neuroticism

neuroticism

THE RELATIONSHIP BETWEEN ORAL CHARACTER TRAITS AND EXTRAVERSION!NEUROTICISM WILLIA.\I Department of Psychology. Um\ersity of J. CORULLA Swcastl...

903KB Sizes 9 Downloads 164 Views

THE RELATIONSHIP BETWEEN ORAL CHARACTER TRAITS AND EXTRAVERSION!NEUROTICISM WILLIA.\I Department

of

Psychology.

Um\ersity

of

J.

CORULLA

Swcastle upon England

Tyne.

Newcastle

upon

Tyne

NE?

JHH,

present study was a factor-analytic study attempting to uncover the relationship between the psychoanalytic concept of oral traits and the personality dimensions of E and N. Two hundred and fifty-four Ss completed the Eysenck Personality Inventory (EPI) and the written form of the 19 oral scales designed by Goldman-Eisler. The data was subjected to factor analysis. It was found that five factors were necessary to account for the majortty of the observed common variance. The extraction of five factors was contrary to the extraction of tuo factors in the Goldman-Eisler study. Path analysis. using multiple regression, uas performed to find to what extent E and N influenced the I9 oral scales. It was found that 15 of the 19 scales were significantly causally inlluenced by E and N. Because there were discrepancies between the correlations in the present study and those reported by Goldman-Eisler. and discrepancies m the number of factors to emerge were found, it U’BS decided to subject the original correlation matrix of the Goldman-Eisler study to re-analysis. The results of this reanalysis demonstrated serious mathematical flaws in Goldman-Eisler’s results.

Summary-The

INTRODUCTION

Frieda Goldman-Eisler (1951) conducted one of the few factor-analytic studies investigating the psychoanalytic concept of ‘oral character’ and the aetiology of the ‘oral syndrome’. The psychoanalytic hypothesis is that ‘early weaning’ causes the child to become pessimistic, aloof, passive and so on. To investigate the existence of the ‘oral syndrome’ Goldman-Eisler administered verbal self-rating scales for 19 ‘traits’, mentioned by psychoanalytic writers as having an oral connotation, to I I5 middle-class, male adult Ss, having an age range of IS-35 yr. Goldman-Eisler also collected ‘weaning’ data, retrospectively, from the Ss’ mothers. Although the traits were considered to be of an oral nature, the scales and items which comprised the scales were designed by GoldmanEisler. Each scale contained between 6 and IO items. The split-half reliabilities of the scales seemed adequate. with a range of 0.56 (Dependence and Conservatism) to 0.90 (Optimism). Intercorrelations between the scales were run and two factors were extracted from the correlation matrix. The plot of factor loadings indicated that the first factor to appear was similar to the hypothetical trait of oral optimism vs oral pessimism. Early weaning was seen to have a statistically significant saturation on the pessimism side of this factor, which appeared to link the hypothesized syndrome and early weaning. Factor 2 appeared to be characterized by, impulsiveness, deliberation, autonomy, aggression, oral aggression, change and conservatism, having factor loadings of: 0.782, -0.776. 0.600, 0.476. 0.3 18, 0.275 and -0.438, respectively, as measured by the Goldman-Eisler scales. This study is cited as one of two factor-analytic investigations that provides support for the psychoanalytic concept of oral character, the other being the study conducted by Lazare, Klerman and Armor (1966). Both of these studies have been reviewed by Kline (1972, 1978) and both have been accepted as supportive evidence for the oral character, with slight reservations, based mostly on the methodologies involved in both studies. However, Eysenck (1970) points out that although the environmental hypothesis offered in psychoanalytic writing seems to have been confirmed, the evidence is not sufficient to reject other hypotheses. For example, if we accept that introversion is an inherited trait, it may be that what psychoanalytic writers refer to as ‘oral pessimism’ may really be just another name for introversion. Implicit in the above alternative hypothesis is an explicit causal statement. The statement would go as follows; the E-I, N-S dimensions of personality are biologically based and therefore largely I83

