The Nature of Extraversion: A Genetical Analysis

Journal of Personality and Social Psychology1975, Vol. 32, No. 1, 102-112

The Nature of Extraversion: A Genetical Analysis

Lindon EavesDepartment of Genetics,

University of Birmingham,Birmingham, England

Hans EysenckInstitute of Psychiatry,University of London,

London, England

A biometrical-genetical analysis of twin data to elucidate the determinants ofvariation in extraversion and its components, sociability and impulsiveness, revealedthat both genetical and environmental factors contributed to variation in extra-version, to the variation and covariation of its component scales, and to the inter-action between subjects and scales. A large environmental correlation between thescales suggested that environmental factors may predominate in determining theunitary nature of extraversion. The interaction between subjects and scales de-pended more on genetical factors, which suggests that the dual nature of extraver-sion has a strong genetical basis. A model assuming random mating, additive geneaction, and specific environmental effects adequately describes the observed vari-ation and covariation of sociability and impulsiveness. Possible evolutionary impli-cations are discussed.

One of the central problems in personalityresearch has been the question of whether suchhigher order factors as extraversion can beregarded in any meaningful sense as unitary orwhether there are several independent factors,such as "sociability" and "impulsiveness,"which should not be thrown together arti-ficially. Carrigan (1960) concluded her surveyof the literature by saying that "the uni-dimensionality of extraversion/introversion hasnot been conclusively demonstrated" (p. 355);she further pointed out that several jointanalyses of the Guildford and Cattell question-naires show that at least two independentfactors are required to account for the inter-correlations between the extraversion-impul-siveness variables. These two factors, shesuggested, may correspond to the Europeanconception of extraversion, with its emphasis onimpulsiveness and weak superego controls, andthe American conception, with its emphasis onsociability and ease in interpersonal relations.Eysenck and Eysenck (1963) have reportedquite sizable correlations between sociabilityand impulsiveness, a conclusion replicated by

We are indebted to the Colonial Research Fund andthe British Medical Research Council for their supportof the investigation. We are grateful to J. Kasriel forthe collection of data and to the twins for their con-tinued cooperation.

Requests for reprints should be sent to Lindon Eaves,Department of Genetics, University of Birmingham,Birmingham, B15 2TT, England.

Sparrow and Ross (1964); this would suggestthat there is a close connection between thetwo conceptions (Eysenck & Eysenck, 1969).Furthermore, Eysenck and Eysenck (1967)have shown that the correlations of extraversionitems (whether sociability or impulsiveness)with subjects' reactions on a physiological testdevised on theoretical grounds were propor-tional to their loadings on the extraversionfactor. The recognition that extraversion is aunitary factor in behavior is thus vindicatedby prediction from a psychological theory asmuch as by a correlation between primaryfactors (Eysenck, 1967).

We now develop a model for the geneticaland environmental determinants of extraver-sion and of its primary components, sociabilityand impulsiveness. Our intention is to analyzethe phenotypic variation and covariation ofsociability and impulsiveness into their geneti-cal and environmental components in order todetermine, as far as our data permit: (a) thesimplest model for the genetical and environ-mental variation of extraversion considered asa unitary trait and (b) the simplest model forthe genetical and environmental determinationof the interaction between subjects and thecomponent tests of extraversion, sociability andimpulsiveness.

In fulfilling these aims, we are led to com-pare the unitary and dual models of extra-version with regard to their relative contri-

102

THE NATURE OF EXTRAVERSION 103

butions to the representation of both geno-typic and environmental determinants ofvariation among the responses of subjects to apersonality inventory.

Earlier research from the standpoint of thepsychological theory underlying this work hasmainly been concerned with the analysis ofextraversion as a unitary trait (Shields, 1962).Claridge, Canter and Hume (1973) reportedanalyses of extraversion, sociability, and impul-siveness, but these authors themselves admittedthat their samples were too small to justifythe kind of analysis we attempt here. Ourmodel will be derived from an analysis of twindata and will, therefore, inevitably reflect thelimitations of twin studies as sources of geneti-cal information (Jinks & Fulker, 1970). Eventwin studies, however, have seldom been usedto best advantage. We hope that our particularanalysis will have the additional virtue ofdemonstrating how twin data in general maybe manipulated to test simple hypothesesabout the causes of variation. We have adoptedthe methods and notation of biometrical gene-tics (Mather & Jinks, 1971) because webelieve them to be the most precise and general,while embodying a denned procedure for theanalysis of continuous variation which may beextended readily to the analysis of humanbehavior (Jinks and Fulker, 1970).

DATA

The analysis is based on the responses of837 pairs of adult volunteer twins to an 80-itempersonality inventory. Of these items, 13 formeda scale of sociability, and 9 items were scoredto provide a measure of impulsiveness. Therelevant items are given in Table 1. On thebasis of a short questionnaire concerningsimilarity during childhood, the twins wereclassified as monozygotic or dizygotic.1 Such aprocedure is surprisingly reliable (Cederlb'f,Friberg, Jonsonn, & Kaij, 1961). A sample of

1 The twins were asked: (a) "Do you differ markedlyin physical appearance and coloring?" and (b) "Inchildhood were you frequently mistaken by people whoknew you?" If consistent replies were not given,reference was made to previous questionnaires, twins'letters, and additional information in an attempt toassess zygosity. Many of the twins have been blood-typed subsequently, and the original diagnoses havegenerally been confirmed (Kasriel, J., personal communi-cation, December 1974).

