Genes, Culture, and Personality : an Empirical Approach 9781483288468, 1483288463, 0122282906

Table of contents :
Content: Cover image
Title page
Table of Contents
Copyright
Authors
Preface
Chapter 1: Another Book on Heritability?
1.1 PERSONALITY AND SOCIAL ATTITUDES
1.2 AWAY FROM THE "HERITABILITY HANG-UP"
1.3 PRESENTATION OF ORIGINAL DATA
1.4 QUANTITATIVE MODELING AND HYPOTHESIS TESTING
1.5 SUSTAINED STUDY OF SELECTED MEASURES
Chapter 2: Dimensions of Personality
Publisher Summary
2.1 THEORIES OF PERSONALITY
2.2 THE DESCRIPTION OF PERSONALITY
2.3 TESTING THE MODEL FOR PERSONALITY
2.4 CRITICISMS OF THE MODEL
2.5 CAUSAL ASPECTS OF THE THEORY
2.6 HEURISTIC VALUE. Chapter 3: The Classical Approach: Early Twin Studies of PersonalityPublisher Summary
3.1 HISTORY OF THE TWIN METHOD
3.2 ANALYZING TWIN DATA: THE CLASSICAL APPROACH
3.3 THE SHARED ENVIRONMENT IN THE CLASSICAL APPROACH
3.4 CRITICISM OF THE TWIN METHOD
3.5 TWIN STUDIES OF PERSONALITY BEFORE 1976
3.6 POWER AND SAMPLE SIZE
3.7 EARLY TWIN STUDIES: A "META-ANALYSIS"
3.8 SUMMARY
Chapter 4: Introduction to Model Fitting
Publisher Summary
4.1 WHAT'S WRONG WITH THE OLD WAY?
4.2 MODEL BUILDING
4.3 MODEL FITTING
4.4 SUMMARY
Chapter 5: Adult Twin Studies of the Major Personality Dimensions. Publisher Summary5.1 THE LONDON DATA
5.2 THE US STUDY
5.3 THE SWEDISH STUDY
5.4 THE AUSTRALIAN STUDY
5.5 SUMMARY AND DISCUSSION
Chapter 6: Further Tests of the Model: Studies of Adoptees and Extended Families
Publisher Summary
6.1 ASSORTATIVE MATING FOR THE DIMENSIONS OF PERSONALITY
6.2 ADOPTIONS AND EXTENDED KINSHIPS: THE LONDON STUDY
6.3 THE EXTENDED KINSHIPS OF TWINS
6.4 SEPARATED TWINS
6.5 OTHER ADOPTION STUDIES OF EXTRAVERSION AND NEUROTICISM
6.6 SUMMARY
Chapter 7: Personality Development and Change: A Genetic Perspective
Publisher Summary. 7.1 ANALYSIS OF P, E, N AND L IN JUVENILE TWINS7.2 THE CAUSES OF DEVELOPMENTAL STABILITY AND CHANGE
7.3 DEVELOPMENTAL CHANGES IN GENE EXPRESSION IN ADULTS
7.4 STABILITY AND CHANGE IN PERSONALITY
Chapter 8: The Genetic Analysis of Individual EPQ Items
Publisher Summary
8.1 THE DATA
8.2 MODELS
8.3 ESTIMATION
8.4 EXAMPLE
8.5 RESULTS
8.7 DISCUSSION AND SUMMARY
Chapter 9: The Specificity of Gene Effects: Implications for the Interaction of Persons and Situations
Publisher Summary
9.1 PERSONALITY TRAITS AND PERSONS × SITUATIONS INTERACTION
9.2 THE DATA
9.3 ANALYSIS OF VARIANCE. 9.4 GENETIC ANALYSIS9.5 DO GENES AFFECT SHORT-TERM PERSONALITY CHANGE?
9.6 DISCUSSION
Chapter 10: Genetic and Environmental Covariance Between Traits
Publisher Summary
10.1 GENETIC AND ENVIRONMENTAL COVARIANCE: A BIVARIATE STUDY
10.2 A BIVARIATE EXAMPLE: THE NATURE OF EXTRA VERSION
10.3 THE COMPONENTS OF IMPULSIVENESS
10.4 THE COVARIANCE STRUCTURE OF NEUROTICISM
10.5 CONCLUSION
Chapter 11: Normal Personality and Symptoms of Psychiatric Disorder: A Genetic Relationship?
Publisher Summary
11.1 THE THRESHOLD MODEL FOR INDIVIDUAL SYMPTOMS.

Citation preview

Genes, Culture and Personality An Empirical Approach

L.J. EAVES

H.J. EYSENCK

Department of Human Genetics Medical College of Virginia Richmond, Virginia, USA

Department of Psychology Institute of Psychiatry London, UK

N.G. MARTIN Queensland Institute of Medical Queensland, Australia

Research

With contributions by R. JARDINE, A . C . H E A T H , L. FEINGOLD, P.A. YOUNG, K.S. KENDLER

Harcourt London

A C A D E M I C PRESS Brace Jovanovich, Publishers San Diego

Boston

Sydney

New York Tokyo

Berkeley

Toronto

ACADEMIC PRESS LIMITED 24/28 Oval Road London NW1 7DX

United States Edition published by ACADEMIC PRESS INC. San Diego, CA 92101

Copyright © 1989 by ACADEMIC PRESS LIMITED

All Rights Reserved No part of this book may be reproduced in any form by photostat, microfilm, or by any other means, without written permission from the publishers

British Library Cataloguing in Publication Data Eaves, L . J . Genes, culture and personality: an empirical approach. 1. Man. Behaviour. Genetic factors I. Title II. Eysenck, H.J. (Hans Jürgen), 1916III. Martin,. G. 155.7 ISBN 0-12-228290-6

Typeset by Colset Pte Ltd, Singapore Printed in Great Britain by T.J. Press (Padstow) Ltd, Padstow, Cornwall

Authors

L . J . Eaves Department o f H u m a n Genetics, Medical College o f Virginia, B o x 3 3 , R i c h m o n d , Virginia 2 3 2 9 8 , U S A H . J . Eysenck Department o f P s y c h o l o g y , Institute o f Psychiatry, Crespigny P a r k , D e n m a r k Hill, L o n d o n S E 5 8 A F , U K N . G . Martin Queensland Institute Queensland 4 0 0 6 , Australia

of

Medical

Research,

De

Brisbane,

R . Jardine A l c o h o l and D r u g Service, Australian Capital T e r r i t o r y Health Authority, B o x 8 2 5 , C a n b e r r a , A C T 2 6 0 1 , Australia A . C . Heath Department o f H u m a n Genetics, Medical College o f Virginia, B o x 3 3 , R i c h m o n d , Virginia 2 3 2 9 8 , U S A L.

Feingold Department o f Experimental Oxford, Oxford O X l 2 J D , U K

Psychology,

University

of

P . A . Y o u n g Department o f Genetics, University o f Birmingham, P O B o x 363, Birmingham B15 2 T T , U K K. S. Kendler Department o f Psychiatry, Medical College o f Virginia, B o x 7 1 0 , R i c h m o n d , Virginia 2 3 2 9 8 , U S A

Preface

O f all the aspects of h u m a n b i o l o g y to excite our fascination, few can b e m o r e significant than the diversity o f h u m a n b e h a v i o r . Even within a culture, our curiosity is captured b y those differences that seem to m a k e one person stand out from a n o t h e r — their individual styles of living, their preferences and their beliefs, their strengths and their weaknesses. Such differences are the r a w material from which theories of personality are crafted. O u r b o o k tries to analyze the genetic and environmental causes o f these differences. A l t h o u g h studies of the genetic and environmental influences are no substitute for g o o d physiological or psychological theories o f individual differences, they do provide one important arena in which such models can be tested. For example, c a n we persist in a "social learning" theory of personality o r social attitudes if it turns out that the only detectable effects o f parents on their children are genetic rather than social? W h a t are the predictions for the genetic analysis of age-dependent traits o f different mechanisms for the acquisition and transformation o f information about the world? Is there a relationship between the pattern of genetic and environmental causation of individual differences and the evolutionary and sociobiological significance of the traits for the species? W e are still a long w a y from being able to answer all these questions, but one thing is clear: there is absolutely no point w h a t e v e r in beginning to speculate about such issues unless we have a clear idea o f what is actually happening in the real world of personality and attitude differences. T h e main purpose o f the b o o k is to give the reader a clearer idea o f the state of knowledge, and ignorance, a b o u t the causes o f individual differences in personality and attitudes, a g o o d feeling for the kinds o f data on which such inferences are based, and a sense o f the methods o f data analysis that are appropriate for answering basic questions about the role o f biological and cultural inheritance in h u m a n populations. A s a consequence, parts o f the b o o k are m o r e technical than is often the case, there are m a n y m o r e tables and m o r e than usual caution in circumscribing our less than certain c o n clusions. B y providing m u c h o f the original data, we have allowed the reader

viii

Preface

scope to develop and test h i s / h e r o w n ideas and models and to retrace our steps, at least in some of the simpler cases. T h e contract for this b o o k was signed almost fifteen years ago! W e were just completing a twin study in L o n d o n and thought that our results were o f sufficient simplicity and interest to justify a b o o k . W h y the delay? T h e r e are three main reasons. T h e first is our growing interest in social attitudes in addition to personality. Initially we were surprised that genes might play a n y part in the determination o f something so obviously "cultural"; then we were pleased that, b e y o n d the contribution of genetic effects, social attitudes did indeed still display all the hallmarks of cultural inheritance; n o w we are surprised again that the cultural effect m a y still evaporate into the genetic consequences o f assortative mating. Faced with such a consistently developing story, which the reader can reconstruct in the later chapters, a n y attempt at s u m m a r y would have been premature. T h e second reason for delay has been the rapid explosion of t h e o r y and method over the last fifteen years, to which we also have had to devote s o m e time. T h e publication of Jinks' and Fulker's paper on fitting b i o metrical-genetic models to human b e h a v i o r was a landmark in 1 9 7 0 . W h e n we first applied these methods to adult personality, the results seemed fairly straightforward and the models described b y Jinks and Fulker carried us a long w a y . T h e m o r e we b e c a m e involved in new data, however, the m o r e it b e c a m e clear that other issues needed further theoretical w o r k , including mechanisms of cultural inheritance, sex limitation, mate selection, developmental change, social interaction and trait c o v a r i a t i o n . T h r o u g h o u t the 1970s, continuing to the present time, we, and m a n y others in the United States w h o s e contributions we cite, recognized the deficiencies o f the classical genetic models when applied to human b e h a v i o r and did our best to develop theoretical models that had explanatory and heuristic value. M o s t of these ideas do not find their definitive expression in this b o o k , because here we are concerned m o r e especially with the substantive issues o f personality and attitudes rather than theory and model-building for its o w n sake. T h e final cause for delay was the completion and publication o f other large twin studies, which played such a crucial role in refining some o f our early notions based on the smaller sample o f London twins and relatives. T h e large Australian study is still yielding fruit, of which the w o r k described here is only a first sample. W e are conscious, even as we write, that new studies are being done that will transcend the ones we describe for their subtlety and creativity. T h i s w o r k would have been impossible without the financial support of the British Medical and Science Research Councils, the Australian N H &

Preface

ix

MRC, The American Tobacco Corporation, NIH, NIAAA and NIMH, and the grants in aid fund of Virginia Commonwealth University. We are indebted especially to Althea Walton, Greg Porter and Judy Silberg for their enormous assistance with preparing and repairing the manuscript. Richmond,

Virginia

Chapter 1

Another Book on Heritability?

For nearly twenty years the genetic study o f h u m a n b e h a v i o r has acquired a medieval flavor in the public forum. P o p u l a r discussion has a "scholastic" quality with its c o n c e n t r a t i o n on texts, historical material, nuances o f expression and writers' credentials. W h e n discussion sinks to this level, new models and new data are largely irrelevant. M u c h of the debate has been less than illuminating from a scientific perspective. W e think that there are five main reasons for this: (1) almost exclusive attention paid to intelligence; (2) restriction o f discussion to "heritability" at the expense o f other causes o f individual differences; (3) emphasis on reviewing old data rather than presenting original research; (4) verbal description of data rather than quantitative hypothesistesting; (5) small sample sizes and idiosyncratic measurements. In this b o o k we try to address a n u m b e r o f questions about the causes o f individual differences in personality and attitudes in a form that does not sacrifice rigor for didactic simplicity o r scholarly doubt for the persuasive power o f c o n v i c t i o n . A l t h o u g h this m a k e s parts o f the b o o k m o r e difficult, the result c o m e s closer to representing what we think and the degree o f (un)certainty with which we think it than would b e apparent from a m o r e strident a c c o u n t . T h e principal features o f our approach m a y b e presented in contrast with the limitations we listed a b o v e .