geneticall> determined. The so-called ‘oral character’ traits are extensions of these biologcall\ based personalit) dimensions. Therefore. E and N \vould be expected to ha\e a significant U&I Influence upon the 19 traits mentioned bL pshchoanai> tic irriters. The present study Lvas a factor-analytic-study designed to test the hypothesis that the oral scales designed by Goldman-Eisler would be accounted for in terms of the Eysenckian dimensions of personalit) as measured by the Eysenck Personalit! Inventory (EPI). The justification for such a research project lies in the fact that the Goldman-Eisler ( 1951) study is presented 3s supporting e\,idence for the psychoanalytic concept of the ‘oral character’ to undergraduate students of personalit>, psychology. both in tvritten publications (Kline. 1972. 1978) and in lecture courses. hIETHOD .A heterogeneous sample of 35-i Ss were administered the EPI Form A and the written version of the Goldman-Eisler (1951) oral scales. Both questionnaires consisted of dichotomous items requiring a Yes or No response. The questionnaires were numbered as per S. i.e. the number on each set of questionnaires corresponded to the number assigned to the individual S. so as to ensure that the data for each S would not be mixed up ivith another S’s data. The Qlre.stiotltlctire.c

The u.ritten form of the Goldman-Eisler Oratity Questionnaire contains 19 scales. Each scale contains bet\veen 6 and IO items. totalling I50 items. Each scale is described, psychoanalytically as follo~rs (the number at the end of each description refers to the number of items in that scale): Optirrrisrt~, the tendency

to underestimate

Pcxsitttktt~. the tendency to overestimate Prt.s.ric~it~‘.passive-receptive attitude (8); L:rtnttclitttrhilit_r~. intense

attainability

desire

to climb

obstacles (5); obstacles (S); combined

with

a feeling

of un-

(9);

Orirl-rr~~rL~.s.viotl. aggressive use of speech (7); .-lggre.c.siotr. the desire to injure or inflict pain

(8); .-l/oc?fize.s.s,negative tropism for people (9); Attzhition. tendency to overcome obstacles (7): A~rtotwttz~~,the wish to neither lead nor be led (IO); Depenrlet~ce. tendency to cry, helplessness and insecurity (6); Guilt, self-torture and self-abasement (7); C’hange. a tendency to move and wander (8); Comer~~tisnr, adherence to certain places. people and standards /rtzpul.sion, tendency to act quickly without reflection (IO); Dcliherirtion, reflection before action (IO); E.~-oc,nt/le.*.ris. occupation with external events (6); Etdocathe.usi.s, preoccupation with internal activities (IO); :l’ztr.tztratzce, tendency to nourish, aid and protect (7); Socinhilitr~, desire to form social relationships (6). RESULTS

AND

(6);

DISCLSSION

It will be recalled that Goldman-Eisler (1951) used the split-half method of reliability analysis and obtained coeflicients of bet\veen 0.56 and 0.90. The split-half method is not the most efficient method to employ when handling dichotomous data. Nunnalty (1978) in his chapter on reliability states that “The only reason for employin g a split-half method occurs when the items are scored not dichotomously. but on three or more points”. For this reason, and because the scales and items used by Goldman-Eister (195 I) were designed on the criterion of face-validity only, reliability analysis employing Cronbach’s r kvas performed in this study. The reliability coefficients obtained using this method were found to be between 0.391, Conser\,atism and 0.690, Guilt (see Table I for a comparison of ‘Split-half‘ reliability coefficients obtained in the Goldman-Eisler study and those obtained in the present study).