TABLE 1

PERSONALITY INVENTORY ITEMS INCLUDEDIN THE ANALYSIS

Item

18. Do you suddenly feel shy when you wantto talk to an attractive stranger?

23. Generally, do you prefer reading to meet-ing people?

27. Do you like going out a lot?30. Do you prefer to have few but special

friends?36. Can you usually let yourself go and enjoy

yourself a lot at a gay party?40. Do other people think of you as being

very lively?44. Are you mostly quiet when you are with

other people?48. If there is something you want to know

about, would you rather look it up in abook than talk to someone about it?

56. Do you hate being with a crowd who playjokes on one another?

66. Do you like talking to people so muchthat you never miss a chance of talking toa stranger?

69. Would you be unhappy if you could notsee lots of people most of the time?

75. Do you find it hard to really enjoy your-self at a lively party?

77. Can you easily get some life into a ratherdull party?

1. Do you often long for excitement?4. Are you usually carefree?8. Do you stop and think things over before

doing anything?12. Do you generally say things quickly with-

out stopping to think?16. Would you do almost anything for a dare?20. Do you often do things on the spur of the

moment?33. When people shout at you, do you shout

back?59. Do you like doing things in which you

have to act quickly?62. Are you slow and unhurried in the way

you move?

Key

-S

-S+S

-S

+S

+s-S

-S

+s+s-S

+s+1+1-I+1+1+1+1+1-I

Note. S denotes an item scored for sociability, I for impulsive-ness; + indicates that "yes" scored 1, and — indicates that"no" scored 1 for scale under consideration.

digzygotic twins of unlike sex has been in-cluded in our study because these provide acritical diagnostic test of sex limitation. Thecomposition of the sample by sex and zygosityis given in Table 2.

The mean sociability and impulsivenessscores of the five groups are given in Table 3.An analysis of the variation between and withingroups revealed highly significant (but sub-stantively fairly small) differences betweengroups with respect to the sociability scores.The groups did not differ with respect to theirmean impulsiveness scores. We shall regard

104 LINDON EAVES AND HANS EYSENCK

TABLE 2

STRUCTURE OP TWIN SAMPLE

Twin type

Monozygotic femaleMonozygotic maleDizygotic femaleDizygotic maleUnlike-sex dizygotic

No. pairs

33112019859

129

the groups as representative of the same popu-lation as far as their means are concerned.The groups are homogeneous with respect totheir dispersion, as will become clear from thesubsequent genetical analysis. The pooledstandard deviations within groups were 3.0015and 1.7586 for sociability and impulsivenses,respectively.

Since we wished to minimize the possibilityof spurious interaction between subjects andtests, we standardized the raw scores of thetwins on both sociability and impulsiveness bydividing the scores by the correspondingaverage within-groups standard errors. Foreach group of twins separately, the meansquares within pairs and between pairs werecalculated for each of the standardized scales.The analogous within-pairs and between-pairsmean products were also calculated. Themean squares and mean products form thebasic statistical summary for the analysis tofollow (see Table 4).

We studied the inheritance of extraversionby analyzing the mean squares derived fromthe subjects' total scores of the two standard-ized tests. The mean squares for the twins onthe measure of extraversion (E) may be

TABLE 3

MEAN SOCIABILITY AND IMPULSIVENESSSCORES or TWIN GROUPS

T ' NM

Sociability

MZf 662 6.5045MZm 240 5.7875DZf 396 6.6869DZm 118 6.6441DZos 258 6.4884

Impulsiveness

3.70393.81253.75254.06783.7054

TABLE 4

MEAN SQUARES AND MEAN PRODUCTS WITHIN ANDBETWEEN TWIN PAIRS FOR STANDARDIZED

SOCIABILITY AND IMPULSIVENESS SCORES

Item df MS(S)

MS(Imp)

MP(S-I)

Between MZf pairs 330 1.5339 1.3777 .6517Within MZf pairs 331 .5394 .6403 .1762

Between MZm pairsWithin MZm pairs

Between DZf pairsWithin DZf pairs

Between DZm pairsWithin DZm pairs

Between DZos pairsWithin DZos pairs

119120

197198

5859

128129

1.5595.4817

1.0855.8380

1.3919.6693

.9457

.9290

1.2904.6630

1.1804.8408

.9799

.8441

1.2839.7697

.3126

.1497

.3069

.2918

.6309

.1516

.3638

.3581

Note. Abbreviations are as follows: MZf = monozygotic fe-male, MZm = monozygotic males, DZf = dizygotic female,DZm = dizygotic male, and DZos = unlike-sex dizygotic.MS = mean square, MP = mean product; S = Sociability andI = impulsiveness. No correction for the main effect of sex wasnecessary for the DZos.

derived directly from the mean squares andmean products (MP) of Table 4, since MSE

= MS(S+i) = MSS + MSi + 2MPB,i, whereS and I refer to sociability and impulsiveness.

Just as we obtained an E score for eachsubject by summing over tests, so we mayobtain a difference (D) score for each subjectby taking the difference between his scores onthe standardized tests. The MS derived fromthese differences summarizes the variationarising because subjects do not perform con-sistently on the two tests. We may obtain theMS for the D scores directly from the raw MSand mean products of Table 4, since

i - 2MP

Note. Abbreviations are as follows: MZf = monozygotic fe-male, MZm = monozygotic male, DZf = dizygotic female.DZm = dizygotic male, and DZos = unlike-sex dizygotic.

(a,i->.