1.1 P E R S O N A L I T Y A N D S O C I A L A T T I T U D E S M o s t of the "heat" o f criticism in the 1 9 7 0 s was directed to the analysis o f correlations for I Q . C o g n i t i o n is an important aspect of h u m a n adaptation, but not the o n l y part. T h e r e are consistent patterns of individual differences in b e h a v i o r that emerge in a wide range o f studies using quite different

2

Genes, Culture and Personality

instruments from those used to assess intellectual function. T h e s e are independent of I Q , yet they are consistent over instruments, occasions, cultures, and even species. In m a n , they affect h o w people interact with one another and respond to values of the society to which they belong. T h e s e personality variables are invoked to account for liability to some important psychiatric and social disorders, and predict h o w people respond to certain kinds of drugs a n d schedules of reinforcement. T h e b a c k g r o u n d to the description of personality, its measurement, and its practical and scientific importance is summarized in Chapter 2 . Social attitudes are intrinsically interesting for several reasons. First, individual social attitudes c o m m o n l y change with time. T h e y are thus a m o n i t o r o f behavioral change, which ought to be affected primarily b y the environment. S e c o n d l y , they express, however inadequately, the orientation of the individual towards the society in which he lives. Even if the attitudes people express do not correspond exactly to their actual social practices, they represent an individual's willingness to be counted as believer or agnostic, as nationalistic or not, as liberal or conservative. T h e s e are the basic currencies that express h o w an individual views himself in relation to society, h o w he spends his time, his m o n e y and his v o t e . Thirdly, social attitudes, perhaps m o r e than a n y other aspect o f b e h a v i o r , belong to the human d o m a i n . T h e y represent the interaction between the individual h u m a n and the habitat that he has created for himself. T h e y could not exist, in the form we measure them, without religion, politics, law, social problems, and the nuclear family. Because they relate to functions that are "late b l o o m e r s " phylogenetically, we might expect them to be especially sensitive to the mechanisms o f social learning that characterize the h u m a n species. It is when we turn to social attitudes, therefore, that we should find the paradigms of non-genetic inheritance that have so far eluded the b e h a vior geneticist.

1.2 A W A Y F R O M T H E "HERITABILITY H A N G - U P " A paper b y Feldman and Lewontin in 1 9 7 5 accused b e h a v i o r geneticists o f a "heritability hang-up". T h a t is, the over-riding concern of b e h a v i o r geneticists was to s h o w that behavioral differences were inherited and to estimate h o w much of the variation we measure is due to genetic causes. A cursory review of the literature prior to 1 9 7 0 , and the subsequent public debate a b o u t I Q , m a y h a v e justified their criticism. A t t a c h e d to every heritability estimate there remains a "so w h a t ? " . A heritability estimate does not translate into a prescription for intervention or social change. Indeed, a heritability estimate does not even translate into a selective breeding program for cattle without

1.

Another Book on Heritability?

3

m o r e knowledge about the kinds o f gene action that contribute to the observed genetic differences. It has been tempting, in the past, to equate "heritability" with ' c o n s t r u c t validity" o r "biological significance". Unfortunately the equation has little foundation. A s we shall see, even such a specific trait as a response to an individual questionnaire item m a y h a v e a genetic c o m p o n e n t , and yet it would be difficult to construct a separate biological justification for each item. M a n y o f the inherently important questions of population genetics that would help us relate observed genetic variation to the mechanisms responsible for genetic p o l y m o r p h i s m h a v e still to be solved. T h e chapters that follow provide m a n y instances in which the simple equation "phenotype = genotype + e n v i r o n m e n t " does not do justice to the variety of causes of differences in attitudes and personality. It seems premature to build exotic sociobiological theories to account for certain kinds o f genetic polymorphism, developmental process or social interaction until w e have understood the basic parameters within which such effects operate. W h y should we spend time theorizing about the adaptive significance o f parent-offspring interaction before we h a v e shown (a) whether the interaction is independent o f genotype, and (b) whether the effects o f p a r e n t offspring interaction persist into adult life or whether they merely evaporate when the offspring "leaves the nest"? Different types of social interaction would require different biological explanations. W i t h o u t the basic parameters to reflect these mechanisms, we c a n n o t begin to theorize constructively. It will be apparent in m a n y places that our understanding o f genetic and environmental effects c a n n o t adequately be represented in the notion o f "heritability". A t various times we shall consider: h o w far social interaction between siblings is responsible for creating personality differences (Chapters 5, 6 and 7 ) ; h o w far the family environment creates differences between families (Chapters 6, 7 and 1 1 - 1 6 ) ; the causes and effects of assortative mating for personality and attitudes (Chapters 6, 1 4 and 1 5 ) ; the effects o f sex on the expression o f genetic and environmental effects (Chapter 5 ) ; h o w genes and environment are organized in their effects on multiple variables (Chapters 1 0 and 1 1 ) ; the causes of temporal change in attitudes (Chapter 13); developmental changes in the expression o f genetic and environmental effects on personality (Chapter 7); and whether or not genes contribute additively to phenotypic differences (Chapters 5, 6 and 7 ) . Readers w h o approach the b o o k wanting a purely environmental explanation of h u m a n differences will be disappointed and will find reasons to explain a w a y a n y genetic parameters in our models. T h a t is one reason w h y we provide the data in as complete a form as possible. O n the other hand, anyone w h o believes that simple additive genetic effects can explain family

Genes, Culture and Personality

4

resemblance for personality and social attitudes will find plenty in the following chapters to prove him wrong.

1.3 P R E S E N T A T I O N O F O R I G I N A L D A T A W i t h a few significant exceptions, our b o o k only uses data that we h a v e either gathered ourselves or have analyzed ourselves from r a w data kindly supplied b y other investigators. It thus represents a joint research p r o g r a m spanning fifteen years, in which we h a v e tried to address new questions as they arose. Inevitably, the final account is "idealized" in the sense that the story is organized with hindsight and does not necessarily reflect the exact sequence o f insight into particular problems or the fact that m a n y early analyses were repeated or improved in the final stages. However, the reader should gain a sense o f the close interaction between theoretical developments on the one hand and the realities o f data on the other. S o m e t i m e s n e w data were collected, or new analyses conducted because a theoretical p r o b l e m was recognized that had to b e addressed in practice. T h i s is the case with some o f the studies o f assortative mating described in the later chapters. In other cases, the development o f theory was motivated in part b y the inability of our original models to explain the findings. T h i s is the case for parts o f the treatment of developmental change in gene expression discussed in C h a p t e r 7. W h e r e v e r possible, we have presented data summaries in the form from which we started the analysis o f genetic and environmental effects. T h e principal exceptions are (1) the analysis o f individual items in Chapters 8 and 1 2 (because there are so m a n y items), and (2) the analysis o f extended pedigrees b y maximum-likelihood in Chapters 6 and 7 (since that would require tabulating the individual observations). T h i s strategy has resulted in an unusually large n u m b e r of tables. H o w e v e r , we present the s u m m a r y statistics so that (i) teachers and students can try some of the analyses for themselves, (ii) critics and researchers can develop their o w n models, and (iii) the reader can judge for h i m / h e r s e l f h o w far our conclusions follow from the data.

1.4 Q U A N T I T A T I V E M O D E L I N G A N D H Y P O T H E S I S T E S T I N G M a n y models for individual differences and family resemblance result in complex predictions that can o n l y be expressed and tested in quantitative terms. Consider, for example, the predictions made for the correlations kûft*roûn

r o i o f ix r o c

T A T V I »O Tf r » O To*

Qro

r » n cVirarorl

or»

ri r r \ r t m o n f o l

of"f"o/-fe

T f îo

i n f n ï _

1.

Another Book on Heritability?

5

tively o b v i o u s that genetic factors will m a k e identical twins m o r e alike than non-identical twins (see C h a p t e r 3) because identical twins are genetically identical. T h i s m u c h of the genetic argument can be put into words and a simple statistical test o f the difference between t w o correlations. But what does a given difference between the correlations o f the two types o f twins predict about the resemblance for other kinds o f relatives? A r e the data c o n sistent with these predictions? If they are not then w h y not? T h e s e questions are inherently quantitative, they require a mathematical model for inheritance and gene expression before they can be tested and they require a statistical approach that enables us to test c o m p l e x hypotheses. Similar problems apply to the analysis of cultural inheritance. C o m m o n sense tells us that the environmental impact of parents on their children creates correlations between parents and children and generates sibling and twin correlations. It takes a quantitative model, however, to s h o w that this type o f cultural inheritance has different consequences for different types o f biological and social relationship. It also takes a variety o f quantitative models to express the predictions o f different theories o f social interaction between family members. If we restrict our discussion o f environmental effects to the simple and intuitively o b v i o u s tests o f differences between pairs o f correlations that we can put into words then we m a y b e able to detect family environmental effects, but we shall never c o m e a n y closer to understanding the mechanism of social interaction that created the environmental differences w e detect. Chapter 4 introduces s o m e o f the basic methods o f model-fitting, and subsequent chapters extend these in a variety o f w a y s to encompass m o r e complex hypotheses and intricate data sets. O u r treatment is different from that which n o r m a l l y appears in t e x t b o o k s . T h e subject matter is inherently quantitative and statistical. T o treat the subject in a purely descriptive m a n n e r does not do justice to the scientific questions. It m a y help the reader to keep separate four quantitative aspects o f the subject. T h e s e are: (1) the statistical aspect o f data s u m m a r y (the numbers); (2) the w a y in which w e translate our theories of biological and cultural inheritance into models for the statistics that we collect (model building); (3) the statistical principles that w e use to estimate parameters and test h y p o theses (model fitting); (4) the c o m p u t e r programs and numerical methods that we use to o b t a i n these estimates and statistical tests (number-crunching) . Generally, the data summaries rely on statistical methods that are familiar to the advanced student and readily available on most computers. T h e actual mechanics of deriving the models are conceptually simple but sometimes tedious. It takes m o r e practice to see h o w to write the model in the first place and to acquire the algebraic "tricks" o f derivation. T h e important issue is to get a "feel" for h o w we can write and test models for the genetic and cultural effects contributing to family resemblance.

6

Genes, Culture and Personality

Even if the reader is unfamiliar with all o f the statistical methods, it is o n l y really necessary to understand the logical principles behind the inferences that we m a k e . T h e s e do not differ from those used in the analysis o f a n y other data set. W e did not set out to write a b o o k on statistics or statistical computing. O n the other hand, we h a v e tried to provide sufficient b a c k ground and w o r k e d examples for the beginner to understand the basic statistical and numerical issues for the simpler cases (e.g. Chapter 4) and h a v e assumed that the m o r e complex statistical methods o f the later chapters will either be familiar to the technical reader or irrelevant to a basic understanding of the logical and substantive issues.

1.5 S U S T A I N E D S T U D Y O F S E L E C T E D M E A S U R E S T h e history of personality research is littered with small twin studies o f large numbers o f measures. It is difficult to extract consistent findings from such studies because the sampling errors attached to correlations are quite large and it is not always clear h o w the individual measures relate to a n y overall model for the main dimensions of personality. It is impossible to decide w h e ther the results that we get with nuclear families or adoptions, say, are inconsistent with the results from twin studies because the measures differ, the populations differ o r because twins differ from non-twins. O u r research is based on a few personality dimensions that have been studied repeatedly with large samples in different populations, or with a number of different types of relationship in the same population. B y concentrating on large samples and several kinds of relationship, we hope to be able to test for m o r e subtle features of genetic and environmental determination than are considered in the conventional "heritability study". B y looking at repeated samples from the same population and samples from different populations, we shall be able to see whether the mechanisms responsible for individual differences in personality and attitudes generalize over populations.If they do not then we would question the value of such studies unless there were g o o d reasons for predicting a particular pattern o f differences, as in the case of secular changes in the causes of educational attainment in N o r w a y , recently described b y Heath et al. (1985a, b ) . B y focusing on the main dimensions of personality and attitudes, we m a y not capture all the nuances of individual behavior, but we hope to establish the roots of theory and empirical data on which other ideas can g r o w .

Chapter 2

Dimensions

of Personality

Early genetic studies o f personality (see C h a p t e r 3) gave few consistent results. T h e r e are p r o b a b l y three main reasons for this: (1) small sample sizes; (2) l a c k o f a coherent personality theory; (3) absence o f a n y systematic theory and m e t h o d o l o g y to guide data collection and interpretation. In this chapter we outline the elements o f the personality theory that guided the development o f the measurements that h a v e formed the c o r e o f the investigations we describe in our b o o k .