Oral

character

traits

and E S

0 90 n ‘4 0 XI 0 7s I) 76 0 78

0 566

0 73

0.539

0 65

0.64

0.8s

0.647

0.56

O.IYi

0 90

0.690

0.71

0 -186

II.56

0.391

0 S?

0.653

0.87

0.505

0 72

051-l

0 73

0 556

0.79

0 433

0 70

0 623

As can be seen from Table I, the reliability coefficients obtained in the present study were between 0.073 and 0.392 lower than those reported by Goldman-Eisler (1951). The possible reasons for this discrepancy are: (1) the mathematical techniques involved in these measures of reliability are different; (2) that the items are not as internally consistent as first shown by Goldman-Eisler; and (3) that the changes in social attitudes over the past 33 yr have affected the way in which Ss responded to the items. For example, the statement “Atomic energy will bring progress and happiness to humanity”, which was said to be a measure of optimism, may have been answered quite differently in 1984 than it was in 1951. The fact that 87% of Ss responded in the negative to this item, and out of these 65% scored 5 out of the possible 8 on the Optimism scale, lends support to the idea of social change affecting responses to personality questionnaires. Intercorrelations were run between the I9 oral scales and E and N. A 21 x 21 matrix was produced (see Table 2). As can be seen from this matrix 8 oral scales were observed to have statistically significant (P > 0.05) correlations with E and 7 oral scales had statistically significant (P > 0.05) correlations with N. This pattern of correlations, by itself, suggests that I5 of the 19 oral scales may well be causally influenced by E and N. Some interesting discrepancies were observed between the correlations obtained in the present study and those reported by Goldman-Eisler (1951). For example, Optimism and Pessimism were observed to have a negative, non-significant correlation (-0. IO) in the present study, whereas, these scales were reported in the Goldman-Eisler study as having a correlation of -0.63. This low, non-significant correlation obtained in the present study suggests that Optimism and Pessimism may not be bipolar in nature ‘as measured by the Goldman-Eisler items’. There were other such discrepancies between both these matrices (compare Table 2 with Table 7, where the original Goldman-Eisler matrix is reproduced). A principal axes (PA/Z) method factor analysis was performed on the correlation matrix of the present study. In keeping with the Goldman-Eisler study the factors were not rotated. This analysis was performed on the 19 oral scales alone and then repeated after the inclusion of E and N. In both cases five factors were extracted, contrary to the extraction of two unrotated factors in the Goldman-Eisler (1951) study. Because of the difficulty encountered in attempting to interpret these unrotated factors, the present writer could only offer ‘General Orality’ as a tentative label. The PA/2 method was repeated as above with the exception that Varimax rotation was performed. In both cases five factors were extracted (see Tables 3 and 4). Table 3 shows the factor matrix produced when the I9 oral scales were factor analysed without E and N included in the analysis (see Table 3). Factor loadings of 0.40 and above were considered to be statistically significant. As can be seen from Table 3, Factor I can be tentatively labelled as ‘Aggression, Autonomy and Aloofness’. Factor 2 can be tentatively labelled as ‘Guilt, Dependence, Unattainability and Pessimism’. Factor 3 can be tentatively labelled as ‘Deliberation vs Impulsion’ and ‘Conservatism vs Change’.

Z2 -

=-I --

=

c

c,

c

c-0

_

--7=

= --_==

-0

- 1) 01

II

-0

-UO‘

01

36

0 45

n us

0 23

0 35

0 05

II IUS 1) II?

0 ‘Y

0 j7

0.02

0 UI)

0 56

0 Oj

-0.17

0 66

I) 0:

-0

0 13

0

0 I-I

0 I5

0

0 bY

IY

-0Oj

-I) 0')

0.61

- 0 IOU

OfA

0 18

0.10

002

0.32

0.35

0.08

0 00 0 I’ -0

IO

-008

, i? _.__ 34 -I

0 13

0 I3

0 I6 I) IT

-II08

-00s

0.15

I) I4

n ii

-0.14 0 17

-0

39

n II

-0.2 ‘3.63

- 0.0 I I) I I

-0-10

0 33

0 43

1) 05

0 OY

0.05

0 07

0.0X

-0.02

I) OS 006

0.10

-007

O.l11 -0.i-1

32

2 16 2')-I

-0

31

0 04

53 72

Ii9

0

140

Factor 4 can be tentatively labelled as ‘Exocathexis vs Endocathesis’, Optimism. Nurturance and Ambition. Factor 5 can be tentatively labelled as ‘Change vs Conservatism‘. The Scree test (Cattell, 1966) was used as a criterion for discontinuing factoring. Although the eigenvalues for Factors 4 and 5 in both analyses were below I, the Scree test indicated that there were five significant factors present. Under the alternative hypothesis, “that E and N have a causal influence on the I9 oral scales”, it would be expected that E and N would emerge as prominent factors when included with the I9 oral scales in a factor analysis, and that the 19 oral scales would load appropriately with the factors E and N. To test this prediction a PA/2 factor analysis was performed on the I9 oral scales with the inclusion of E and N. Table 4 shows the factor matrix produced from this analysis (see Table 4). As can be seen from Table 4, Factor I would appear to be an Extraversion factor with the subfactors of Impulsion, Exocathexis and Sociability loading in the expected fashion. Factor 2

FI I.