The mean squares for E and D are found inTable 5. Clearly, since the mean products areall positive, the MS®'s are larger than thecorresponding MSo's. Since we are only con-cerned with these particular tests, the meansquares between subjects for E contain none ofthe interaction variation. Thus the fact thatthe MSv's are approximately twice as largeas the MSo's is an indication that E accountsfor more of the total variation of the twotests than D.

We analyze the MSv to provide a geneticalmodel for variation in extraversion, and weanalyze the MSV to determine the extent to

THE NATURE OF EXTRAVEESION 105

which genetical or environmental factors con-tribute to the resolution of E into sociabilityand impulsiveness. Finally, we show that thecovariation of sociability and impulsivenessreflects both genetical and environmental fac-tors by an analysis of the raw mean squares andmean products of Table 5.

METHODS

Formulation of the Model

Most analyses of classical twin studies have merelydemonstrated the existence of a genetical component ofvariation by showing that monozygotic twins are morealike than are dizygotic twins. Such an intuitive ap-proach is imprecise and does not lead to any exact pre-dictions about the similarity between other degrees ofrelatives. For this reason the classical approach is notvery helpful in guiding the design of future research. Inadopting the methods of biometrical genetics we areable to specify, for a given set of assumptions about thekinds of gene action and environmental effects, preciseexpectations for the components of variance (and con-sequently for the mean squares) derived from the analy-sis of variance of any group of relatives. Furthermore,having specified our assumptions and the consequentexpectations of mean squares, we are able to provide astatistical test of the agreement between observationsand expectations and, consequently, to test the validityof the assumptions we made at the outset.

Clearly, twin data of the kind we have summarizedin this article do not allow us to test any but the simplestset of assumptions about the causes of variation. Wealso recognize that failure of particular assumptions inprinciple may not lead to failure of the model in practice,either because sample sizes are too small for the test tobe sufficiently powerful (see, e.g., Eaves & Jinks, 1972)or because failure of certain assumptions may contri-bute more to a bias in estimation than to departures ofwhat is observed from what is expected.

We now consider individually the assumptions wemake in the analyses which follow. Some of these are anundesirable necessity of the limited data we have avail-able and may not be tested very powerfully by ouranalysis. Other assumptions are quite likely to be dis-proved in practice, even with the data available, if theyare unjustified. We emphasize that these limitationsapply not to the method, which is the most explicit andflexible available, but to the particular data upon whichwe seek to build our model. Other experimental designswould enable us to test with greater conviction theassumptions which we now make tentatively, (Eaves,1972; Jinks & Fulker, 1970).

1: Alleles and loci act additiwly and independently. Weassume that there is no dominance, epistasis, or linkagefor the loci contributing to variation of the traits underconsideration. If our other assumptions are justified,nonadditive variation may be detected in principlewith our data, but in practice the necessary sample sizesare likely to be too large (Eaves, 1972). Failure of eitherof the two following assumptions may make the de-

TABLE S

MEAN SQUARES FOR EXTRAVERSION AND INTERACTIONOF SUBJECTS AND COMPONENT TESTS

Item df

MS

E = S +1

Between MZf pairs 330 4.2150Within MZf pairs 331 1.5321

D =S - I

1.6082.8273





119120

197198

5859

128129

3.47511.4441

2.87972.2624

3.63361.8166

2.95722.4749

2.2247.8453

1.65211.0952

1.11001.2102

1.5020.9825

Note. Abbreviations arc as follows: MZf = monozygotic fe-male, MZm = monogyzotic male, DZf = dizygotic female,DZm = dizygotic male, and DZos = unlike-sex dizygotic. Erefers to extraversion, and D refers to the interaction of sub-jects and component tests. S and I refer to sociability and im-pulsiveness, respectively.

tection of nonadditive variation virtually impossiblewith the data of our study.

2: Mating is random. There is little evidence of assort-ative mating for extraversion. We might expect to detectthe genetical consequences of assortative matingprovided there is substantial genetical variation and afairly high correlation between spouses (Eaves, 1973a).The design of the present study, however, since itinvolves twins reared together, makes it impossible todistinguish the effects of assortative mating from thoseof environmental influences shared by members of thesame twin pair. Eysenck (1974) reports a nonsignificantcorrelation for the extraversion scores of husbands andwives. This suggests that assortative mating can safelybe discounted as a factor contributing to the geneticalvariability of extraversion.

3: All environmental effects we specific to individualswithin families. Most of the twins in our study livedtogether, especially when they were young. The factthat both individuals in a pair have the same biologicalmother and grew up in the same family may make bothmonogyzotic and dizygotic twins more alike than wewould expect on the basis of our simple genetical andenvironmental model. For twins reared together, sucheffects are formally indistinguishable from those ofassortative mating and, if they are substantial, maycontribute to failure of our simple model, which assumesthat assortative mating and common environmentalinfluences make a negligible contribution to the observedvariation. We have some independent test of the con-tribution of postnatal shared environmental influencesfor extraversion because we would expect the intrapairdifferences of twins to increase with the period of sepa-ration. Although age and duration of separation are, ofcourse, highly correlated, we can detect no relationshipbetween the intrapair difference for extraversion scores


TABLE 6

A GENETICAL AND ENVIRONMENTAL MODEL FOR ASET or OBSERVED MEAN SQUARES

MS

Between MZf pairsWithin MZf pairs





Coefficient of parameter

DK

1

1

f4

34

f4

Ei

1111111111

Note, Abbreviations are as follows: MZf = monozygotic fe-male, MZm = monozygotic male, DZf = dizygotic female,DZm = dizygotic male, and DZos = unlikc-sex dizygotic.DR refers to an additive genetical component, and Ei refers toa within-family environmental component.

and the age or duration of adult and juvenile separationof these twins. We hope to make a detailed considerationof this issue the subject of a future publication. Weassume, until we have evidence to the contrary, thatany environmental variation for the traits in questionis the result of influences which are unique to particularindividuals rather than shared by members of the samefamily. If such an assumption were clearly unjustified,then we would find our observations quite obviouslydid not coincide with our expectations and we would beforced to reject our simple model.