2.1 T H E O R I E S O F P E R S O N A L I T Y Even though there m a y b e disagreement a b o u t the merits o f particular models, there is n o w a fair a m o u n t o f agreement a m o n g geneticists a b o u t the basic a p p r o a c h that is appropriate in the analysis of h u m a n variation. T h e methods will b e examined and illustrated throughout the subsequent chapters. T h e y a p p r o a c h the paradigm that K u h n (1962, 1 9 7 0 , 1 9 7 4 ) and others (e.g. U r b a c h , 1974) consider appropriate for scientific enquiry. T h e r e is far less uniformity a m o n g psychologists, h o w e v e r , a b o u t what are the marks of a g o o d theory o f personality. Allport ( 1 9 3 7 ) , for example, already discovered o v e r 5 0 meanings o f the term, including quite different and often contradictory definitions. T h e situation has not improved all that m u c h in recent years, where t e x t b o o k s o f personality present a picture either of benevolent eclecticism o r contentious idiosyncrasy. T e x t b o o k s o f personality frequently present the reader with e p o n y m o u s chapters devoted to the theories of one particular a u t h o r each (e.g. Hall and Lindzey, 1 9 7 0 ) , without a n y attempt to address substantive issues, to c o m p a r e different solutions, to l o o k at the empirical evidence for and against a given theory, o r try to arrive at overall conclusions. A t the other extreme (e.g. Cattell, 1 9 8 2 ) , authors cite tew writers other than themselves and their c o l l a b o r a t o r s , and p a y little attention to criticism.

8

Genes, Culture and Personality 2.2 T H E DESCRIPTION OF PERSONALITY

These two attitudes towards personality research are extremes. A g o o d theory o f personality should encompass the solid mass of factual material available, and lead to further empirical studies. It might be argued that such a paradigm exists in personality research (Eysenck, 1 9 8 4 ) . T h i s paradigm in its descriptive aspect, goes b a c k in large part to the ancient G r e e k s , w h o s e theory of the four temperaments (Choleric, Sanguine, Phlegmatic and Melancholic) was based on solid observation. In its modern form, and based on correlational and factor-analytic methods employed upon the results o f self-descriptive questionnaires, ratings b y friends and acquaintances, miniature situation studies, experimental investigations, physiological measures, and h o r m o n a l and other biochemical assays, this model has transcended the purely descriptive phase o f investigation, and has begun to assume a dynamic and causal aspect, relating b e h a v i o r to fundamental biological factors, whether physiological or h o r m o n a l (Eysenck and Eysenck, 1 9 8 5 ) . T h e evidence for this statement is reviewed in detail in the b o o k just cited. In this chapter we shall briefly recapitulate the m a j o r reasons for suggesting that there does exist a paradigm in personality study, and that this paradigm has b o t h descriptive and explanatory aspects. W e summarize the descriptive aspects of the paradigm, and then outline the various types o f tests that such a model must undergo successfully in order to be accepted as being paradigmatic. T h r e e m a j o r dimensions of personality emerge consistently as higher-order or superfactors from large-scale factor analyses o f matrices o f intercorrelations, the elements of which are individual answers to inventory questions, single ratings, or test scores o f one kind or another. T h e s e three m a j o r dimensions have been variously named b y different investigators, and there is no agreement on the semantic problem of h o w best to label them. In this b o o k we shall use the terms suggested b y Eysenck and Eysenck ( 1 9 7 6 ) , leaving the reader free to substitute other terms should he so desire. T h e names given to these three superfactors are psychoticism ( P ) , extra version (E) and neuroticism (N), with ego control, introversion and emotional stability being the opposite ends of the three continua in question. T h e traits that go to m a k e up each of these three factors are shown in Figures 2 . 1 , 2 . 2 and 2 . 3 below; it is the empirically observed intercorrelations between the traits that give rise to the superfactors and legitimate them. Figure 2 . 1 shows the traits that go to m a k e up the psychoticism factor, i.e. the figure indicates that a typical high-P scorer is aggressive, cold, egocentric, impersonal, impulsive, antisocial, unempathic, tough-minded and creative (Eysenck and Eysenck, 1 9 7 6 ) . T h e term "creative" m a y seem to stand out from the others as being wrongly chosen, but the evidence for the

2.

Dimensions of Personality

9

Figure 2.1 Traits correlating together to define the psychoticism dimension (Eysenck and Eysenck, 1985).

proposition that genius and madness are closely allied is quite strong, and the empirical w o r k of G ö t z and G ö t z ( 1 9 7 9 a , b) and W o o d y and Claridge (1977) leaves little doubt about the genuineness o f the association (Eysenck and Eysenck, 1 9 7 6 ) . T h e distribution o f is v e r y much skewed, with few high scorers and m a n y low scorers. T h e introversion dimension is t o o well k n o w n to require much discussion; Figure 2 . 2 shows the traits characteristic o f the typical extravert, indicating that he is sociable, lively, active, assertive, sensation-seeking, carefree, dominant, surgent and venturesome. Introverts, o f course, are the opposite in all these respects, and ambiverts are intermediate between these extremes.

Figure 2.2 Traits correlating together to define the extraversion dimension (Eysenck and Eysenck, 1985).

Genes, Culture and Personality

10

A m b i v e r t s constitute the majority of the population, with extraversion-introversion showing a fairly normal distribution. Neuroticism or emotional instability is characterized b y such traits as being anxious, depressed, having guilt feelings, low self-esteem, being tense, irrational, shy, m o o d y and emotional (Figure 2 . 3 ) . Stable people show the opposite traits, and here, t o o , ambiverts tend to be intermediate, and in the majority, with a continuous, approximately normal, distribution being usually observed in r a n d o m samples o f the population (Eysenck, 1 9 5 2 a ) .

2.3 TESTING THE MODEL FOR PERSONALITY

2 . 3 . 1 Consistency across measurements For such a model to be widely accepted, clearly a n u m b e r o f conditions must b e fulfilled. T h e first o f these is that these three dimensions, in one form or another, should emerge from the great majority of, if not all, statistical studies carried out in this field, and embracing m o r e than just a few restricted traits of personality. A survey o f this kind has been undertaken b y R o y c e and Powell ( 1 9 8 3 ) , and they concluded their survey b y arguing that there were three m a j o r dimensions of personality, which they label ' e m o t i o n a l stability" (the obverse of neuroticism), "introversion-extra version", and "emotional independence" (similar to psychoticism in showing lack o f trust, lack of cooperativeness, tough-mindedness, lack of affect, dominance and realism). Eysenck and Eysenck (1985) have also surveyed correlational studies of the m a j o r instruments used for the investigation of personality, such

Figure 2.3 Traits correlating together to define the neuroticism dimension (Eysenck and Eysenck, 1985).

2.

Dimensions of Personality

11

as the M M P I , the C P I , the 1 6 P F , and m a n y others. T h e y concluded, that for practically all o f these, factors similar to and can be observed in factorial analyses, and that in m a n y cases an additional factor is o b v i o u s . W e m a y thus conclude that these three m a j o r dimensions or superfactors emerge fairly universally in large-scale studies carried out b y A m e r i c a n and European psychologists on samples taken from these countries.

2 . 3 . 2 Cross-cultural consistency T w o objections might be raised to the acceptance o f the paradigm. T h e first might be that the similarities observed are superficial, depending on subjective judgements o f the meaning o f individual items or factors defining traits. T h e second objection might b e that what is true o f A m e r i c a n and European populations might not b e universally true and apply in other countries o f a very different culture. It thus b e c o m e s important to l o o k at cross-cultural comparisons in order to discover whether the same factors could be found in populations w h o s e culture differs very much from the Euro-American. Recent studies on the use o f the Eysenck Personality Questionnaire ( E P Q , Eysenck and Eysenck, 1975) in 2 5 different cultures has suggested an answer to these questions. T h e E P Q (see Appendix B) is a preferred test for the measurement of, and and it has been widely translated and used in a great variety o f countries. In each case, samples o f over 5 0 0 males and females were tested, factor analyses carried out on the matrices o f intercorrelations for males and females separately, and indices o f factor c o m parisons calculated, comparing each c o u n t r y with each other (Barrett and Eysenck, 1 9 8 4 ) . T h e results showed a surprising congruity between different nations, with indices o f factor c o m p a r i s o n nearly always exceeding the 0 . 9 5 level, and v e r y frequently the 0 . 9 8 level. T h u s there are very m a r k e d similarities in the personality patterns found in such widely varying countries as Nigeria, Japan, H o n g K o n g , Brazil, M a i n l a n d China, Uganda, Greece and Bangladesh. W e m a y conclude that not o n l y are the superfactors found in m a n y different measuring instruments in the Western world, but we m a y also conclude that these factors are equally characteristic o f third-world countries in Africa and S o u t h A m e r i c a , of countries in the Chinese-Japanese culture circle, etc.

2 . 3 . 3 Animal models If these dimensions o f personality are so universal within the h u m a n species, it might be suggested that evidence for them might also be found a m o n g

Genes, Culture and Personality

12

animals. After all, if the m a j o r characteristics o f human beings are found in interaction with other people, then it would seem that there are three m a j o r types of interactions. W e m a y be sociable with others (E); we m a y be afraid of others (N); o r w e m a y be aggressive towards others ( P ) . T h e s e three patterns of social interaction are also observed a m o n g animals, and the studies of C h a m o v e , Eysenck and H a r l o w (1972) have shown, on the basis o f the analysis o f years o f careful observation o f Rhesus m o n k e y s , that their b e h a v i o r can indeed be analyzed in terms of these three m a j o r factors. It is even possible to discern evidence for these factors in the b e h a v i o r o f non-primate m a m m a l s , such as the rat. T h e w o r k o f Eysenck and Broadhurst (1965) and Broadhurst (1975) has demonstrated that tests such as the "open" field m a y be used with advantage as a measure o f emotionality, and extensive w o r k in recent years at the B a r c e l o n a A u t o n o m o u s University ( G a r c i a and Sevilla, 1 9 8 4 ) has shown that with suitable alterations the same test can be used for the measurement of extraversion-introversion, using p e r a m b u l a tion as a measure o f extraversion, and defecation as a measure o f neuroticism. Aggression has been measured along m a n y different lines in rats, and presents n o p r o b l e m s . T h e literature has been reviewed in greater detail b y G r a y ( 1 9 7 0 , 1 9 7 3 ) and b y Eysenck and Eysenck ( 1 9 8 5 ) .

2 . 3 . 4 Developmental consistency S o far, we have considered three tests o f the model: support b y m a n y different types of questionnaires and ratings, used b y different investigators on different samples; universality, i.e. of cross-cultural similarity; and that o f animal comparison, i.e. o f applying to animals as well as to h u m a n s . W e n o w turn to another type o f test, n a m e l y that of consistency over time. If the variables we are dealing with are truly fundamental then they should also b e permanent or semipermanent characteristics of the individual, and longitudinal studies o f personality should demonstrate a certain consistency o v e r time. A c t u a l l y there are t w o questions that should be carefully distinguished here, as Hindley and Guiganino ( 1 9 8 2 , p. 1 2 7 ) point out: One concerns the extent to which the behavioral characteristics assessed can be regarded as similar in nature at different ages: the issue of what Emmerick (1964,1967,1968) and Baltes and Nesselroade (1973) termed continuity versus discontinuity of variables. The other concern extends to whether individuals maintain their relative status across ages on the variables in question. T h e first question is dealt with b y Eysenck and Eysenck (1985), w h o c o n clude that as far as, and are concerned, the junior and senior versions of the E P Q s h o w a sufficient degree o f similarity to use them for longitudinal study from early youth to old age. T h e extensive w o r k o f Hindley and G u i -

2.

Dimensions of Personality

13

ganino (1982) is particularly relevant in lending support to this conclusion. C o n l e y (1984) has recently reviewed the literature on longitudinal studies, as well as contributing findings of his o w n . He uses the formula nC = Rs, where C is the observed retest coefficient, R the internal consistency or period-free reliability o f the measuring instrument, s the annual stability and the interval (in years) o v e r which the coefficient is calculated. C o m p a r i n g the annual stabilities o f intelligence and personality traits, particularly, and N , he was able to s h o w that these could b e estimated at 0 . 9 9 and 0 . 9 8 respectively. He concluded (p. 11) that: "Intelligence and personality m a y be regarded as relatively stable characteristics o v e r the length o f adult life span." It is important to note that the intervals o f the studies summarized b y him extended to something like 5 0 years, i.e. a v e r y lengthy span o f time indeed. W e shall return to this issue in C h a p t e r 7, when w e consider developmental changes in gene expression. O t h e r authors (e.g. Schuerger et al, 1 9 8 2 ; Guiganino and Hindley, 1 9 8 2 ; C o s t a et al, 1 9 8 3 ; Eichorn et al, 1 9 8 1 ; M c C r a e and C o s t a , 1 9 8 4 ; and others summarized b y Eysenck and Eysenck, 1 9 8 5 ) suggest that our model indeed passes this hurdle as well, and that consistency o f conduct over m a n y years is characteristic o f the three factors, and N .