Oplimlsm

FZ

0.12

F-l

F3

I

-0.1

F5

0.51

-0.05

2.

Pessimism

-0.07

0.36

0.43

0 I-l

3

P‘Isslvlly

-005

0.28

0.30

0 IO

-0.06

4.

Unartninabilay

-0.01

0.48

0.37

0.27

-0.09

5.

Oral-aggression

0 02

0.56

6.

Aggrewon

0.00

0.63

0.16

0 54

0.1 I

0. I j

7.

Aloofness

S.

Ambition

9.

Aukxxxny

0 IS

I -009 -005 - 0.03 0.3

IO

Dependence

Il.

Gwlt

-0

I?.

Change

I3

Consrrvaunm

0 19 -0.47

I4

lmpulslon

15.Delibastion

0.06 16

0.59

Exocathexsis

17.

Endocnthrxsis

IY.

Nurturance

19.

Sociability

0.40

20.

Extraversion

0 35

I.

Neurotiasm

0.45

0.66

0 00

0.0 I

0.03

-0.2.l

-0.09

0.03 -11.10

0.20 0 16

0 45 -0.05

0.21

0.61

-0.02

0.03

0.12

0.77

-0.01

0 IO

0. I7

0.1 I

0.19

0.05

0 19

0.06

0.00

-0.33

I 0 19

-0.00

0.07 -0 09

0.34 -0

-0

0.09

-0

-0.0s

0 39

0 17

0.09

0.53

-0.1

04

-0.02

0 31

-0.16

-0

-0.10

0.06

-060

I6

2

-0.06

-0.1

I

0.20 0.10 0.04

09

0 -17

-005

O?

0 24

-0.01

0.1’

0.16

II

0.73

0.1 I

Z 93

2.83

I ..l9

-0 0.9

0 05

13

0.23

I

0.54

0 191 11205

emerges as a Neuroticism factor with the subfactors of Guilt. Dependence and Lnattainabilit> loading in the expected manner. Factor 3 appears to be an A ggression. Aloofness and Pessimism factor. Factor 3 appears principally concerned with Optimism, Ambition and Nurturance. Factor 5 seems to be a Change vs Conservatism factor.

Fig. I. Path diagram showng the pathways from E and 9 to the I5 oral scales which arc signiticantly causallq influenced by E and N. The values shown on the paths are /?-heights. The curved hnr between E and N indicates that there is a correlation berwern E and N but the causal model does not specify this relationship.

