    ^.: { is'!-J| I '

    Is t ? ! ~ r.s ai Is s! i tl !" I'5 I *"tgl S

    r3. 'i ?

    a -" 1 isc ^ 1

    T

    ||. 5 r *" s^ -5 S o o

    1 s I f D c/i1 O V O C j * -

    ^ c^'5 "5 E* Oi ii;_co> tJ|IBs " ^ ^2 t5 M

    C j3i a 4 a ^S ^ 3 8 .a -r i r i 1 " 3 . J 21 i1"* Oot- j ,

    \ !is ga 5 2 2" * " ^ i s " D -

    5 f1 & S < 5 o*

    'a I I^ ij, ^ g < M'5 *. ? 1 I |\ l l \ l \n \ ft f ;^ m J2 s*: s 5 5 ||s\^s& a

    < vt-l

    44-4

  • 7/30/2019 Benjamin 74

    20/34

    S T R U C T U R A L ANALYSIS OF SOCIAL BEHAVIOR 411ings for affiliation to yield net affiliation (inTable 2, this would be Factor 2-Factor 1),and which subtracts the vector of factorloadings for power-helplessness from th evector of factor loadings fo r emancipated-beemancipated to yie ld net emancipation (inTable 2, this would be Factor 3 from Fac-tor 4).The resul ts of the t r a ns fo rm a t ion , T, ap-plied to the factors of Table 2 are shown inFigure 5. For ex am ple, the location of thepoint emancipate in Figure 5 was determinedby T: .14 - (.06) = .20; .61 - (-.06)= .67. Thus the values from the four-spaceshown in Table 1 for the point emancipateare reduced to the values for the two-spacein Figure 1, and the coordinates for the pointare (.20, .67). Similarly, the location ofth e point d om inate in Figure 5 comes fromTable 1: -.22 - (.08) = -.30; -.11 -.76= -.87; result: (.30, . 8 7 ) . The trans-formation, T, was applied to factor loadingsfo r each item and in the resulting Figure 5,m a n y points conformed quite closely totheory (140, 137, 148, 147, 138, 133, 134,123, 120, 117, 118, 115, 116, 111, 110, and143); a few points were not reasonably closeto expectation (124, 136, 126, and 128),and th e rest were moderately close to ex-pectation. Inspection of Figure 5 reveals atendency for clustering to occur around th epoles rather than for the points to spreadthemselves evenly in circumplex order. Thism ay be due to the fact that th e va r im a x ro-tation tends to yield high factor loadingsfor as few variables as possible (Cooley &Lohnes, 1962, p. 162). In other words,th e varimax rotation is structured to m a xi -mize the loadings of a few variables on eachfactor and let the rest of the variables haveloadings which approach zero. This wouldtend to force the p oints tow ard the poles ofthe figure. In summary , despi te some im -perfections, Figure 5 is judged to be a "rea-sonable facsimile" of the first surface ofFigure 1.A reasonable facsimile of the childlike sur-face of Figure 1 is presented in Figure 6and was obtained by principal componentsanalysis of maternal ratings of childlike be-havior of 171 normal children (Sample 1)

    using varimax rotation followed by thetransformation, T. This analysis rotatedratings of 72 items from both the parentlikeand the childlike surfaces, but similar fig-ures are obtained by separately rotating th e36 i tems from each surface. This invulnera-bility to whether parentl ike or childlikei tems were rotated separately or together isd ue to the fact that varimax procedure isnot affected by the number of variables ro-tated (Cooley & Lohnes, 1962, pp. 162-163).An al ternative means of reducing the datato two dimensions which could be relatedto Figure 1 m igh t be to ipsatize scores be-fore factoring for two factors. Accordingly,in Samples 1 and 4, each individual subject'sscore was first s tandardized to his/her ow nm e a n and sigma, and then principal compo-nents analysis w as followed by a var imaxrotation of two factors. This procedure ap-plied to data describing memory of mothers 'parentlike behavior (reported in Table 2 andFigure 5) yielded the two factors shown inFigure 7. The first factor to emerge showedhigh po si t ive loadings for points on the affili-ative side of Figure 1 and large negativeloadings on the disaffiliative side. The bowshape of the curves suggests that loadingswere greater fo r points near the poles ofth e affiliation axis (110 and 130), and thefactor is named affiliation. The second fac-to r tended to show progressively largerloadings for points nearer the dominancepole (140) of the interdependence axis andis named power . These tw o factors d o cor-respond to expectations based on Figure 1and respectively accounted for 24.0% and9.1% of the varian ce. Since the initialp roc e dure of extracting four factors and ap-plying T accounted for nearly twice as m u c hvariance, this ipsatizing procedure was notpursued fu r the r .In small samples, factor analyses of rat-ings in term s of the paren tlike and childlikesurfaces consistently conformed to Figure 1.However factor analyses of the part of theSeries C questionnaires (Samples 3 and 4)measuring generalized attitudes toward theself in te rms of the introject surface did notconform well to Figure 1. Such factor ana-lytic reconstructions of the third surface of