2 . 3 . 5 Relationship to psychiatric disorder and social behavior A n o t h e r test o f the model would b e the following. If, and are indeed important aspects o f personality then they should have predictable and testable relationships to important areas o f social conduct. T h e r e is ample evidence that this is indeed so (Wilson, 1 9 8 1 ) . Personality has been found to be closely related to psychiatric a b n o r m a l i t y , in particular neurosis and psychosis (Eysenck, 1 9 7 1 a ) . O t h e r relationships are discussed in Eysenck and Eysenck ( 1 9 7 6 ) . Extraverted types o f neurosis (hysteria, p s y c h o p a t h y ) m a y be contrasted with introverted types (anxiety states, phobias, obsessive-compulsive b e h a v i o r ) . Physical s y m p t o m s o f neurosis are m o r e frequently found in extraverted persons, mental s y m p t o m s in introverted ones (Eysenck, 1 9 7 3 ) . A n o t h e r area o f interest is criminality (Eysenck, 1 9 7 7 ; Eysenck and Gudjonsson, 1 9 8 8 ) . Criminals tend to be high on, and N , and there is evidence that individuals with this personality profile indulge in antisocial b e h a v i o r from quite an early age. T h e s e relations are derived from theoretical considerations a b o u t the nature o f these personality factors, and the evidence suggests that similar personalities c o m m i t crimes in communist and third-world countries as do in Europe and the United States (Eysenck, 1 9 8 5 ) .

Genes, Culture and Personality

14

A number of studies in G e r m a n y have also supported the results originally reported from the United K i n g d o m and other English-speaking countries (Stellar and Hunze, 1 9 8 4 ) . Recently there has been growing interest a m o n g leading psychiatrists (e.g. Cloninger, 1 9 8 0 ) in the capacity of a dimensional model to encompass the k n o w n facts on p s y c h o p a t h o l o g y . Research in the sexual (Eysenck, 1 9 7 6 b ) and marital (Eysenck and W a k e field, 1 9 8 1 ) fields has disclosed m a n y important relations between personality factors, and on the one hand, and sexual attitudes and behaviors and marital happiness on the other. Again these relations were predicted, and the confirmation of these predictions strongly supports the model. O t h e r relationships that have been discussed in m o r e detail b y W i l s o n (1981) relate to affiliation and personal space, birth order, group interaction and social skills, speech patterns, expressive b e h a v i o r and person perception, expressive controls, suggestibility, conflict handling, attraction, attitudes and values, recreational interests, occupational choice and aptitude, industrial performance, academic aptitude and achievement, drug use and abuse, and various others. He concludes his survey b y stating (p. 2 3 9 ) that: Clearly the dimensions of, and . . . have wide explanatory and predictive power across a variety of socially important domains. The suggestion that has sometimes been made to the effect that there are no stable traits which enable us usefully to predict social behavior, is shown to be untenable. Although learning experiences and transient environmental circumstances do have to be considered in predicting what a person will do in a particular situation, so too must their personality be taken into account, or the formula is bound to be incomplete. 2.4 C R I T I C I S M S O F T H E M O D E L 2 . 4 . 1 Criticism o f trait theories S o far we have dealt with the model in its descriptive aspects, and such models have sometimes been criticized on a variety of grounds. It has been suggested (e.g. b y Mischel, 1 9 6 8 , 1 9 7 7 ; Mischel and P e a k e , 1 9 8 2 ) that the study o f personality b y w a y o f traits is essentially unproductive, and that situations contribute m o r e than do traits to individual b e h a v i o r . T h e argument has been discussed in great detail b y Magnusson and Endler ( 1 9 7 7 ) . A detailed refutation is given b y Eysenck and Eysenck (1980), so we do not repeat the arguments here in detail. However, material already surveyed in this chapter demonstrates that Mischel's conception is at best one-sided, and at worst completely mistaken. T h e issue of person situations interaction will be considered from a genetic perspective in Chapter 9 .

2.

Dimensions of Personality

15

2 . 4 . 2 Nomothetic vs. idiographic methods A rather different type of criticism is that originally put forward b y Allport, w h o argued in favor o f an idiographic instead o f a n o m o t h e t i c type o f approach to the p r o b l e m of personality. T h e usefulness o f the n o m o t h e t i c approach has been demonstrated t o o often and t o o incisively to m a k e it possible to disregard the arguments in its favor, whereas there is no evidence that a n y a d v a n c e has been made in the idiographic study of personality (for contrary arguments see R u n y a n , 1 9 8 2 ) . Indeed, Allport himself, whenever he published empirical data, used the n o m o t h e t i c rather than the idiographic approach. A m o r e detailed e x a m i n a t i o n o f this issue m a y be found in an earlier publication (Eysenck and Eysenck, 1 9 8 5 ) .

2 . 4 . 3 Inability to explain social behavior A n o t h e r frequent criticism o f trait p s y c h o l o g y is one that has been made in a similar form o f the early t h e o r y o f instincts. It is pointed out that, just as the postulation o f an instinct o f "sociability" does nothing to explain the phenomena o f social b e h a v i o r , because the instinct is predicated upon the phenomena it is supposed to explain, so similarly a postulation o f a trait o f "sociability" does nothing to explain social or asocial b e h a v i o r . In answer, it should b e said that while the objection is true, it is irrelevant. T r a i t p s y c h o logy is essentially descriptive; it m a k e s no causal claims. T o postulate a trait of "sociability" is not to attempt an explanation of sociable b e h a v i o r ; it is merely to claim descriptively that instances o f social b e h a v i o r correlate together o v e r a group o f persons, and define and m a k e measurable this particular trait. It might appear a task o f supererogation to c a r r y out such w o r k , but this is not s o . A s Eysenck (1956a) has shown, in an empirical study o f the Guilford questionnaire of social shyness, there is not only one type o f social shyness, but two quite uncorrelated ones. S o m e people are lacking in sociability because they do not care for other people (introversion); others b e h a v e in an unsociable m a n n e r because they are afraid o f other people (neuroticism). T h u s , because we possess a single term, sociability, to c h a r a c terize a certain type of b e h a v i o r , it does not follow that we can postulate a single factor, and careful correlational and factor-analytic w o r k has to be done in order to discover to what extent our semantic habits agree with descriptive facts. M a n y other examples are given b y Eysenck and Eysenck (1985) to illustrate this point. A c c u r a t e observation and a well-founded t a x o n o m y of b e h a v i o r are needed to give us a proper descriptive basis for trait psychology, and it would b e quite w r o n g to rely on simple semantic habits and everyday convenience in our description o f personality traits. T h u s this

16

Genes, Culture and Personality

particular objection also must be ruled out as not properly applicable to the study o f personality traits.

2.5 C A U S A L A S P E C T S O F T H E T H E O R Y While a descriptive, t a x o n o m i c phase is essential in scientific analysis, and must precede the m o r e causal or d y n a m i c phase, it is not in itself sufficient. W e need to ask causal questions, and these, in the personality field, c a n often be answered in what are often called "reductionist" terms, i.e. in terms that would reduce psychological variables to concepts coming from a m o r e fundamental biological b a c k g r o u n d . Reductionism is often criticized on philosophical grounds, but this would seem to be inappropriate in particular circumstances. It m a y or m a y not be the case that all h u m a n b e h a v i o r , including cognition, can be ultimately reduced to physiological or b i o chemical, or physical concepts and theories. Such a belief c a n n o t b e supported empirically at the present time. However, certain relationships have been observed between psychological variables, such as those entering into our description of personality, on the one hand, and physiological and biochemical variables, on the other. T h u s Eysenck (1967b) has suggested that extraverts are characterized b y a l o w level of cortical arousal, while introverts are characterized b y a high level of cortical arousal. He has further suggested that these levels are in part controlled b y the ascending reticular activating system, and that these biological differences are causally related to the b e h a v i o r characteristic o f extraversion and introversion respectively (Eysenck, 1 9 8 1 ) . Similarly, neuroticism has been related to the activity o f the limbic system, expressed through the sympathetic and parasympathetic branches o f the a u t o n o m i c system ( S t e l m a c k , 1 9 8 1 ) . O t h e r theories have been prepared that try to account for observed trait variation in terms of the structure, physiology and b i o c h e m i s t r y o f the nervous system ( G r a y , 1 9 7 0 ; Z u c k e r m a n , 1 9 7 9 ; Z u c k e r m a n et al., 1 9 8 4 ) . W e need not concern ourselves here with the question of whether such theories are correct or incorrect; they are merely quoted as an example of the possible reduction of factors in the personality field to biological systems that are considered to be m o r e elementary and fundamental. Physiological variables are not the o n l y ones to be related to the m a j o r dimensions o f personality. O l w e u s (personal c o m m u n i c a t i o n ) has shown correlations o f 0 . 5 between adrenaline secretion and b o t h introversion and neuroticism; thus the simple h o r m o n a l secretion m a y account for something like 2 5 % o f variation in the aspect of personality usually called "trait anxiety", which is a mixture o f introversion and neuroticism. T h u s there is evidence of a relationship between b e h a v i o r and h o r m o n e secretion.

2.

Dimensions of Personality/

17

A n o t h e r example o f the relationship between personality and biochemical variables is the important effect that the enzyme m o n o a m i n e oxidase ( M A O ) seems to exert on sensation-seeking and extra verted b e h a v i o r . M u c h w o r k has been done in this field on b o t h m o n k e y s and humans, and quite a strong negative correlation has been established between platelet M A O and extraversion in general, and sensation-seeking in particular. T h e evidence is surveyed b y Z u c k e r m a n et al. ( 1 9 8 4 ) ; it is o n l y o n e of m a n y such relationships that h a v e been studied in s o m e detail in the past. T h e o u t c o m e o f these studies leaves little doubt that there are important biological foundations for differences in personality, and this fact b y itself suggests that genetic factors might b e involved in an important w a y in causing differences in b e h a v i o r , with neurobiological structures and secretions mediating this relationship. Early w o r k b y Eysenck and Prell (1951) and Eysenck (1956a) gave support to this view. T h e s e findings o f strong genetic involvement in personality differences contradicted the accepted view ( N e w m a n et al, 1937) o f little genetic influence o n personality, and m a r k s the reawakening o f interest in this field. T h e earlier w o r k s h a v e been criticized b y Eysenck ( 1 9 6 7 b ) .

2.6 HEURISTIC VALUE T h e final attribute that justifies giving a model the status o f a paradigm is that it is a fertile source o f experimental predictions. Deductions can be m a d e from the physiological, causal model that are testable; s o m e o f these are themselves physiological, others are experimental along traditional p s y c h o logical lines. A g o o d review of the former studies that h a v e been carried out in relation to personality can b e found in S t e l m a c k (1981), while the latter are reviewed in Eysenck ( 1 9 6 7 a , 1 9 7 6 a , 1 9 8 1 ) . T h e list of such studies is very long indeed, ranging from perceptual variables like sensory thresholds, figurai after-effects, pain and sensory-deprivation tolerance, to various aspects o f m e m o r y , and learning and conditioning. Indeed, it is possible to argue that personality variables interact with almost all the aspects o f experimental, social, clinical, educational and industrial p s y c h o l o g y that have been studied b y empirically minded psychologists, and that this interaction is so close that it a c c o u n t s for m o r e o f the variance, in most studies, than do the so-called main effects. These, then, are our reasons for adopting the, and model o f personality provisionally as o u r phenotype, for quantitative genetic analysis. It is not necessary for the reader to adopt the personality model that we h a v e outlined a b o v e in toto. It can be translated into other models, and it should be possible to use our data to effect such a transformation, and to derive some knowledge about the genetics o f other systems from the data here

18

Genes, Culture and Personality

given. But it must be doubtful if such a translation is really necessary or advisable in view of the paradigmatic aspects o f our model. In due course, the model will be improved, and m a y even be a b a n d o n e d . But a n y better model will need to incorporate at least as m u c h biological, experimental and descriptive evidence, and must offer still m o r e far-reaching theories to mediate between the various aspects of the model. M o s t personality theorists have been content to l o o k at one particular corner of this whole field, rather than to try and c o v e r its m a j o r aspects. M u c h remains to be studied; it is not claimed for this model that it covers all of personality, but merely that the three dimensions it deals with are important, perhaps the most important, descriptive and causal features of personality. Future research will no doubt add to this array, and m a y even substitute new, and better, variables for those here singled out for c o m m e n t . Similarly, future research will undoubtedly l o o k in m o r e detail at the various traits themselves that m a k e up the superfactors along the lines of analysis used for impulsive and sensation-seeking b e h a v i o r b y Eysenck ( 1 9 8 3 a ) . In all, it is not claimed that the analysis of personality briefly discussed here is anything but the beginning o f the scientific study of personality, and, equally, the discussions of the genetics o f personality are in truth o n l y a beginning, laying down in a rough form certain conclusions that future w o r k will undoubtedly refine, and m a y even reject. Nevertheless, it seems to us that at the m o m e n t there is sufficient agreement in the empirical field to justify the claim that the findings are neither statistical artifacts, n o r idiosyncratic observations, but that we h a v e here the beginnings of a paradigm that justifies our adopting it as the central focus of our analyses o f genetic and environmental effects.