Jinks and Fulker (1970) showed in their biometricalgenetical reanalysis of Shields's (1962) extraversiondata that common environmental influences must befairly unimportant. This adds some weight to ourassumption that common environmental influences canbe ignored, but we should first indicate the likely sensi-tivity of our experiment for detecting such effects ifthey still contribute to the variation between pairs.

Power calculations, familiar in the context of bio-metrical genetics, (e.g., Eaves, 1972; Eaves & Jinks,1972; Kearsey, 1970) reveal that a sample of approxi-mately 220 pairs of monozygotic and the same numberof dizygotic pairs would allow us to be roughly 50%certain of detecting common environmental influenceswhich accounted for 20% of the total variation againstthe background of an additive genetical componentwhich accounted for about 40% of the total variation.This gives some indication of the power of the presentstudy to detect common environmental influences (andthe confounded consequences of assortative mating),provided there is no nonadditive genetical variation.To be 95% certain of detecting a common environ-mental component of this magnitude, sample sizeswould have to be about four times as great.

4. Sex linkage and sex limitation are absent. If we wereto adopt the usual practice of analyzing correlations

rather than mean squares, we would have only poortests of sex linkage and sex limitation. We would, how-ever, expect the numerical values of comparable meansquares to vary significantly between sexes in the pres-ence of sex linkage or sex limitation except under veryrestrictive assumptions about the magnitudes and typesof gene effects (see Mather & Jinks, 1971). In applyingour model to the mean squares rather than to the corre-lations, we may expect any gross distortion due toeither of these causes to result in significant failure ofthe model.

5. Genotypic and environmental deviations are un-correlaled. Under some circumstances we might expectgenotype-environmental covariation to contribute tofailure of our simple genetical model for monozygoticand dizygotic twins. These circumstances, however, arerather restricting for a study of this type, and we shouldbe cautious about assuming that the adequacy of themodel for twin data means that we can ignore thissource of variation. As a consequence of the covariationof genotype and environmental effects we could, inprinciple, find that the environmental components fordizygotic and monozygotic twins are no longer com-parable. In practice we are unlikely to detect suchdifferences with these data.

6. Any variation due to the interaction of genotype andenvironment is confounded with the enmronmental vari-ation within families. It is inevitable that any inter-action between genotypic effects and those environ-mental influences specific to individuals will be con-founded with variation due to specific environmentalfactors in human studies (Jinks & Fulker 1970). Wehope that an analysis of such interactions for personalityvariables will be the subject of a future article. If weare justified in our assumption that there are no commonenvironmental influences, then we are also justified inour assumption that these do not interact with thegenotype. Should our model fail because of commonenvironments, we would find that our estimate of theadditive genetical component was biased by the vari-ation due to any interaction of these influences withgenotypic factors.

We represent the six assumptions by writing a modelfor the mean squares for pairs of monozygotic anddizygotic twins in terms of an additive geneticalcomponent (DR) and a within-family environmentalcomponent (Ei). Mather and Jinks (1971) showed howDK may be defined in terms of the frequenices andeffects of many loci. The coefficient chosen for DE willdepend on the mating system. Since we are assumingmating to be random, the coefficients involve no furtherunknown parameter and may be written for monozy-gotic and dizygotic twins as they are shown in Table 6.The expectation for a within-fair mean square is simplythe expectation for the corresponding within-familycomponent of variance, trw^, and the expectation for abetween-pair mean square is aw

z plus twice the expecta-tion for the corresponding between-families component ofvariance, <r$. Full tables of expectations of mean squaresand variance components for different kinds of relativesmay be found elsewhere (e.g., Eaves, 1973.a; Jinks &Fulker, 1970; Mather & Jinks, 1971).

THE NATURE OF EXTRAVEESION 107

Estimating Parameters and Testing the Model

If we consider only one trait we have, in this study,10 observed mean squares. Let these be written as thecolumn vector x. Our model (see Table 6) involves twoparameters whose coefficients in the expectations of xmay be represented by the 10 X 2 design matrix A. Wemay obtain our estimates of the J:wo parameters,denoted by the two-element vector 6, by solving thesimultaneous equations:

8 = (A'WA)-iA'Wx,

where W is the (10 X 10) matrix of information aboutthe observed statistics. When the x are mean squares,the amount of information about mean square Xi is»i/2(ex,-)2, where m is the degrees of freedom corre-sponding to Xi. Clearly we do not know ex until themodel has been fitted so we have to use the observed xto provide trial values for the amounts of informationand proceed iteratively until our ex are stable. Inpractice, however, it is often unnecessary to go beyondthe first cycle provided the model is adequate, since xwill then be a close approximation to ex. For the case inwhich we are considering a single trait and our mean-squares are all independent, W is diagonal and the com-putations for simple models are not tedious. Providingthat our observed statistics are normally distributed,our estimates of 6 are the maximum-likelihood esti-mates and the scalar

S = (x - ex)'W(x -ex)

is distributed as a chi-square with degrees of freedomequal to the number of statistics less the number ofparameters estimated from the data. The assumptionof normality is probably not far from the truth with thesample sizes available. Should this chi-square besignificant, we would be compelled to reject our modelas inappropriate for the description of the variation forthe trait under consideration.