  • 7/30/2019 Benjamin 74

    21/34

    412 L O R N A S M I T H B E N J A M I N3fcoC3

  • 7/30/2019 Benjamin 74

    22/34

    S T R U C T U R A L A N A L Y S I S O F S O C I A L B E H A V I O R 413.8.1.6.3.

    ,1..1

    *"'3"-.3.4.5-.r-.11 118 117 116 115 114 113 112 111 110 141 142 143 144 145 146 147 148 140O 120 128 127 126125 124123 122 121 130 131 132133 134 135 136 137 138

    Parentlike Points

    ~ * 8'~ 118117 116115 114113 112111 110 HI 14 2 M 3 144 MS M 6 147148140. O 128 127 126 125 124 123 122 121 130 IX 132 133 134 135 13 6 137 138Panntliln Point*FIGURE 7. Principal components analysis of ipsatized scores followed by varimax rotation oftwo factors. (Two factors named affiliation and power respectively accounted for 24.0% and

    9.1% of the variance.)Figure 1 correctly placed most affiliativeattitudes toward the self on the right-handside of the figure and disaffiliative atti-tudes on the left-hand s ide ; but the

    vertical dimension was not as predicted.Self-emancipating attitudes did not consist-ently appear in the upper half of the figure,nor did self-controlling attitudes consistently

  • 7/30/2019 Benjamin 74

    23/34

    414 L O R N A SMITH BENJAMINappear in the low er half. This failure toreconstruct th e third surface in the interde-pendence (vertical) dimension was consist-ent across sam ples. Possibly it reflects afaul t in the theory, in the wording of thequestionnaire items, or a generalized con-fusion in people's attitudes about their ow nau to n o my from or interdependence with in -trojected significant others. Factor analy sisof an expanded sample of short-form rat ingsof self will, it is hoped, determine whetherth e problem was in the word ing of the long-form item s. Clinical interv iew s of subjectsin Sample 4 provided extensive subjectivesupport for the hypothesis that attitudes to-ward the self do relate directly to treatmentreceived from significant others (spouse,mother , father , im portant sibl ings) . M oreformal testing of the third surface of Figure1 will be forthcom ing elsewhere.Principle o f Complements

    Th e principle o f complements is confirmedby correlations between surfaces. The prin-ciples of complementary relations (and ofant idotes) should be tested by an exper imen-ta l setting wherein one behavior (e.g., shout,crit icize134; or confirm, praise114) isexhibited by the experimenter, and then in-dependent observers reliably rate th e sub-ject's response and f ind a high incidenceof the predicted behavior (e.g., cringe, de-fend, whine234; or display, cour t214) .This needed rigorous test of Figure 1 hasnot yet been perform ed . However , enthusi-astic if unscient ific re ports from psychiatricresidents, medical students, and pat ientshave supported th e hypothesis about comple-mentary relations and antidotes. A n anno-tated sequence of family interactions illus-trating these principles is available elsewhere(Benjamin , 1973, pp. 231-236).Correlations between maternal ratings oftheir ow n behavior and of their children'sbehavior (Sample \,n 171) are consistentwith the principle of complementary rela-tions. Tw o examples are shown in Figure8. The left-hand side of the Figure show sth e relation between maternal ratings of chil-d ren fo r Point 210, hug, embrace, and ma-ternal ratings of themselves for points on