Chapter 3

The Classical Approach: Early Twin Studies of Personality

3.1 HISTORY O F THE TWIN METHOD 3 . 1 . 1 Hereditary genius T h e first empirical studies o f the inheritance o f h u m a n b e h a v i o r were described b y Francis G a l t o n in Hereditary Genius ( 1 8 6 9 ) . He attempted to define and measure family resemblance for exceptional ability in a wide range of professions and skills ranging from the law and the church to rowing and wrestling. In these studies, G a l t o n introduced the "proband m e t h o d " in which "high-risk" families are ascertained through affected individuals. T h i s w o r k showed: (1) that the empirical risk o f genius in the relatives o f eminent probands w a s far higher than would be expected for qualities so rare in society as a whole; (2) that the p r o b a b i l i t y that a relative o f an eminent prob a n d would himself b e eminent declined as the relationship b e c a m e m o r e r e m o t e . T h e s e findings led G a l t o n to formulate the first mathematical model for family resemblance, the " L a w o f Ancestral Heredity", which described the facts o f family resemblance quite well but w a s not based in the m o r e precise understanding o f the actual m e c h a n i s m o f inheritance that was to follow from the rediscovery o f Mendel's w o r k in the early 1 9 0 0 s .

3 . 1 . 2 The history of twins G a l t o n saw a m a j o r p r o b l e m with his a p p r o a c h . M o s t individuals w h o are related biologically are also related socially. Parents provide part o f the environment for their children as well as providing their genes. T h e twin study w a s proposed as an elegant solution to this difficulty in his b o o k Inquiries into Human Faculty ( 1 8 8 3 ) . He observes that the objection to

20

Genes, Culture and Personality

statistical evidence in p r o o f of the inheritance of peculiar faculties has always been: The persons whom you compare may have lived under similar social conditions and have had similar advantages of education, but such prominent conditions are only a small part of those that determine the future of each man's life. It is to trifling accidental circumstances that the bent of his disposition and his success are mainly due, and these you leave wholly out of account — in fact, they do not admit of being tabulated, and therefore your statistics, however plausible at first sight, are really of very little use. No method of inquiry which I had previously been able to carry out — and I have tried many methods — is wholly free from this objection. I have therefore attacked the problem from the opposite side, seeking for some new method by which it would be possible to weigh in just scales the effects of Nature and Nurture, and to ascertain their respective shares in framing the disposition and intellectual ability of men. The life-history of twins supplies what I wanted. T h e logic of the twin m e t h o d was simple. Identical twins h a v e the same genes and share the same parents. A n y differences within pairs o f M Z twins must be due to their unique environmental experiences. Non-identical twins have the same parents but different genes. Differences within pairs o f nonidentical twins must be due to genetic effects and their unique environmental experiences. If the differences within pairs o f D Z twins are greater than the differences within M Z twins, the excess must be due to genetic effects. Galton's o w n use o f the m e t h o d was disappointing and confined to a discussion of anecdotal accounts derived from letters sent to him b y twins in response to appeal. Nevertheless, his final conclusions s h o w great insight into the dependence o f estimates o f genetic and environmental effects on the populations and cultures from which they are sampled: There is no escape from the conclusion that nature prevails enormously over nurture when the differences of nurture do not exceed what is commonly to be found among persons of the same rank of society and in the same country. My fear is, that my evidence may seem to prove too much, and be discredited on that account, as it appears contrary to all experience that nurture should go for so little. But experience is often fallacious in ascribing great effects to trifling circumstances. T h e main weakness o f Galton's study was the lack o f quantitative b e h a vioral data on which to base his analysis. T h e exploitation o f the twin method only began seriously in the 1920s and 1930s with the early development of behavioral measurements. Foremost a m o n g these investigations w a s the classic study o f N e w m a n , Freeman and Holzinger (1937), which supplemented a sample o f twins reared together with 19 pairs of m o n o z y g o t i c twins w h o had been reared apart. T h i s study, and the m a n y other studies o f family

3.

21

The Classical Approach

resemblance for cognitive abilities, h a v e been reviewed frequently (e.g. Fuller and T h o m p s o n , 1 9 7 8 ) . 3.2 ANALYZING T W I N D A T A : THE CLASSICAL A P P R O A C H T h e conventional approach to the analysis o f twin data addresses t w o issues: (1) do genetic factors contribute significantly to individual differences? and (2) what are the relative contributions o f genetic and environmental effects to variation within twin pairs? T h e s e t w o questions deal with the related statistical issues o f hypothesis testing (Are genetic effects statistically significant?) and estimation ( W h a t is our best estimate of the contribution o f genetic factors to v a r i a t i o n ? ) . Traditional analyses o f twin data focus o n l y on estimating and testing the significance o f the genetic c o m p o n e n t . O u r subsequent treatment will illustrate the estimation o f genetic and non-genetic parameters and testing o f c o m p l e x hypotheses concerning the joint effects o f genes and environment. 3 . 2 . 1 Analysis o f variance T h e starting point for the genetic interpretation o f twin data is a data summary derived from the pairs of observations made on large numbers of M Z and D Z twin pairs. O n e convenient approach starts with the nested analysis of variance of each group o f twins (see S n e d e c o r and C o c h r a n , 1 9 8 0 ) which divides the total variation into that between pairs and within pairs. If there are JV twin pairs and X{] represents the measurement o f the ;th twin in the ith pair then the analysis computes the mean square between pairs ( M S B ) from MSB = Where

+ Xi2)

2

2

2NX ]/(N-1),

-

is the m e a n o f the 2N observations:

T h e m e a n square within pairs ( M S W ) measures the average variation o f individual twins around the averages o f the pairs to which they belong, and is computed as

M S W =( * . -

W

T h e analysis is illustrated for neuroticism scores for female M Z and D Z twins comprising part o f the large National Merit T w i n Study (Loehlin and

22

Genes, Culture and Personality

Table 3.1 Analysis of variance of neuroticism scores of female MZ and DZ twins from the National Merit Study. Source Monozygotic Between pairs Within pairs Dizygotic Between pairs Within pairs

df

Mean square

F

r

266 267

41.9 14.5

2.88***

0.4858

175 176

34.2 21.7

1.58**

0.2236

Expected MS a WMZ + a WMZ

2

a

2 ( 7 BDZ

WOZ

+

^WDZ

** Significant at 0.01 level; ***, significant at 0.001 level.

Nichols, 1 9 7 6 ) . T h e data, kindly made available b y D r R . C . Nichols, are discussed m o r e fully in Chapter 5 . T h e mean squares o f the analysis o f variance are given in T a b l e 3 . 1 . Separate analyses are conducted for M Z and D Z twins. T h e variance ratio F = M S B / M S W m a y be used to test whether the differences between pairs are statistically significant (see Snedecor and C o c h r a n , 1 9 8 0 ) . A significant variance ratio indicates that there are genuine differences between pairs of twins that are not explained b y sampling alone and implies that members o f twin pairs resemble one another m o r e than unrelated individuals paired at r a n d o m from the population. B o t h M Z and D Z twins are significantly correlated for their neuroticism scores.

3 . 2 . 2 Genetic and environmental c o m p o n e n t s within families T h e basic statistical analysis only tells whether there is significant family resemblance in personality. It does not say anything about whether these differences are due to genes or environment. G a l t o n ' s original argument c o n centrated on the causes of differences within twin pairs, which, in the analysis o f variance, contribute to the mean squares within pairs. T h e m e a n square within pairs of M Z twins, M S W M Z, is simply a function o f environmental effects within families, E w . T h e mean square within D Z twin pairs is assumed to be the sum of two c o m p o n e n t s . T h e first, £ w , reflects the same kinds o f environmental influences that m a k e M Z twins differ from one another. T h e second, G w , reflects the contribution of genetic effects arising from the segregation within pairs o f D Z twins o f alleles having different effects on personality. T h e classical approach does not ask anything about the kinds o f gene action contributing to G w and assumes that genetic and environmental effects are additive and independent (i.e. that there is neither

3.

23

The Classical Approach

genotype environment interaction n o r g e n o t y p e - e n v i r o n m e n t correlation — see C h a p t e r 4 ) . T h e observed c o m p o n e n t s o f variance within M Z and D Z pairs m a y thus be given expected values in terms of the two parameters o f the simple "model": Observed 21.7 = = 14.5 = =

Expected £ w + G w, £ w.

(3.1) (3.2)

Vandenberg (1966) proposed using the variance ratio (F) test =

^WDZ/^WMZ

as a test o f whether G w w a s significantly greater than zero. For these data we obtain F = 1 . 5 0 for 1 7 6 and 2 6 7 df, a clearly significant value. Although the F-statistic gives a valid test o f significance for the importance o f genetic effects o v e r and a b o v e the b a c k g r o u n d of r a n d o m environmental effects, it does not yield a sense of the relative contribution o f genetic and environmental factors to individual differences. T h e pair o f simultaneous linear equations ( 3 . 1 ) , ( 3 . 2 ) a b o v e m a y b e solved to yield estimates o f £ w and G w thus: 5 f w G

w

= =

0WMZ

=

fryvDZ

-

· 0"WMZ

1

4

> =

7.2.

T h e units o f variation are determined b y the units o f measurement, so it is m o r e convenient to express the contribution o f genetic factors as a proportion of the variation caused b y b o t h genetic and environmental effects. Holzinger (1929) proposed a statistic that has been used frequently b y twin researchers:

ft =

. G

6 w

w

+ Êw

~ F

1F

which yields H = 0 . 3 3 for our neuroticism data. T h i s statistic is sometimes referred to as the "heritability" estimate, but in fact it does not correspond to a n y estimate o f heritability employed b y geneticists elsewhere. Holzinger's H o n l y includes genetic and environmental differences within twin pairs and ignores the fact that pairs are also expected to differ from one another because they receive their genes and environments from different parents (see e.g. Jinks and Fulker, 1 9 7 0 ) .

24

Genes, Culture and Personality

3 . 2 . 3 T h e intraclass correlation and twin similarity Just as genetic and environmental differences contribute to differences between twins within a pair, so one pair differs from another because each pair derives its genes and part of its environment from a unique set o f parents. T h e fact that each pair has a unique set of parents m a k e s m e m b e r s o f a pair similar to one another and makes one pair different from a n o t h e r . T h e significant variance ratios reported for M Z and D Z twins indicate that these between-family effects are important, but does not directly measure their relative contribution to individual differences. A statistic that is frequently used to summarize data on the resemblance o f family members is the intraclass correlation, since it does not depend on the units o f measurement and often has a simple intuitive rationale. T h e intraclass correlation r is a p p r o priate for grouped data that can be analyzed legitimately b y the nested analysis of v a r i a n c e . T h e statistic c a n n o t properly be used with twin data if there are significant effects o f birth-order on mean or variance, n o r can it b e used with unlike-sex twin pairs if there are sex differences in mean or variance. T h e intraclass correlation is the proportion o f the total p h e n o t y p i c variance a m o n g individuals that is due to factors which differ between twin pairs. A n o t h e r w a y of saying the same thing is that the correlation measures the proportion o f the total variance explained b y factors shared b y twin pairs. T h e total phenotypic variance in a twin sample is the sum o f t w o c o m 2 ponents of variance: the c o m p o n e n t within pairs a w and the c o m p o n e n t 2 between pairs, o ^. T h e intraclass correlation is thus the ratio r = a\l{o\

+ ow).