The pattern of these factor loadings su,,ooests that E and N do, indeed. account for man>’ of the ‘so-called oral character traits’. It is also of interest to note that Optimism and Pessimism ha\e of these variables ‘as high loadings on .sepcrrait’ fi~ctors. Lvhich suggests the lack of bipolarity measured bq the Goldman-Eisler items’. Path analysis using multiple regression kkas performed on the data of the present study. E and N were designated the independent variables while the 19 oral scales were designated the dependent variables. Table 5 shows the results of this analysis (see Table 5). As can be seen from Table 5. E and N were obser\,ed to have a statistically significant causal influence on I5 out of the I9 oral scales. This is demonstrated by the I5 statistically significant B-weights (path coefficients) associated with E and N. Figure I is the path diagram obtained from this analysis. As can be seen this diagram demonstrates the causal relationship bet\veen E. N and I9 oral scales quite clearly. The pattern of these relationships suggests that at least IS out of the I9 oral scales can be accounted for in terms of the E and N dimensions of personality structure as predicted from the alternative causal hypothesis outlined above. CONCLUSIONS It is concluded that the pattern of results obtained in the present study are consistent with the causal model outlined in the Introduction. If we accept the evidence for the hypothesis that E and N are biologically based. and therefore largely genetically determined [cf. the twin studies conducted by Shields (1962) and Claridge, Cantor and Hume (1973) and the physiological studies of Concoran (1964) and Revelle, Humphreys, Simon and Gilliland (1980)]. we can conclude that the oral traits, which have been shown to be causally influenced by E and N, are also genetically determined and not caused by ‘early weaning’ as proposed by psychoanalytic writers. As has been seen, both factor analyses. with and without rotation, extracted five factors from the correlation matri.1 produced in the present study. It has been seen that the factors. whether rotated or not bear little resemblance to the factors produced in the Goldman-Eisler (1951) study. Because of these discrepancies and the differences between the correlations reported by GoldmanEisler and those reported in the present study (particularly the correlations between Optimismpessimism and Optimism-Passivity), it kvas decided to subject the original Goldman-Eisler (1951) correlation matrix to a factor analysis. Reproduced in Table 6 is the correlation matrix produced in the Goldman-Eisler (1951) study. A comparison of this matrix and that of Table 2 will show these discrepancies quite clearly. Reproduced in Table 7 is the original factor matrix produced in the Goldman-Eisler (195 I) study. Because the eigenvalues and the per cent of common variance accounted for by each factor were not reported in the Goldman-Eisler publication, these calculations were performed. As can be seen the t\vo factors account for 31% of the variance, leaving 69% unexplained. Factor analysis (both the principal-components and maximum-likelihood methods) were performed on this matrix. In both cases seven factors were extracted. It was then decided to constrain the computer package employed (SPSSX) to extract only two factors. This decision was made on the grounds that only two factors were extracted in the Goldman-Eisler (1951) study. A principal-components factor analysis was performed on the correlation matrix produced in the Goldman-Eisler study, with the above-mentioned constraint. The principal-components method is the closest technique to the simple-summation method available to us in SPSSX. Table 8 sho\vs the matrix of factor loadings produced by this analysis. The factors are not rotated. A comparison of this table and the reproduced Goldman-Eisler (I95 1) factor matrix in Table 7 will demonstrate the discrepancies found between these matrices. Technical

Difficulties

Encountered

(I) The computer programme (SPSSX) factor-analytic techniques employed: (A) the correlation produced;

matrix

is ill-conditioned

produced

in this Re-analysis the following

and a determinant

warnings of 0.0000000

in all was

I YO

MILLI\V

J

COKCLLA

-=

CI

-

-

r. r,

c, c

CI

=-P-_ -_===

^

=

=

-_=3=

=

FI

FZ

.-\100in.3s

0 709

0 162

Endocalhsxsk

0.513

Pssslmsm

0 501

P;lwr,,y

0.387

0.204

0.790

-0

IX

O.f!-ll 1.i

Conser~.Ar,sm

0.353

- 0.438

0.530

-0

I35

0 ‘9392

Dependence

0.347

- 0.244

0.535

-0.29-

AggressIon

0.235

O 176

0 521

Oral

a_eprewon Dchbcrarlon

0 193

0318

0 182

0 17s

-0

Guilt

0 163

-0

,Autonomy

0 150

Unaumablhty

0.05-I

0.201

052

0 782

0 I50 -0

059

376

0 130

0115

0 35693

0 338

0 35;

O.I-IIYJ

-0

219

0

Chary

-0

53

II.275

Soclabdlt)

-0

489

Nurrurltncr

-0.585

-0

Opumlsm

-0.694

-O.?YO

Ewcathewi

-0

-0

Psrccnklge

of

v;*rlil”cr

I) 52319

0 600

Ambltmn

0 n7’)

I 39 IS1

-0446 0 312

0 2Y392

0 166

0.031 I-l

0 IS9

0.310

0 6952

0.142

0 03436 0 33391

-0.528

0.W 0 _L “i!

-0

0.033

0 13693

I26

0 60993

0 0O-J

0.623%~

IY

660 -0

-0.770

I4

-0

(B) negative eigenvalues were found; (C) the correlation matrix is not positive (D) squared multiple correlations cannot

0 -0239 0.10522

0 IS3

-0.194

3 1127

33s

I009YI4

n -162 -0.1

29i

17

-0

-0006

-0012

3 257

0 30993

0.062

-0

Eyrnvaluc

-0699 -0 30’

0.43109

776

ImpUliiW

701

0. IY3

790

0.3iS73

definite; be found.