    th e parentlike surface. As the parentlikepoints closer to the complement of Point210 (i.e., Point 110embrace, tender touch)are approached, th e rs become larger andmore positive. The right-hand side of Fig-ure 8 shows th e relation between ratings ofchildren on Point 218, unassaultive assert ,and maternal self-ratings of parentlike be-havior. Inspection of this part of Figure 8shows that correlations between childlike as-sertiveness (218) and parentlike points tendto become more posit ive as the complemen-tary parentlike Point 118 (You can do it)is approached. Thus, Figure 8 is consistentwith th e principle of complementary rela-tions between the parentlike and childlikesurfaces of Figure 1. The lef t -hand andright-hand sides of Figure 8 show that af-fection begets affection and independencebegets independence (andvice versa) ; theseprinciples have been repeatedly confirmed inthe child development literature (e.g., B a u m -rind, 1967; Schaefer & Bayley, 1963).Not every pair of complementary po in tsconformed as closely to theory as the twopairs (110 and 210; 118 and 218) shownin Figure 8. In the total analys is fromwhich th e examples presented in Figure 8were selected, many other points showed re-sults close to expectations based on Figure1; a few did not, and the rest were mod-erately close to expectation. The r betweenthe 36 maternal parentlike points and the 36predicted complementary behaviors for chil-dren w as significant in 28/36 = 78% of thepossible complementary pairings. A harshertest of theory was to determine whether th er between a given pair of complementarypoints was greater than all other rs betweenthat part icular parent l ike point (motherrated herself) and all possible childlikepoints (mother rated ch i l d ) . T h e m a x im u mr occurred exactly as predicted for the fol-lowing p a i r s : Mother (M ) 118-Child (C )218, M 116-C 216, M 114-C 214, M 113-C213, M 112-C 212, M 110-C 210, M 143-C 243, M 134-C 234, and M 148-C 248.The m a x i m u m r suggestive of complemen-tary relation held within one poin t for PairsM 120-C 218, M 115-C 214, M 111-C 210,M 132-C 233, M 142-C 241, M 135-C 234,

  • 7/30/2019 Benjamin 74

    24/34

    S T R U C T U R A L ANALYSIS O F S O C IA L B E H A V I O R 415

    o.+ S I-S'

    H2 -

    2-I8 -r

    S I-S'?

    H ? -io.T8;i8-r

    1

    (

    1I1 ' 11 1 . 1 .''

    1

    1

    mLx.ft ,T T T . %tv s * W 1 1 l>l-( - ~ V a a O I - w . u > 9 U O NO . O *> Uw g > ~ - o i / i 3 k V . - u _ 3 i D u i- e t c out

    v ' ' ? . j> . J j j _ r s i ; ~ J < J s J [ * l g 1 - "2 " " Vo1" ?-u c* E1) . C f l ) + ' < L ) 4 J Q ) a ' o u t n * - > UV >U > O U D 1 O C C"O t f f l J U 4- . f D W a i k . d i > O U *J "O *O

    . .o e x o 6 E 3 C k | i w j t c o o 3- x o U C 3 * J V J 3 ^f r r \ r ^ m f A r ^ ( ^ f ^ r g r g < ^ f M ( M f N < N P *

  • 7/30/2019 Benjamin 74

    25/34

    416 L O R N A SMITH BENJAMINand M 148-C 247; it held within tw o pointsfor M 127-220, M 126-C 228, and M 136-C 234. This represents 9 + 7 + 3 = 19/36= 53% of the pairs conforming closely toth e complementary prediction as measuredby location of max ima l r. The random ex-pectation of having the maximum r occurwithin two steps (either direction) of theexact complement is 5/36 = 14%. A simplechi square contrast of observed and expectedvalue is significant beyond the .001 level, x2= 42.34, df=l.When maternal ratings of children's par-entlike behavior were compared to maternalratings of their own parentlike behavior, re-sults suggested a strong imitative tendency.A significant r occurred in 25/36 = 69% ofthe imitative pairings, and the max ima l roccurred at the exact point of imitat ion forPairs M 118-C 118, M123-C 123, M 113-C 113, M 122-C 122, M 112- C 112, M 110-C 110, M 131-C 131, M 133-C 133, M 134-C 134, and M 137-C 137; it occurred withinone point fo r Pairs M 127-C 128, M 117-C 118, M 111-C 112, and M 141-C 110,and within tw o points fo r Pairs M 120-C 117, M 116-C 118, M 121-C 131, andM 142-C 110. This represents 11 + 4 + 4= 19/36 = 53% of the pairs conformingclosely to the imitative expectation. A chisquare test is significant beyond the .001level.In s u m m a r y , th e results are not perfectbut are nevertheless fairly consistent withexpectations based on Figure 1. The varia-bles involved in whether a child imitates themother , and/or assumes a complementaryrole, or does something else are not speci-fied, but some relevant clinical observationsin t e rms of Figure 1 appear elsewhere (Ben-j amin, 1973, pp. 236-237).High Reliability in Normal Samples