Although the preceding statistical considerations arenot new, they have not been generally applied to thegenetical analysis of human behavior, with the resultthat data have been used inefficiently, standard errorsof estimates rarely quoted, and assumptions rarelytested. The usual analyses of twin data either concen-trate on the variation within pairs or on a comparisonof monozygotic and dizygotic correlations. The methodwe employ combines both approaches in a single testof a simple model. In effect, our test of the DR, EI modelis not merely testing whether the within-pair variancesdiffer for the two types of twins but whether theestimates derived from within-pair comparisons can beused to predict the variation between pairs. We expectthe prediction to be poor if certain of our assumptionsfail. These and other considerations, such as that ofsex limitation, are all combined in our weighted least-squares analysis of the full set of raw mean squares.

RESULTS

Genetical A nalysis of Exit'aversion

The estimates of the parameters and theelements of their covariance matrix, (A'WA)~l,

TABLE 7

GENETICAL ANALYSIS OF EXTRAVERSION

Parameter Estimate X2 df P

DR 2.2487 98.50 1 <.001Ei 1.5280 276.04 1 <.001Residual 7.05 8 .50VDn .051336 — —VEi .008458 — —Cov DREi —.011814 — —

Note. DR refers to an additive genetical component, and Eirefers to a within-family environmental component. V and Covrefer to variance and covariance, respectively.

are given for extraversion in Table 7. We seefrom the nonsignificant residual X2 that ourmodel is clearly adequate so that the data giveno reason to suppose that our assumptions areunjustified. We divide the square of eachestimate by the corresponding variance termto give, for each estimate, a X2(l) which teststhe significance of that parameter. Clearlyboth DR and EI are highly significant com-ponents of the variation in extraversion.

On the basis of our tests of the model wetentatively adopt the view that most of thegenetical variation is additive and most of theenvironmental variation can be attributed toEI. We may use our estimates to estimate theproportion of the population variance forextraversion which can be attributed to genet-ical causes. Since all of the variation is additive,we have no need of the distinction between"broad" and "narrow" heritability in thepresent context ; we just estimate :

= .424 for extraversion.

This means that 42% of the variation in extra-version may be attributed to genetical causes.In the present case EI includes variation dueboth to "unreliability" and "real" specificenvironmental influences. There seems littlepoint in correcting for unreliability if all pre-dictions are to be made on the basis of oneadministration of a test such as that analyzedhere. If our genetical model is in fact appro-priate, we may predict the correlations betweenother degrees of relatives for extraversion asmeasured by this test. For parents and off-spring, for example, we would expect a corre-lation of ^h2 = .21. Such data as we havesuggest that the observed correlation is


TABLE 8GENETICAL ANALYSIS OF INTERACTION BETWEEN SUB-

JECTS AND TESTS OF SOCIABILITYAND IMPULSIVENESS

Parameter

DREIResidualVD RVEiCov DftEi

Estimate

.8591

.8359—

.011727

.002418-.003277

x°

62.94288.95

8.76———

df

118

———

p

<.001<.001>.30

——

—

Note. DR refers to an additive genetical component, and Eirefers to a within-family environmental component. V and Covrefer to variance and covariance, respectively.

somewhat lower but not significantly so. Sucha difference, if it turned out to be significant,might be attributed to the interaction of thegenotypic difference between individuals withan overall differences between the environ-ments of parents and offspring or to the factthat our estimate of the heritability is some-what biased by undetected common environ-mental effects. A common environmentaleffect which accounted for about 10%-15% ofthe total variance might explain the disparityand is more likely than not be to undetectedin our study.

We obtained estimates of the internal con-sistency of the scales. For sociability thereliability was about .75 and for impulsiveness,.60. We may correct our heritability estimatefor unreliability provided we can assume theSubjects X Items interactions estimate experi-mental error only. Using the estimates ofgenetical and environmental variance andcovariance obtained below, we found theheritability of extraversion, after correction,to be .57. By correcting for unreliability, wehave attempted to partition the environmentalvariation for extraversion into that part whichmay reflect stable environmental influences onthe development of the trait and that part dueto experimental error. If subjects and itemsinteract, the contribution of experimental errorto EI will be overestimated. Such interactionsmay have a genetical component which wecould analyze using the methods adopted inthis article. Confounded with our "true" en-vironmental variation will remain variationreflecting day-to-day changes in behaviorwhose contribution can only be assessed byrepeated measurement.

Genetical Analysis of Subject X Tests Interaction

The results of the analysis of the meansquares for the D scores appear in Table 8.Broadly speaking the results for the inter-action are very similar to those for E. The maindifference is the reduction by half, in this case,of the estimates of DR and EI. This reflects thegreater discriminating power of E resultingfrom the positive covariation of sociabilityand impulsiveness. However, the simple modelis again adequate, since the residual X2(8) isnot significant. DR and EI are, once more,highly significant. This means that the dis-crimination between sociability and impulsive-ness is justified in genetical terms. We have toconclude that not all the genetical factorscontributing to variation in sociability andimpulsiveness contribute equally and consis-tently to both. We estimate the heritability ofthe interaction to be .339. Although this valueis somewhat lower than that for E, the differ-ence is not large and we must notice that therelative contribution of unreliability variationwill be greater for the interaction than for E.Using the reliabilities given above, and theestimates of the genetical and environmentalvariance and covariance components from alater analysis (see below), we estimate theheritability of the interaction of subjects andtests to be .72. The marked change reflects therelatively large positive environmental corre-lation between sociability and impulsiveness,particularly when the environmental vari-ances are corrected for unreliability.