T h e c o m p o n e n t s of variance within M Z and D Z pairs can be obtained directly from the mean squares within pairs M S W (see a b o v e ) . T h e betweenpair c o m p o n e n t s , however, are not equal to the mean squares between pairs. T h e reason is simple, but m a y still need explaining. Consider the mean o f the z'th pair o f twins, Xi . T h e elementary statistical formula for the variance o f 2 an estimate o f a m e a n based on observations is In our case, the twins are repeated samples from a "population" with variance, so the variance of a typical pair-mean is expected to be Via^. Even if no differences are expected between pairs b y virtue of their arising from different parents, the average o f one pair will differ from another because o f these sampling effects alone. S o the variance of the pair-means is expected to be Vzafyj even if there are no real differences between pairs. N o w , if there are differences between pairs over and a b o v e the effects o f sampling, the variance o f the pair-means is Via^ +, the extra term o\ being the true variance o f pair-means. If it were possible to h a v e families of infinite size (rather than size 2 , as in the case

3.

25

The Classical Approach

with twins) then the contribution o f sampling variation would vanish and we would just b e left with the true variance o f family-means, o\, which would only reflect differences between pairs without a n y contribution from factors that create differences within pairs. O u r discussion is based on the variances of pair-means. T h e analysis o f variance generates the mean square between pairs, which is not the same thing. A little algebra shows that the mean square between pairs is twice the variance o f pair-means, so we h a v e a\ = jj.(MSB - M S W ) .

(3.3)

Substituting the m e a n squares for neuroticism in ( 3 . 3 ) , we h a v e , for M Z twins, 2

= 13.7,

and, for D Z twins, = 6.8, yielding intraclass correlations of 0 . 4 8 6 and 0 . 2 2 4 for M Z and D Z twins respectively. Just as we expressed the c o m p o n e n t s o f variance within pairs in terms o f genetic and environmental effects, so we m a y write expressions for the variance c o m p o n e n t s between pairs as functions o f hypothesized genetic and environmental variances between families thus:

^BMZ

=

GB + £B, = G B + G w + £ B.

^BDZ

T h e expectation for D Z twins requires no explanation. S o m e people are troubled b y the inclusion o f G w in the expectation for M Z twins. T h e difference reflects the m e c h a n i s m b y which the different types o f twins are formed. B o t h M Z and D Z twins derive their genes and some o f their environment from unique pairs o f parents w h o s e genes and b e h a v i o r h a v e their o w n characteristics. T h e s e differences contribute to G B and £ B . H o w e v e r , when D Z twins are produced, each twin is a genetically unique individual obtained b y sampling alleles from the parental genotypes. T h i s sampling process (which follows Mendelian processes o f segregation and assortment) contributes to G w in D Z twins. In the case o f M Z twins, however, a single sample of parental alleles is generated at fertilization and replicated in two genetically identical individuals. T h u s the segregation and assortment that create differences within D Z pairs adds to differences between M Z pairs. T h e total phenotypic variance VP for M Z or D Z twins is expected to be the sum of four c o m p o n e n t s given that genes and environment act additively and independently: VP = G

w

+ GB + £

w

+ £B

26

Genes, Culture and Personality

The intraclass correlations for M Z and D Z twins are then expected to b e *MZ = ( G W + G B + E B) / V p

(3.4)

and r DZ

= ( G W + £ B ) / V P.

(3.5)

The difference between the correlations is thus an estimate of the proportion of the total variance that is due to genetic factors within families, since Q

rMZ-rOZ

=

— - —

= 0.262

for the neuroticism data.

Vp A n alternative form of Holzinger's H, derived from correlations rather than variances, is

1-*DZ

For the example data the value o f H is 0 . 3 4 . T h e small difference between this value and that derived from the within-pair variances is due to the slight (but not significant) difference in total variance between M Z and D Z twins, which is the d e n o m i n a t o r in the formula for estimating the correlation coefficients. If the total variances o f M Z and D Z twins differ significantly then neither estimate of H is valid, since the simple genetic model predicts that the variances of the t w o types of twins should be the same if they are sampled from the same population of genetic and environmental effects. T h i s test of the model w a s often ignored in early analyses. A (two-tailed) F-test c o m paring the total variances for M Z and D Z twins gives a non-significant value (F 533,35i = 1 . 0 0 9 ) , confirming that the comparison between correlations is not affected b y sampling differences between the distributions o f M Z and D Z twins. A n alternative to Vandenberg's F for testing the contribution of genetic factors is to test whether the correlation for M Z twins is significantly greater than that for D Z twins. T h e correlations m a y be converted into n o r mally distributed statistics before they are compared to see if they differ significantly (see S n e d e c o r and C o c h r a n , 1 9 8 0 ) . Since genetic theory predicts an excess o f M Z similarity, we conduct a one-tailed test and find that the correlations are significantly heterogeneous = 9 . 5 9 , P EB E„, VA Ewr EB, VA Ew, VA, vD Ew, VA, CAS

Ew 1.01*** 0.67*** 0.54*** 0.51*** 0.51*** 0.52***

2

EB

VA

vD

VAS

X

df

—

—

—

—

0.34*** — -0.23***

—

—

0.48*** 0.72*** 0.04 0.56***

— 0.46

— — — —

—

-0.04 +

137.0*** 44.3 11.3 7.0 7.0 9.4

9 8 8 7 7 7

— —

—

Table 10.10 Model fitting to impulsiveness scores: no sex limitation. Model

EW

Ew Ew, EB

1.00*** 0.73*** 0.64*** 0.65*** 0.65*** 0.65***

EW, EW, EW, EW,

VA VA

V A, VD VA, C AS

C AS

HB —

0.28*** —

0.01 — —

—

0.36*** 0.34*** 0.38* 0.35***

— —

-0.03 —

0.00

2

71.8*** 9.4 1.9 1.9 1.9 1.9

df 9 8 8 7 7 7

Significance levels: * P < 0 . 0 5 ; * * P < 0 . 0 1 ; * * * P < 0 . 0 0 1 A l t h o u g h the environmental model fits the impulsiveness data well, the fit of the two-parameter g e n o t y p e - e n v i r o n m e n t a l model is much better. In n o case does the addition o f a between-families environmental c o m p o n e n t ( E B) to the genetic model lead to a significant positive estimate of E B . In the case o f the sociability scores the estimate o f E B is negative and significant at the 5 % level, suggesting a hint o f d o m i n a n c e , but c o m p a r i s o n with the sex-limited model in T a b l e 1 0 . 8 suggests that the effects of d o m i n a n c e cannot be

256

Genes, Culture and Personality

Table 10.11 Model Ew E w, E B EW, v A E w, E B , v A EW, VA, v D EW, VA, VAS

Model fitting to extraversion (S + I) scores: no sex limitation. Ew

EB

x

VA

2.71*** 1.86*** 1.54*** 149*** 249*** 1.50***

—

2 27* * *

-0.42

1.63***

— — —

0.37

0.84

0.85***

— —

1.34***

-0.09

—

—

2

df

111.0***

9

31.0***

8

7.4

8

—

5.1

7

—

5.1

7

6.2

7

—

Table 10.12 Model fitting to difference (S - 1 ) scores: no sex limitation. Model Ew E W/ E w, E w, E w. E w,

EB VA E B, v A VA, V A , V AS

Ew

1.30*** 0.94*** 0.83*** 0.83*** 0.83*** 0.83***

EB

0.36***

— -0.02

— —

vA

C AS

— — 0.47*** 0.49** + 0.42 0.48***

— — —

—

0.04

—

0.01

2

84.6 19.0* 9.2 9.2 9.2 9.2

df

9 8 8 7 7 7

distinguished from the l o w correlation for unlike-sex twins, which could result from m a r k e d sex limitation o f gene expression. In view o f the discussion of d o m i n a n c e in Chapter 5 , it is o f some interest to note that the estimate of the d o m i n a n c e c o m p o n e n t of the extraversion scores (Table 1 0 . 1 1 ) is quite large, though admittedly not statistically different from zero. Again, it is difficult to disentangle the apparent effects of dominance from those of sex limitation in these data. O v e r a l l , the results for the extra version scores from the Personality Inventory correspond well with those reported for the E P Q extraversion scores in the L o n d o n twins (see Chapter 5 ) . T h e main determinants o f individual differences appear to be genetic effects and within-family environmental differences. T h e possibility that the d o m i n a n c e / c o m p e t i t i o n c o m p o n e n t o f extraversion might be confined to the "sociability" c o m p o n e n t is consistent with an explanation in terms of competitive social interaction between twins based on availability of resources of social reinforcement, but the evidence is w e a k . T h e heritability of the total extraversion score turns out to be 0 . 4 2 , but it is important to notice that the S - I difference score also has a signifi2 cant genetic c o m p o n e n t (h = 0 . 3 4 ) so the distinction between sociability and impulsiveness described b y several authors c a n n o t be attributed to environment alone, but the aspects o f personality are somewhat distinct genetically. T h e fact that the heritability of the scores is slightly higher than that o f the S - I scores implies that the genetic c o v a r i a n c e between

10.

Genetic and Environmental Covariance Between Traits

257

measures of sociability and impulsiveness exceeds the environmental c o v a riance. T h a t is, genes are the main determinant of the phenotypic correlation between these aspects of personality.

1 0 . 2 . 3 The causes of correlation between the scales The analysis of the sums and differences of the standardized sociability and impulsiveness scores reflects the underlying genetic and environmental covariance structure of the t w o subscales of extra version. Rather than c o m bine the scales according to a priori criteria, however, we m a y directly examine the variation and covariance of the component traits. Since the V A, E w model gave the best fit to the composite scores, we used the method of weighted least squares to estimate six parameters: V A i, V A /, V A l ,/ £ W l, E w / and £ W l, / from the 3 0 mean squares and mean products given in Table 1 0 . 6 . The W L S estimates of the parameters, their estimated standard errors and associated tests of significance are given in Table 1 0 . 1 3 . All the parameter estimates differ v e r y significantly from zero, and the model gives a good fit to the data (X^ 4 - 2 9 . 4 , - 0 . 2 ) . Since the covariance components are both significant, we conclude that the correlation between sociability and impulsiveness is caused both by genetic and environmental factors. The heritability of the sociability scores is estimated to be 0 . 4 6 , and that of impulsiveness is 0 . 3 6 . O u r estimates of the genetic and environmental variances and covariances m a y be combined, if desired, to give estimates of the genetic and environmental variances of composite scores. If we combine sociability and impulsiveness into a single extraversion score by adding the two subscales with equal weight (as we did originally) then the additive genetic component of the extraversion score should be the sum of the additive genetic components of the component scales and twice the genetic covariance between scales. The environmental component should likewise be the analogous sum of the environmental variances and c o v a r i a n c e for the two subtests. For the c o m bined extraversion score, therefore, we find the additive genetic variance to

Table 10.13 Estimates ( ± s.e.) of genetic and environmental variances and covariances for sociability and impulsiveness.

Sociability Impulsivity Covariance

Genetic ( V A)

Environmental ( E w )

0.461 ± 0 . 0 4 1 0.356 ± 0 . 0 4 2 0.179 ± 0 . 0 3 1

0.541 ± 0 . 0 3 2 0.644 ± 0 . 0 3 8 0.176 ± 0 . 0 2 6

Genes, Culture and Personality

258

be 0 . 4 6 + 0 . 3 6 + 2 0 . 1 7 = 1.16. Similarly, the environmental variance within families is 0 . 5 4 + 0 . 6 4 + 2 x 0 . 1 8 = 1.54. B o t h o f these values are very close to the estimates derived directly from the total extraversion scores under the VA, Ew model. T h e genetic c o m p o n e n t o f the difference score, D = S - 1 , is the sum o f the additive genetic c o m p o n e n t s o f the t w o separate scales minus twice their genetic c o v a r i a n c e . Substituting the parameters estimated from the bivariate model, w e obtain an estimate o f 0 . 4 6 + 0 . 3 6 - 2 x 0 . 1 7 = 0 . 4 8 for the additive genetic variance in the difference score. Similarly, the environmental variance in the difference score is estimated to b e 0 . 5 4 + 0 . 6 4 - 2 0 . 1 8 = 0 . 8 2 . O n c e again, these estimates are very close to those obtained directly from the difference scores. T h e genetic correlation between sociability a n d impulsiveness is

rAii = VAii/(VAi

vAjy\

A similar expression yields the environmental correlation r

EWî/ =

/2

VEmj/(EmEmy .