The ill-conditiotwci t?mtri.\An ill-conditioned correlation matrix is a matrix having an inconsistent pattern of correlations. It was thought that the ill-conditioned matrix might have been the result of rounding the correlation coefhcients from four to two decimal places. (The correlations were those in the correlation matrix reproduced in Table 6.) To test this possibility an example correlation matrix was taken from Child (1979) and all correlations were rounded from four to two decimal places to see the effect of such rounding (Table 9). The example is of the extraction of the first factor using the ‘simple-summation’ method (see Burt, 1940). The effect of rounding to two decimal places for this matrix shows a difference of between 0.001 and 0.002 in the value of the obtained factor loadings. Because of the negligible differences found in the above data when rounding to tvvo decimal errors accounted for the discrepancies between places, it seemed unlikely that rounding Goldman-Eisler’s (1951) analysis and the present one. It was then decided to perform. by hand.

TI

T-l

T3

TI! (0 ) Be/ore

Rounriwlq

(C/Ii/~/.

T5

/Y 79. p

SY)

TI

I .0000

0.984s

0.3170

0 0000

T?

0 9YJY

I on00

0.5000

0 1736

0 on00

T3

0 3430

0.5000

I 0000

0 9397

O.Sh60

Tl

0.0000

0 1736

0.9397

I .0000

0.YY18

1736

0 0000

0.8660

0 YR-IS

I 0000

sot

? I532

2.658-l

3.6477

3 09s I

? 6172 CT

FL

0.572

0.705

0.970

ox

TI

1.00

0.9s

0 3-l

0.00

T?

0 9s

I .oo

0 50

0 17

0.00

T3

0 34

0.50

I 00

n 94

0 87

T1

0 00

0 I?

0.91

I 00

I) YY

17

0 98

I 00

Tj

-0

(h 1 A//u

T5

-0

Roundr,rq

I

-0.

I736

h -0.17

0.00

0.87

sot

2.15

2.65

3.65

3 OY

2 6X GT

FL

0.570

0.703

0.96X

OS19

0 il0

Rsproducsd I_L

=

w!th fuor

the kind

pcrmisbion

loading:

GT

= ggd

FL = Co.

total

.CT.

oi Ihs ~oul

= 14.2346

0.11 I

publlbhcrs.

SOC

= sum

= II

21

of column:

;I simple-summation method factor snulhsis for the sytraction oi the first factor. on the orl~ln,tl Goldman-Eisler correlation matrix. This was performed following the same principle of dividing Ihe column total b> the square root oi the gand totit of the correlations. B>, doins this it ~\ould be expected that Factor I of this re-analysis and Factor 1 produced in the origins1 Goldman-Eijler c195 I) stud). v,ould correspond to a very high degree. to the order of & O.OOI !1.001. limited bb rounding errors. This. ho\vever. ivaj not the case. Table I0 presents ;I comparison of the factor loadings on Factor I from the original Goldman-Eisler (1951) publication and those of this re-analysis. of the same original data. for the extraction of Factor I. It 15quite noticeable that there are some striking discrepancies bet\veen the magnitudes of these loadings. although the same method of analysis \!‘a used in both cases. For ewmple. the oral scale ,Aloofness ( II saturation of 0.709 on the Goldman-Eisler first factor. Mhereas. this \~triable has ;I saturation of 0.739 on Factor I of the re-analysis (see Table IO). The o\.erall degree of agreement betlvsen these analyses can be summarized in n scattergram of correspondin g factor loadinps I Fis. 7). ‘As can be seen quite clearly from this scatterg-am. Endocatheusis (En) 11~sin nearly the same position in both cxes. with a difference of oni>, 0.006 bettveen the lo:idinzs. .-\utonomg and Dependence (Dz) alw come fairly close in their respective magnitudes. The conclusion reached from this re-analysis. is that the discrepancies betL\een the magnitudes of the factor loadings is not the result of simpl\, roundin p the correlations to two decimal places. As has been sho\vn from the example taken from Child ( 1979) roundin, (7onlv_ had negligible ctt‘ects and rounding would be espected to produce the same sort of negligible edicts in this rz-analvsis. .Vc,yurirc ei,ye~r~ciiws