Table 9 summarizes the results of bothanalyses in terms of the proportions of thetotal variation of sociability and impulsiveness

TABLE 9THE RELATIVE CONTRIBUTIONS OF GENOTYPIC AND

ENVIRONMENTAL FACTORS OF EXTRAVERSION ANDSUBJECTS X TESTS INTERACTION TO THE VARI-

ATION BETWEEN SUBJECTS FOR SOCIABILITYAND IMPULSIVENESS

Causal factor

Psychological factor

Extra-version

Inter-action

Total

Genetical .2870(.3402) .1096(.2631) .3966(.6035)Environmental .390<)(.2156) .2134(.1807) .6034(.3963)

Total .6770(.5558) .3230(.4440) 1.0000(.9998)

Note. Proportions of estimated reliable variation are given inparenthesis.


TABLE 10RESULT OF FITTING SIMPLE MODEL TO VARIATION AND COVARIATION

OP SOCIABILITY AND IMPULSIVENESS

p F f 'm t

Covariance ( X 104) of estimate with estimate of

DHS DRI

DRS .9214 67.05 8.54DBI .7132 68.38DBSI .3419E1B 15410En .6441Eisi .1758

DRS I

23.8924.0138.08

Eis

-14.80- 1.62- 4.92

10.42

En

-1.61- 19.34- 5.64

1.1114.32

Eisi

o / i \

-4.92 126.63*-1.93 74.33*-9.30 30.70*

3.41 280.77*9.92 290.45*6.68 46.27*

Note. The parameters DRS, DRI, DRSI, Eis, En, and Eiai correspond to the components of the mean squares of sociability, im-pulsiveness, and the mean products of the two traits, respectively.

* P < .001.

scores which may be attributed to the geneticaland environmental components of extraversionand the interaction of subjects and tests.Approximately three fifths of the total vari-ation is environmental (from the row totalsof Table 9) and two thirds of the total variationis attributable to the extraversion factor (fromthe column totals of Table 9). The proportionof environmental variation is fairly consistentover columns, and the proportion of variationaccounted for by extraversion is fairly con-sistent over rows. In Table 9 we also presenta summary for the scales after the environ-mental components have been corrected forunreliability. So far we have shown that geneti-cal factors probably contribute to individualdifferences in both E and D scores. A quali-tative consideration of the conclusion suggeststhat E is more discriminating than D geneti-cally and environmentally and leads us to theview that the positive covariation of sociabilityand impulsiveness has a basis which is bothgenetical and environmental. We could verifythis directly by a statistical comparison of ourestimates to test whether the estimates of DRand EI are significantly greater than the corre-sponding estimates for D. We prefer, however,to estimate separately the genetical and en-vironmental components of the variation andcovariation of sociability and impulsivenessscores, since this will allow us to estimate thegenetical and environmental correlations be-tween the traits.

Analysis of Variation and Covariation ofSociability and Impulsiveness

The weighted least-squares procedure de-scribed above may be extended without undue

complication to the simultaneous analysis ofthe variances and covariances of multiplevariables (Eaves & Gale, 1974). In this case,however, separate DR'S and EI'S are fitted forthe variance and covariance terms. The in-formation matrix, W, is no longer diagonal,since the model is fitted to mean squares andmean products which are no longer indepen-dent because each subject yields measurementson every trait.

In this instance the model is fitted to the30 statistics of Table 5. Now six parametersare specified, DES, DRI, DRS,i, EIS, En, andEIS,I. These correspond to the components ofthe mean squares of sociability, impulsiveness,and the mean products of the two traits, re-spectively. The method and the definition ofthe parameters is discussed in more detail byEaves and Gale (1974).

There are thus 30 statistics. Six parametersare estimated from the data, so the residualchi-square for testing the goodness of fit ofthe model has 24 degrees of freedom. Thischi-square changed by less than .2% betweenthe first and second cycle of the weightedleast-squares analysis, when X2(24) = 29.41,/>~.20. Thus the adequacy of the DR, EImodel for the variation and covariation ofsociability and impulsiveness was confirmed.

The estimates of the six parameters and theircovariance matrix are given in Table 10.Clearly all the estimates differ significantlyfrom zero. We estimate the heritability ofsociability to be .460 and that of impulsivenessto be .356. Using, once more, our estimates ofreliability, we infer that about 54% of theenvironmental variation for sociability is"reliable" variation, assuming that we have


accounted for all the unreliability. For impul-siveness the comparable figure is 38%. We maynow obtain estimates of the proportion ofreliable variance which is due to geneticalcauses. Our estimates are .61 for sociabilityand .60 for impulsiveness. There is, therefore,convincing evidence that both traits are undersome degree of genetical control. This findingis not new. Claridge, Canter, and Hume (1973)reported an apparent genetical component ofvariation for both scales. Our model-fittingapproach, however, leads us to suggest thatthere is no evidence of nonadditive geneticalvariation and no evidence of common en-vironmental effects. Furthermore, we havedemonstrated that the covariance of sociabilityand impulsiveness probably has a geneticalbasis but that environmental factors also con-tribute significantly to the covariation of thetwo scales. The extent to which the two traitsmay be regarded as sharing common geneticaland environmental factors is represented by thegenetic and environmental correlations »-DB andrEl, respectively:

= .42

= .32.