Substituting o u r estimated genetic a n d environmental covariances in the a b o v e expression yields for the genetic correlation between sociability a n d impulsiveness rAij = 0 . 4 2 . F o r the environmental correlation w e obtain a value o f 0 . 3 2 . T h u s , if w e consider the t w o scales as they are measured, the phenotypic correlation between them is due to genetic and environmental causes. T h e genetic correlation between the scales is slightly greater than that due to the environment, but not markedly s o . Since the test scores are only estimates o f the subjects' true scores, it m a y be argued that the heritability o f the test scores, a n d the environmental correlation between them, is less than that o f the true scores, since errors of measurement will contribute to estimates o f E w . T h u s there m a y b e s o m e justification for correcting heritability estimates for unreliability o f measurement. Before making such adjustments, however, it is important to consider the context in which such adjustments are m a d e . First, as w e have already shown (Chapter 9 ) , the variance c o m p o n e n t s often used to yield estimates o f unreliability (i.e., persons items interaction components) are often as much genetic as they are environmental. Under these circumstances, it is wrong to assume that reliability corrections will only reduce estimates o f non-genetic v a r i a n c e . S e c o n d l y , before making a correction for unreliability, w e have to consider the purpose for which the actual test scores a r e being used. In so far as the test is being used as it stands for predicting the resemblance between relatives, the unadjusted heritability is the correct o n e for practical purposes. O n the other hand, there m a y b e certain processes in the population (of which m a t e selection is one) that depend not o n the test scores but o n the latent trait o f which the test is but an unreliable index. T h e

10.

Genetic and Environmental Covariance Between Traits

259

genetic consequences of m a t e selection, i.e. the correlation between the genetic deviations o f spouses, A (see Chapter 1 6 ) , will depend on the heritability of the latent trait and not on the heritability o f the raw scores. Using the estimates of internal consistency given a b o v e , and assuming that they indeed have no genetic c o m p o n e n t , we conclude that 5 4 % o f the environmental variation in sociability is "reliable" variation, and the remainder is due to error. For impulsiveness, about 3 8 % o f the estimate o f £ w is due to real long-term environmental effects and the remaining 6 2 % is due to errors o f measurement. Subtracting the fraction o f E w that is assumed to be error leads to estimates of the "real" environmental c o m p o n e n t of variance. Estimates of the genetic c o m p o n e n t are unchanged. T h e heritability estimates o f the true scores m a y n o w be computed using the original estimates o f VA and the corrected estimates o f E w , yielding estimated heritabilities of 0 . 6 1 for sociability and 0 . 6 0 for impulsiveness. T h e corrected estimates o f E w for the two traits m a y also be used to c o m pute an estimate o f the correlation between the environmental influences on the true scores. T h e corrected estimate o f r E W ;i is 0 . 6 6 . T h i s value is substantially higher than the uncorrected value and larger than the genetic correlation. If our unreliability correction is justified then the data imply that such long-term environmental effects as affect the traits o f sociability and impulsiveness exercise a significant joint influence on b o t h traits. O v e r a l l , when the effects o f the short-term errors o f measurement are subtracted, the greater part o f the reliable variation in b o t h traits, and in the c o m b i n e d extraversion score, is due to additive genetic factors. H o w e v e r , the genetic correlation between sociability and impulsiveness is significant but far from perfect. T h u s the genes create m u c h o f the c o m m o n factor underlying the c o m p o n e n t tests, but there is substantial genetic variation unique to the t w o traits, which justifies our regarding them as genetically distinct aspects o f personality.

1 0 . 3 THE COMPONENTS OF IMPULSIVENESS 1 0 . 3 . 1 Models for more than two variables T h e r e are limits to what can be said about the c o v a r i a n c e between two measures. A c o v a r i a n c e m a y be estimated, and the methods we h a v e described so far m a y be used to partition phenotypic c o v a r i a n c e into its genetic and environmental c o m p o n e n t s . A s long as we o n l y consider pairs o f measurements, however, there is n o w a y to partition their variances and covariance into c o m m o n and specific factors without placing unwarranted constraints o n the specific variances. T h e m e t h o d that we used in the analysis

260

Genes, Culture and Personality

of sociability and extraversion can easily be extended to the multivariate case (see e.g. Eaves and Eysenck, 1975, 1980), to yield estimates of genetic and environmental covariance between every conceivable pair of variables. While this approach tells us much about the sources of trait covariation in genetic and social terms, it does not address certain kinds of question that naturally arise when we consider multiple variables. For example, having decided that there is substantial genetic covariation between traits, we want to know whether there is a single set of genetic effects common to all measures, or whether a significant part of the genetic variation is specific to individual variables. As far as environmental effects are concerned, we should like to answer similar questions. Given estimates of the genetic and environmental covariances, it does not require much imagination to see that these might be factor-analysed using some of the standard methods to obtain estimates of factor loadings and specific genetic and environmental variances. In practice, it turns out to be not quite as simple, because estimates of genetic and environmental covariance matrices obtained from linear combinations of mean-product matrices are often not positive-definite nor even positive-semidefinite (see e.g. Loehlin and Vandenberg, 1968). Cantor et al. (1983) and Martin et al. (1984) show how maximum-likelihood estimates of genetic and environmental covariances can be obtained that do not suffer from this mathematical problem. However, the main issue is not mathematical, but hangs on what kinds of hypothesis we want to test given that we have multivariate data on twins that can be used to resolve the structure of genetic and environmental covariance. Some of these hypotheses cannot be tested if our ideal analysis is confined to factor analysis of genetic and environmental covariances. For example, Figures 10.1 and 10.2 show two models for the way in which common and specific genetic and environmental components may contribute to the covariance structure of multiple measures. In Figure 10.2 a common genetic factor and a common environmental factor contribute directly to the correlation between traits, but there is no common intervening pathway. Conceptually, it is as if the measured traits were the outcome of two distinct processes, one that was under genetic control (for example the biochemical pathways underlying "sensitivity" to the environment) and the other that reflected the covariance structure of the information which individuals acquire from their environment. With this model, the ratios of genetic factor loadings to environmental factor loadings will change from one variable to the next. In the model shown in Figure 10.1 there are no separate pathways from the underlying genetic and environmental factors to the variables. Rather, both genes and environment affect a single process P, which then has a joint effect on the multiple outcomes. In this case, the ratio of genetic factor loadings to environmental loadings will be constant for all variables. Conventional unconstrained factor analysis of

10.

Genetic and Environmental Covariance Between Traits

261

genetic and environmental covariances does not allow us to make this fundamental distinction between two quite different ideas about how genes and environment create the structure of phenotypic correlations. Rather than simply estimate genetic and environmental covariance matrices as a prelude to conventional factor analysis, we therefore try to specify certain explicit structural hypotheses in advance of the analysis, obtain maximum-likelihood estimates of the parameters, and conduct likelihood-ratio tests to discriminate between alternative hypotheses wherever possible. The approach is illustrated by data on four measures of impulsiveness.

10.3.2 The measurements and sample Eysenck & Eysenck (1977) have suggested that impulsiveness is capable of resolution into four correlated primary factors, which have been termed "impulsiveness in the narrow sense" (IMPN), "non-planning" (NONP), "risk taking" (RISK) and "liveliness" (LIVE). Questionnaire measurements of these factors are shown to correlate differently with questionnaire measurements of "psychoticism", "extraversion" and "neuroticism". Copies of an experimental questionnaire were mailed to adult twins on the Institute of Psychiatry Twin Register. Completed questionnaires were received from 587 pairs of twins, for whom the distribution of zygosity and sex is summarized in Table. 10.14. From the 52 impulsiveness items in the questionnaire 40 were selected that best represented the four component factors of impulsiveness. Three illustrative items from each scale are given in Table 10.15. The numbers of items contributing to scores on each factor were IMPN (12), RISK (10), NONP (12), LIVE (6). An angular transformation was applied to the raw scores for each factor to secure additivity, but the improvement was not marked, probably on account of the relatively small number of items contributing to each of the component scales.

Table 10.14 Composition of twin sample employed in the analysis of impulsiveness. Number of pairs

Male pairs (M) Female pairs (F) Unlike-sex pairs (OS)

Monozygotic (MZ)

Dizygotic (DZ)

144 233

52 83 75

—

262 Table 1 0 . 1 5

Genes, Culture and Personality Illustrative items for four impulsiveness factors.

Items Factor: impulsiveness (narrow sense) Do you often do things on the spur of the moment? Do you often get involved in things you later wish you could get out of? Do you usually think carefully before doing anything? Factor: risk Would you prefer a job involving change, travel and variety even though it might be insecure? Would you enjoy parachute jumping? Would you enjoy fast driving?

Keyed responses (Yes/No) Y Y

Y Y Y

Factor: non-planning Do you think an evening out is more successful if it is unplanned or arranged at the last moment? Would you make quite sure you had another job before giving up your old one? When you go on a trip do you like to plan routes and timetables carefully?

N

Factor: liveliness Do you usually make up your mind quickly? Are you usually carefree? Are you slow and unhurried in the way you move?

Y Y N

Y N

10.3.3 Data summary T h e mean squares and mean products between and within pairs were c o m puted for the four variables for each o f the four like-sex twin groups separately. For the m a l e - f e m a l e pairs the mean v e c t o r corresponding to the overall sex difference was also extracted from the intrapair variation. For each of the five twin groups the linear regression on age was partialled out o f the variation and c o v a r i a t i o n between pairs. T h e corrected mean squares and mean products are given in T a b l e 1 0 . 1 6 . T h e df take account o f the corrections made for age and sex. Recognizing the constraints imposed b y the s y m m e t r y o f the ten matrices, the basic data summ a r y comprises 1 0 0 statistics (ten c o v a r i a n c e matrices, each with 1 0 free statistics).

a

RISK 2

0.05049 0.12487 0.03785 0.03785 0.03427 0.10136 0.02636 0.02756 0.04840 0.08159 0.02951 0.01655 0.04154 0.07101 0.02986 0.02515 0.02009 0.07473 0.02242 0.02658

IMPN 1

0.12041 0.05049 0.03926 0.04673 0.08964 0.03427 0.01683 0.02782 0.10729 0.04840 0.03176 0.03082 0.09077 0.04154 0.02206 0.01277 0.07418 0.02009 0.01796 0.01044

df

231 1 2 3 4 142 1 2 3 4 81 1 2 3 4 50 1 2 3 4 73 1 2 3 4

0.03926 0.04456 0.03286 0.03286 0.01683 0.02636 0.05672 0.01235 0.03176 0.02951 0.06455 0.02149 0.02206 0.02986 0.04510 0.03040 0.01796 0.02242 0.05869 0.01858

NONP 3

Between-pairs mean-product matrices

0.04673 0.03785 0.17454 0.17454 0.02782 0.02756 0.01235 0.12959 0.03082 0.01655 0.02149 0.19970 0.01277 0.02515 0.03040 0.13465 0.01044 0.02658 0.01858 0.10020

LIVE 4

74

52

83

144

233

df 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 0.05344 0.01945 0.01381 0.02299 0.07445 0.09791 0.02829 0.02863 0.04787 0.01216 0.01178 0.02489 0.07478 0.01300 0.01298 0.00996 0.08180 0.03712 0.02634 0.02307

IMPN 1 0.01945 0.05960 0.01312 0.01472 0.03766 0.09791 0.02760 0.01528 0.01216 0.03764 0.00877 0.02283 0.01300 0.06791 0.01651 0.00583 0.03712 0.07394 0.01977 0.02212

RISK 2 0.01381 0.01312 0.03072 0.01545 0.02829 0.02760 0.05021 0.02387 0.01178 0.00877 0.03307 0.00579 0.01298 0.01651 0.02939 0.01024 0.02634 0.01977 0.04465 0.00214

NONP 3

0.02299 0.01472 0.01545 0.09528 0.02863 0.01528 0.02387 0.11747 0.02489 0.02283 0.00579 0.08473 0.00996 0.00583 0.01024 0.08983 0.02307 0.02212 0.00214 0.12258

LIVE 4

Within-pairs mean-product matrices

Age-corrected mean-product matrices within and between twin pairs for four impulsiveness scales/

Key to labels for primary factors: IMPN, impulsiveness in the "narrow sense"; NONP, non-planning: RISK, risk-taking; LIVE, liveliness.