The production of negative eigenwlues and the tinding that Goldman-Eisler’s ( 195 I ) correlation matrix MX not positive definite (points B and C) are the probable result5 of an Inconsistent pattern of correlations. These inconsistent correlations result In an ill-condition4 correlation matrix [cf. Lubin’s (1950) notes on the re-analysis of Sheldon’s (1942) correlation matris in Lvhich the conclusion was reached that Sheldon’s matrix was ill-conditioned because of an inconsistent pattern of correlations). One of the widely used techniques. in the simple-summation method, to estimate the eigencalue for ;I factor is to sum the squares of factor loading on that factor. Presumably this was the technique employed by Goldman-Eisler in the simple-summation factor analysis performed on her correlation matrix. it is mathematically impossible to arrive at negative eigenvalues using the sum-of-squares technique.

Oral character GGLDMAN

traits and E N

- EISLER’S

FACTGR

1

CORRULA’S FACTOR 1

-02

-

Am . Ch

Fig. 2. Scattergram

showing the loadings on Factor I from the Goldman-Eisler study obtained in the re-analysis of Goldman-Eisler’s correlation matrix.

against

those

When a correlation matrix is not positive definite and a determinant of 0.000 for that matrix is arrived at, there is, in mathematical terms, no inverse of that matrix. Any results that come from an analysis of that matrix would then be misleading. Taken together, both of the above-mentioned mathematical inconsistencies are the result of an ill-conditioned correlation matrix. The ill-conditionability problem arises because of the inconsistent pattern of correlations. The conclusions reached from this part of the re-analysis are: (a) the correlation matrix produced in the Goldman-Eisler (195 I) study had inconsistent, ill-conditioned patterns of correlations; and (b) the production of negative eigenvalues was the result of imaginary factor loadings being used in the calculations for the eigenvalues. Multiple-regression analysis was performed on Goldman-Eisler’s (1951) original correlation matrix. This was performed for the followin g three purposes: (1) to find out what oral scale(s) might be causing the problems in the correlation matrix, as set out above; (2) as a check for the independence of the 19 orality scales; and (3) because squared multiple correlations could not be found during factor analysis. This analysis found that three variables (Optimism, Deliberation and Guilt) had impossible tolerances. The product-moment correlation coefficient can not be greater than +- 1 or less than - I and it follows that r? can not be greater than + 1 or less than - 1. Tolerance of a variable can be defined as the proportion of its variance not accounted for by other independent variables in the regression equation. The minimum tolerance of a variable is the smallest tolerance any variable already in the analysis would have if that variable were included in the analysis. A variable must pass both tolerance and minimum tolerance tests in order to enter into a multiple-regression equation. Tolerance is given using the equation 1 - r’. This means that an impossible tolerance occurs when the equation 1 - r’ results in a negative value. This analysis also has shown that four variables (Impulsion, Deliberation. Optimism dnd Dependence) are

almojt perfectI) predict-d t‘rom a linear combination of the remaining 15 \3ri;lbleS. This result demonstrates that the 19 oral scales are not indepwdsnt \anable:s. It \\:\j aljo found that \\hsn the aboLe-mentioned four \.ariabies avers excluded from a factor xnal>, Goldman-Eisler (195 I) haLe either serious calculation errors or are baxd on non-existent correlations. The present \vrirsr is noi completely sure Lvhich of these possibilities is the correct one. but ivill attempt to discover exactI>. \\hat the problem is and report the resulrs as soon as they are a\~ailable. The results of this w-analysis sugest thal. the rejults of Goldman-Eisler’s ( 1’951) stud>, led her to reach an invalid conclusion. BW;~LW doubt is cast upon the rt’suI1s and the conclusion arri\.ect at by Goldman-Eisler i5 placed into ;I position of suspicion, it is aryed that her study can not be accepted as support for the psychoanalytic concept of oral character or the aetiology of an oral syndrome.