Variation due to unreliability contributes toEI but not to DR so we might expect theobserved environmental correlation to be lessthan the genetic correlation. These correlationsare a little less, though not considerably less,than the phenotypic correlation of .468reported for sociability and impulsiveness byEysenck and Eysenck (1969).

Providing we are justified in assuming theunreliability components of sociability andimpulsiveness to be uncorrelated, we may cor-rect our estimate of r^ for unreliability usingthe estimates of reliability given above. We nowfind that r^ is .66. This indicates that the uni-tary nature of extraversion is clearly evidentin the environmental determinants of the trait,even though the genetical correlation betweensociability and impulsiveness is rather less.Eaves (1973b) suggested, on the basis of amultivariate genetical analysis of monozygotictwins, that "the apparently unitary nature ofextraversion at the phenotypic level could bedue to environmental rather than to genetical

influences." The different analysis we havepresented here confirms this conclusion.

We should peihaps clarify what this findingmeans. It does not necessarily support theview that extraversion is an "environmentalmold" trait, to use the conception of Cattell,that is, a trait which reflects the structure ofenvironmental influences inherent in the en-vironment itself. We may obtain exactly thesame picture because the organism, by virtueof the integration of its nervous system,imposes a unitary structure on externally un-structured environmental influences contribut-ing to the development of behavior.

CONCLUSIONS

The analyses presented above suggest thefollowing principal conclusions.

1. Genetical factors contribute both to thevariation and covariation of sociability andimpulsiveness.

2. Environmental factors also contribute tothe covariation of sociability and impulsiveness.

3. The genetical correlation between thetwo factors is estimated to be .42, the environ-mental correlation to be .66 after correctionfor unreliability.

4. Combining sociability and impulsivenessscores by addition to provide a measure ofextraversion provides the most powerful singlemeans of discriminating between individualswith respect to the genetical and environmentaldeterminants of their responses to the soci-ability and impulsiveness items of thequestionnaire.

5. The interaction between subjects andtests has a significant genetical component, sothere is some justification for regarding soci-ability and impulsiveness as distinguishablegenetically.

Furthermore we conclude:

6. About 40% of the variation in sociability,impulsiveness, and their combinations, asmeasured by this questionnaire, can be at-tributed to genetical factors.

7. Our data are consistent with the view thatthe genetical variation is mainly additive.

8. We find no evidence for a large effect ofthe family environment on any of the traitsstudied, but the environmental influences


specific to individuals (including unreliabilityof measurement) account for about 60% of thevariation.

9. Mating is effectively random for thetraits in question.

10. The genetical and environmental de-terminants of variation are homogeneous oversexes, suggesting that the effects of sexlinkage and sex limitation are negligible.

DISCUSSION

Our analysis is necessarily tentative be-cause it is based only on monozygotic anddizygotic twins. We would be particularlycautious about discounting genotype-environ-ment correlations as an additional source ofvariation. It must also be emphasized thatgenotype-environment interaction may wellbe confounded with EI so that variation whichwe have ascribed to environmental factorsmay itself have a genetical component. Wehope to clarify this matter in the future.

Between 30% and 40% of the variationin components of extraversion may be due toenvironmental factors that cannot be at-tributed to the inconsistency of the test. Allof the detectable environmental variation isspecific to individuals rather than common tofamilies. This suggests that attempts to relateextraversion to aspects of the individual's"family background" are unlikely to be pro-ductive unless the family background has adirect genetical association with extraversion.Even though we might be able to measuresocial and domestic factors shared by membersof the same sibship, we would not expect theseto be very highly correlated (say, not morethan r = .2) with the mean extraversionscore of the sibship. Consistently larger corre-lations between such shared environmentalfactors and the mean raw E scores of sibshipwould lead us to suspect our simple model.

Since a considerable proportion of the vari-ation in extraversion and its components isclearly due to environmental influences specificto individuals, we could expect, in principle, torelate the intrapair differences of monozygotictwins to differences in their environmentalexperiences. That this is feasible in principle,however, does not aid our efforts to specifyor detect such likely influences. Attempts topredict the variation in extraversion for a

random sample of individuals by measuringconcomitant social or other variables, however,may be misleading, because any association wefind could reflect either genetical or environ-mental communality of the traits in question.Merely attaching the label "social" to a traitdoes not constitute a prior case for environ-mental causation. Analyses of the kind we haveconducted for the covariation of componentsof extraversion would have to be employed forthe other variables if we were to discriminatebetween environmental and genetical associ-ation between extraversion and other vari-ables in the social "environment."