    Reliability is high in normal samples, andfailure o f reliability ca n characterise sub-jects. Since the purpose of this paper is toexpose th e theory of Figure 1 and to pre-sent tests of its validity as a construct ratherthan to propose a set of new personalityquestionnaires, a formal examination of thereliability of the questionnaire is not required

    here. For this reason, there will not be ex-tensive reference to technical aspects of theproblem of reliability.In Autocorrelat ions Among I tems, a co-efficient of internal consistency was devel-oped and offered as support for the constructvalidity of Figure 1. Such a coefficient ofinternal consistency is one measure of re-liability. Normal ly in instances where in -ternal consistency is high, stability over timeis also high (Fiske & Rice, 1955). A largesample of test-retest rs could confirm thisprinciple that a coefficient of internal con-sistency is related to a coefficient of stabilitysuch as test-retest r. Support for the ideathat internal consistency implies day-to-dayconsistency or stability w as found in a se-ries of over 70 interviews B with both normaland psychiatric subjects. Ratings yieldingorder ly autocorrelations such as those shownin Figure 3 to be associated with a highcoefficient of internal .consistency also de-scribed behaviors reported to be consistentand stable over time.During the interviews, maps and auto-correlations in the format of Figure 3 wereshown to each individual, and implicationsw ere discussed . Re sults suggested that ifth e internal consistency w as greater than orequal to .90, then the domain of behaviorsbeing described could be regarded as stable,predictable, and consistent from day to day.Autocorrelations which did not look so or-derly appeared as shown in Figure 9 andusually referred to a domain of behaviorsphenomenologically described as unpredicta-ble or changeable. For example, th e datashown in Figure 10 were based on theself-ratings of the childlike behavior of amiddle-aged male in relation to his wife.The couple's psychotherapist confirmed th esubject's description of himself in relationto his wife as tending to vascillate betweensubmissiveness and assertiveness, betweenaccepting and refusing interpersonal pos-tures, thus confirming the hypothesis of un-predictability in instances where internalconsistency fails. A nother exam ple of thisprinciple is offered by the medica l s tudent

    5 The assistance o f my colleague and fr iendSteven Troner is gratefully acknowledged.

  • 7/30/2019 Benjamin 74

    26/34

    S T R U C T U R A L A N A L Y S I S O F S O C I A L B E H A V I O R 417.70.60

    ! .5 0g .405| .30Q .20* .10G - o oi-.'o0 -.20" -.30-.40-.50-.60-.70 ft n

    D a a a a aD D D D OD ao D a D Q

    D Do u DD 0a a

    1 5 10 15 20 25 30 35S U C C E S S I V E L A G S

    FIGURE 9. Autocorrelations fo r self-ratings of thechildl ike behavior of a middle -aged ma le in rela-tion to his wife. (Autocorre la t ion curves such asthis which f a i l to show th e order i l lus tra ted inFigure 3 usual ly descr ibe a domain of behaviorscharacterized as unpredictable, unstable , or chaotic.)w h o , on seeing the disorder among auto-correlations between his ra tings of his fa-ther's parent l ike behavior , excla imed: "Inever k n e w w h a t to expect from h i m . Hemight be nice and f r iendly one m o m e n t , andthen WHAMO, th e next minute he 'd bebeating us up for no reason at all." Thisf inding that failures of internal consistencyreflect unstable unpredictable interpersonalpostures was confi rmed in nearly 100% ofthe interviews.If i t is correct to reason that failure toobtain a high degree of internal consistencyrepr esen ts the sign ificant psychological phe-nomenon of behavioral chaos and changerather than "failure" of a test, then it wouldfollow that internal consistency should behigher in normal samples than in psychiatricsam ples. The hyp othesis is conf irm ed sinceth e average correlation between th e auto-correlation curve and the inverted normalcurve for the childlike behavior of 171 pedi-a tr ic outpa t ients (Sam ple 1) was .92, where-as the comparable correlation for a sample(Sam ple 2) of 50 child psy chiatry patientswas .68. In a sample of 60 normal volun-tee rs (Sample 4) the average r betweenautocorrelat ion curves and the inverted nor-m al curve for childlike behavior in relationto a significant other person w as .97, whereasthe same coefficient of internal consistency