The fact that between 60% and 70% of the"reliable" variation is genetically determineddoes not, of course, suggest which genes are in-volved nor what may or may not be done tomodify the trait. It does, however, suggest thatthe segregation and recombination of allelesmay be a primary cause of variation in the di-mension of personality we have studied here. Atthe level of population biology it means thatextraversion, like most other traits, reflectsgenetical polymorphism and as such is exposedto the directional, stabilizing or disruptiveinfluence of natural selection. As far as we canjudge from studies on other organisms (Kearsey& Kojima, 1967; Mather, 1966; 1967), wefind that directional selection has been charac-teristically associated with the evolution of agenetical system demonstrating a large amountof directional dominance and duplicate geneinteractions. When natural selection hasfavored intermediate phenotypes, the geneticalsystem involves predominately additive effects,and dominance, if any, is ambidirectional. Itwould be too early to say whether our failure todetect nonadditive variation merely reflectsthe design of our study, the (relatively) lowheritability, or the small amount of dominancerelative to the additive variation. It may bedifficult to obtain a definite answer to suchquestions for this trait because of the largesamples required and because of the formalinability to disentangle completely the ad-ditive, dominance, and epistatic components ofgene action for natural populations even whenthese can be raised in strictly experimentalsituations (Mather, 1974). We suggest verytentatively that the polymorphism we detectfor extraversion and its components may be


subject to stabilizing selection because, as faras we can tell at the moment, the geneticalvariation is additive. That is, we should con-clude that neither extreme introversion norextraversion has been favored systematicallyduring human evolution. It is possible thatwithout either extreme the fitness of a humanpopulation would have suffered at sometimeor other. We can at least conceive of situationsin which individuals of more impulsive orsociable temperament may well have promotedthe survival of themselves and their closerelatives. Similarly, we can imagine that thereare times or situations in which it would beadvantageous to have the persistent, atten-tive behavior characteristic of introverts. Incontrast to this, we may consider a trait such ashigh intelligence for which there is a suggestionof directional dominance (Jinks & Eaves, 1974;Jinks & Fulker, 1970) and for which, therefore,we suspect a history of directional selection. Inthe case of intelligence, it is difficult to con-ceive of as many plausible situations in whichrelatively high intelligence could not conferupon an individual greater reproductivefitness than average. Such speculations aboutthe evolutionary significance of personality,are less well founded at this stage because wecannot infer the genetical system with anygreat degree of confidence. We believe, how-ever, that such speculations are legitimate ifthey engender a more systematic and thought-ful approach to the collection and analysis ofdata on human behavioral traits.

REFERENCES

Carrigan, P. M. Extraversion-introversion as a dimen-sion of personality: A reappraisal. PsychologicalBulletin, 1960, 57, 329-360.

Caltell, R. B. The description and measurement ofpersonality. New York: World Book, 1946.

Cederlof, R., Friberg, L., Jonsonn, E., & Kaij, L.Studies in similarity of diagnosis in twins with theaid of mailed questionnaires. Ada Genelica (Basel),1961, 11, 338-362.

Claridge, G., Canter, S., & Hume, W. I. Personalitydifferences and biological variations: A study of twins.New York: Pergamon Press, 1973.

Eaves, L. J. Computer simulation of sample size andexperimental design in human psychogenetics.Psychological Bulletin, 1972, 77, 144-152.

Eaves, L. J. Assortative mating and intelligence: Ananalysis of pedigree data. Heredity, 1973, 30, 199-210.(a)

Eaves, L. J. The structure of genotypic and environ-mental covariation of personality measurements: Ananalysis of the PEN. British Journal of Social andClinical Psychology, 1973, 12, 275-282. (b)

Eaves, L. J., & Gale, J. S. A method for analysing thegenetical basis of covariation. Behavior Genetics, \ 974,4, 253-267.

Eaves, L. J., & Jinks, J. L. Insignificance of evidencefor differences in heritability of I.Q. between racesand social classes. Nature, 1972, 240, 84-88.

Eysenck, H. J. The biological basis of personality.Springfield, 111.: Charles C Thomas, 1967.

Eysenck, H. J. Personality, premarital sexual per-missiveness and assortative mating. Journal of SexResearch, 1974, 10, 47-51.

Eysenck, H. J., & Eysenck, S. B. G. On the unitarynature of extraversion. Ada Psychologica, 1967,26, 383-390.

Eysenck, H. J., & Eysenck, S. B. G. Personalitystructure and measurement. London: Routledge &Kegan Paul, 1969.

Eysenck, S. B. G., & Eysenck, H. J. On the dual natureof extraversion. British Journal of Social and ClinicalPsychology. 1963, 2, 46-55.

Jinks, J. L., & Eaves, L. J. I.Q. and inequality. Nature,1974, 248, 287-289.

Jinks, J. L., & Fulker, D. W. A comparison of the bio-metrical genetical, MAVA and classical approachesto the analysis of human behavior. PsychologicalBulletin, 1970, 73, 311-349.

Kearsey, M. J. Experimental sizes for detecting domi-nance variation. Heredity, 1970, 25, 529-542.

Kearsey, M. J., & Kojima, K. The genetic architectureof body weight and egg hatchability in Drosophilamelangaster. Genetics, 1967, 56, 23-37.

Mather, K. Variability and selection. Proceedings of theRoyal Society of London, Series B, 1966,164, 328-340.

Mather, K. Complementary and duplicate gene inter-actions in biometrical genetics. Heredity, 1967, 22,97-103.

Mather, K. Non-allelic interaction in continuous vari-ation of randomly breeding populations. Heredity,1974, 32, 414-419.

Mather, K., & Jinks, J. L. Biometrical Genetics. London:Chapman Hall, 1971.

Shields, J. Monozygotic twins. Oxford: University Press,1962.

Sparrow, N. H., & Ross, J. The dual nature of extra-version: A replication. Australian Journal of Psy-chology, 1964, 16, 214-218.

(Received March 11, 1974)

Documents

The Nature of Extraversion: A Genetical Analysis