    for 50 (also Sample 4) psychiatr ic patientswa s .81. W itho ut exception, for every setof ratings, th e average internal consistencyw as greater in the normal sample than inthe psychiatric sample.The idea that reliability is a property ofpeople as well as of scales has been ex-pressed by others (Jackson, 1971, p. 243)and sim ilarly, lack of ord er (Foa, 1968) orreliability (Hersch & Scheibe , 1967) haspreviously been seen as reflecting psycho-pathology.Although internal consistency is m o r echaracteristic of normals, it can appear inpsychiatric subjects and can also reflect psy-chopathology. For example, Figure 10shows a map of the self-ra tings m ad e shortlybefore th e sub je c t m a de a serious suicideattem pt. The correlation between the auto-correlation curve for the ra tings shown inFigure 10 and the inverted normal curvew as .9 8 showing tha t murderous a t tack hadbeen directed toward th e self with impres-sive consistency just before this suicide a t-tem pt. No one wou ld be incl ined to labelthis highly consistent and integrated disaffi-liation toward self as "good" or "healthy,"so autocorrelation curves should be inter-preted in conjunc t ion wi th th e m a p s of theratings.The autocorrelations shown in Figure 11were obtained from a medical student re-sponding to the short-form questionnairewith ratings of his m e m o r y of his fa ther.The map appearing at the top of Figure 11shows that opposing tendencies were en-

    FIGURE 10. Map of self-ratings completed shortlybefore the subject made a serious suicide attempt.(Autocorrelations associated with these ratingsshowed a high degree of internal consistency ori-ented around self-destruction.)

  • 7/30/2019 Benjamin 74

    27/34

    418 L O R N A SMITH B E N J A M I N

    N e g l e c t , IgnoreExclude 95

    50

    5 0 E m a n c i p a t e

    2525

    T h r e a t e n 100A u t h o r i t a r i a n 100

    D o m i n a t e 100

    E n c o u r a g e d i v e r g e n c e

    9 9 P l a y , a l l o w p e e r p l a y0

    5 0 S t i m u l a t e , t e a c h75 I n t r u d e

    2 4 6 8 10 12 14 16 16 20 22 24 26 28 30 32 34 36Successive Laas

    F I G U R E 11. Map and autocorrelations of a male medical student rating hism e m o r y of his father. (The autocorrelations increase as more oppositepoints are paired, indicating ambivalent memory and/or double-binding be-havior by the father.)

    d orse d : neglect, ignore (125) and stim u-late, teach (145) both received above-m edianra t ings; exclude (127) and intrude (14 7 ) ;emancipate (120) and dominate (140); en-courage divergence (117) and authoritarian(137) all received relatively high degrees ofend orsem ent. These opposing tend encies orambivalences are reflected in the autocorrela-tions shown at the bottom of Figure 11.Correlations become progressively larger aspaired points become more oppositional.Autocorrelations falling into a pattern likeFigure 11 m ay be one answ er to the prob-lem of how to define and study am bivalence.(Minkowich, Weingarten, & Blum , 1966)

    Clearly there is a need for methodological advanceto encourage frui tful exploration of a topic socentral to theories of persona lity form ation [i.e.,ambivalence]. The development of a simple, easilyadministered technique for the assessment of indi-vidual differences in ambivalent feelings, generallyapplicable in studies of interpersonal relationships,is a worthwhile objective in itself [p . 32].There ha s also been a test of agreementbetween persons rating videotaped inter-

    views in terms of Figure 1. Two femalegraduate students6 initially unfamiliar with8 Acknowledgment and thanks are extended toLinda Kasuboski and Eileen Weil fo r m ak i n g th eratings. Frederick Brown and Dennis Cook alsocontributed generously to this effor t through their

  • 7/30/2019 Benjamin 74

    28/34

    STRUCTURAL ANALYSIS OF SOCIAL BEHAVIOR 419Figure 1 spent tw o training sessions learn-ing to rate 15-second segments of videotapein t e rms of quad ran t s of Figure 1. Col-lapsing points into quadrants meant thatraters only had three judgments to m a k efor each interaction sc ore d : (a) Does it in-volve concern with the other and what is tobe done to or for the other (paren t l ike) , ordoes it involve th e self and w h a t is to bedone to or for the self (ch i ld l ike) ? (b) Isit f r iendly or unfr iend ly? (c) Is it concernedwi


Recommended