D Z os

DZm

DZ f

MZm

MZ f

Twin type

Table 10.16

Genes, Culture and Personality

264 1 0 . 3 . 4 T h e model

The above analysis of impulsiveness and sociability showed that the pattern of individual differences in male and female monzygotic (MZ) twins and in like-sex and unlike-sex male and female dizygotic (DZ) twin pairs was consistent with a model that assumed that the variation and covariation of the two traits were due primarily to the additive effects of many genes and the effects of environmental influences that were largely specific to individuals rather than common to families. Furthermore, the consistency over sexes of particular estimates, and the ability of the model to encompass data on unlike-sex pairs without additional parameters, implied that the causes of variation in extraversion and its components did not depend substantially on sex. We suppose that the phenotypic variation for the four traits may be explained by a model invoking a single factor common to the four variables (impulsiveness in the broad sense) and components specific to each of the variables. However, by including twins in the study we are able to go beyond a simple treatment of the structure of phenotypic variation into an analysis of its causal basis. Thus, combining the simple causal model for impulsiveness described in Section 10.2 with the simple factorial model proposed by Eysenck & Eysenck (1977), we can discover whether the factorial unity of impulsiveness applies with equal force to both the genetic and environmental determinants of the trait. We assume that individual differences in impulsiveness are due to additive genetic effects with random mating (V A) and within-family ( E w ) environmental effects. For a single variable, we may write our expectation for the total phenotypic variance in terms of our simple model as VP = V A + E w

(cf. Chapter 4).

The basic model for individual differences can be extended in many ways for the univariate case, as we have already shown. For multiple variables, the same model can be used for the phenotypic covariance matrix, which is and now expressed as the sum of an "additive genetic" covariance matrix, a "within-family environmental" covariance matrix E w . We also assume that the genetic and environmental covariance matrices may themselves be decomposed in terms of the conventional factor model into one or more common factors and a number of specific variances: = E

A

= GG

+ E +

w 2

D

+

HH

2

+ E .

This model may be made more complex if the need arises. As long as we only have unrelated individuals, as is the case in most studies of individual differences, there is no hope of separating the genetic and environmental

10.

Genetic and Environmental Covariance Between Traits

265

components of phenotypic covariance. However, once we have kinship data, we can estimate genetic and environmental factors separately. In the twin data we have matrices of mean products within and between pairs for the different types of twins. The contributions of the genetic and environmental factors to the different matrices differ, so we may try to resolve genetic and environmental components of phenotypic covariation. For twins we have the following expectations: = 2(GG =

HH'

2

D )

+

2

+

HH

+ E ,

+

HH

+ E ,

+

HH

+ E .

2

+ E ,

= f (GG = I(GG'

2

+

D )

+

D )

2

2

2

The subscript denotes a matrix of mean products between pairs, W denotes a matrix of within-pair mean products. There are simple extensions of the expectations for the mean squares for a single variable given in Chapter 5. We have made no distinction between male and female twins, nor between like-sex and unlike-sex dizygotic twins, in writing the initial expectations, since our basic model assumes that the genetic and environmental components do not depend on sex. Allowing for sex limitation in gene expression in the multivariate model is quite straightforward. For the present case there are four variables. The factor model anticipates that there will be only one common factor. The matrices G and H will therefore be four-element column vectors of genetic and environmental loadings. 2 2 D and E will be 4 4 diagonal matrices containing the corresponding specific variances. Our first model will thus attempt to explain the 100 raw statistics by reference to 16 parameters. A further simplification is proposed. We begin by assuming that genetic and environmental effects on the phenotype are mediated through a common underlying process (cf. Figure 10.1). We thus constrain the genetic loadings to be a constant multiple of the environmental loadings thus: G

=

bH.

If this model fits then the scale factor b is related to the heritability of the 2 2 2 common factor through h » b /(l + b ). In summary, the simplest form of factor (b); the model has 13 parameters: four factor loadings ( H ) ; one scalar 2 2 four genetic specifics ( D ) ; and four environmental specifics ( E ) . In principle, further reductions might be possible by imposing constraints on the relative values of specific components, but this does not seem appropriate in the absence of any theoretical justification. A second model for the relationship between genes, environment and phenotype in a multivariate system is represented in Figure 10.2. In this case

Genes, Culture and Personality

266 GENES

ENVIRONMENT

FACTOR

TRAITS GENES ENVIRONMENT

SPECIFICS

Figure 10.1 Multivariate model when genes and environment operate through a common underlying variable.

GENES

ENVIRONMENT

FACTORS

TRAITS

GENES SPECIFICS

ENVIRONMENT

Figure 10.2 Multivariate model when genes and environment affect different pathways.

a c o m m o n set of genetic effects ( G ) and a c o m m o n environmental factor ( E ) each affect the multiple variables through different p a t h w a y s . Here it would be impossible to scale the genetic factor loadings to be constant multiples of the environmental loadings. Although we h a v e considered only the simplest experimental design and specified only the most basic g e n o t y p e - e n v i r o n m e n t a l factor models, the approach is general in that virtually a n y model we can write for a single variable can be cast in a form applicable to multiple variables, so that h y p o theses can be tested and parameters estimated, given adequate data.

10.

Genetic and Environmental Covariance Between Traits

267

10.3.5 Statistical method for testing the model Joreskog has developed a conceptual and statistical m e t h o d o l o g y for problems very similar to ours, (see e.g. J o r e s k o g , 1 9 7 3 , 1 9 7 8 ) . W e have used an approach that is v e r y similar to his, adapting it somewhat to the needs o f our particular class o f p r o b l e m . T h e approach is described fully b y M a r t i n and Eaves ( 1 9 7 7 ) . Generally, our data will consist o f k matrices o f mean products. W e m a y write S „ for the zth matrix, having JV, df. Given some model for the S „ we m a y c o m p u t e the expected values being positive-definite, for particular values o f the parameters o f the model. W h e n the observations are multivariate n o r m a l we m a y write the log-likelihood of obtaining the k observed independent S 2 as 1 L = -

k

NJloglE.I+triS^)]-

1

+ constant.

For a given model, we require the parameter estimates that maximize log L. Given maximum-likelihood estimates o f our parameters, we m a y test the hypothesis that a less restricting model (i.e. one involving m o r e parameters) 2 significantly improves the fit b y computing X = 2 ( L 0 - L a ) , where L 2 is the log-likelihood obtained under the restricted hypothesis (H x) and L 0 is the loglikelihood obtained under the less demanding hypothesis ( H 0 ) . T h e H 0 that we shall adopt in practice assumes that as m a n y parameters are required to explain the data as there are independent mean-squares and mean-products in the first place, i.e. { = S, for every i. In this case we have simply:

2

W h e n we h a v e k matrices the X has Vi kp(p + 1 ) - m df, where m is the n u m b e r o f parameters estimated under H 2 and is the number o f variables. T h e likelihood m a y b e maximized numerically using algorithms for unconstrained optimization as long as the model is parametrized in a form that always yields positive-definite expected c o v a r i a n c e matrices. In our application, this restriction is satisfied b y the following conditions: (1) we 2 2 estimate D and rather than D and E ; (2) the expected genetic and environmental c o v a r i a n c e matrices each parametrized as the products o f a (possibly different) matrix and its transpose; (3) the specific environmental c o m ponents are not zero. W e h a v e never encountered a case where the last condition is not satisfied; and we should not expect one to occur, since all variables h a v e s o m e specific environmental variation even if it is o n l y due to errors o f measurement.

Genes, Culture and Personality

268

In order that the estimates of a satisfactory model might be interpreted more rigorously, their covariance matrix is required. This is the inverse of the matrix of the second derivatives of the log-likelihood with respect to the maximum-likelihood estimates of the parameters. Joreskog (1973) gave formulae for the derivatives in problems involving single covariance matrices. We followed his approach, constructing first the matrix of second derivatives with respect to all the parameters, fixed and free, then striking out the rows and columns corresponding to fixed parameters, and finally combining the information on those parameters which are constrained to be equal. Martin and Eaves (1977) described the method for obtaining the covariances of the estimates in further detail.

10.3.6 Results Table 10.17 gives estimates obtained from fitting the simple model that assumed a single genetic factor with loadings proportional to those of a single common environmental factor. These loadings are scaled to reproduce the covariance matrices. The log-likelihood under this hypothesis was 3867.10 for 13 parameters. If we were to estimate the parameters of a model involving the maximum of 100 parameters (a perfect-fit solution) then we would obtain a value of 3919.37 for the log-likelihood. The goodness-of-fit test yields X%7 ** 104.54. Although this value is not significant (P = 0.097), the fit is relatively poor. Standard errors of the estimates are not cited at this stage because we are not satisfied with this model. We first consider some modifications of the basic model that might improve the fit. There are several possibilities. We could seek additional common factors. This would seem unwise with only four variables! We could seek explanations that involve effects other than additive genetic and within-

Table 10.17 Maximum-likelihood parameter estimates assuming identical structure and environmental factors. Genetic

Trait IMPN RISK NONP LIVE

Loading

0

0.115 0.114 0.085 0.089

(h\)

Environmental b

Specific )

0.128 0.143 0.105 0.197

Loading (e\)

0.142 0.141 0.105 0.111

Specific

0.186 0.150 0.135 0.249

Genetic loadings are a constant multiple (0.810) of the environmental loadings. Loadings are scaled to reproduce the phenotypic covariance matrix, not the correlation matrix. b Specific standard deviations, not variances, are tabulated.

10.

Genetic and Environmental Covariance Between Traits

269

family environmental effects such as more complex environmental effects or more subtle genetic effects. One possibility would be to relax the constraint that G - bH and allow the genetic and environmental loadings to vary independently. Such a model implies that genes and environment affect the measures of impulsiveness directly and independently (cf. the model in Figure 10.2) and are not mediated through some common underlying latent variable. With this model, a factor analysis of the phenotypic correlation matrix could never adequately represent the joint action of genetic and environmental effects because there is not just one underlying factor but two different factors, one of which is substantially genetically determined, and the other of which reflects the structure of environmental influences. When we fitted this model we obtained a value of 3870.40 for the log-likelihood, 2 giving a goodness-of-fit X of 97.94 for 84 df (P = 0.142). The three additional parameters led to a reduction in chi-square of 6.60 for 3 df (P = 0.086), suggesting a slight but not very significant improvement in fit. The slight, but not very striking, evidence of some differences between genetic and environmental factor structure led us to revert to the previous model for the factor structure, and we started to examine the specific variation. By a process of tentative model fitting to the data on sexes separately, but leaving out the unlike-sex pairs, we obtained an indication that, although the factor loadings seemed fairly consistent over sexes, the values obtained for the specific variances, especially the specific genetic variances, differed quite markedly between males and females. This suggested that the genetic determinants of trait-specific variation were different in the two sexes. If this were the case then we should expect the common factors to contribute to the covariation of male-female pairs, but we should expect the specific genetic variances to take different values in males and females and to make no contribution to the covariance of unlike-sex twins. Thus a final model was fitted that differed from the initial model in only the following features. Specific genetic variances were fitted that depended on sex, with the further specification that these were genetically quite distinct in the two sexes. This amounts to saying that the genetic component of the trait-specific variation can be best approximated by a model that assumes that quite different genes are expressed in males from those expressed in females (nonscalar sex limitation). The model is thus that described above, except that we have slightly different expectations for the opposite-sex (OS) pairs, as follows: EBOS = fGG

Ewos

2

2

+ i ( D M + D F) + H H ' + E ,

GG' + i(D

2

2

M

+ D P) + H H ' + E ,

where Dj^ and Dj denote the specific additive genetic variances for males and females respectively. In the expectation for like-sex pairs (above) we merely substitute D M for D in the males and D F for D in the expectation for female twins (cf. Table 5.11 for the univariate case).

Genes, Culture and Personality

270

Table 10.18 Parameter estimates for multivariate model for impulsiveness (standard errors in parentheses). Specific standard deviations Factor loadings Trait IMPN RISK NONP LIVE a

Genetic (c)

Genetic (h)

Environmental (e)

Male

Female

Environmental (n)

0.114 (0.008) 0.112 (0.008) 0.083 (0.006) 0.090 (0.009)

0.142 (0.010) 0.140 (0.010) 0.105 (0.008) 0.113 (0.011)

0.133 (0.015) 0.134 (0.016) 0.101 (0.012) 0.173 (0.191)

0.127 (0.021)

0.181 (0.009) 0.200 (0.009) 0.149 (0.007) 0.274 (0.011)

— 0.087 (0.018) 0.201 (0.026)

Parameter is fixed on the boundary.

T h e model n o w has four factor loadings, with one constant relating genetic and environmental factors, four specific environmental c o m p o n e n t s and eight specific genetic c o m p o n e n t s (four for each sex), making 1 7 parameters in all. T h e log-likelihood is n o w 3 8 7 5 . 1 2 , giving an overall X%3 = 88.50 (P = 0 . 3 1 9 ) indicating a g o o d fit and a marked improvement {X\ = 1 6 . 0 4 , = 0 . 0 0 3 ) on the original model that assumed that the specific genetic variances were the same in b o t h sexes. T h e estimates of the parameters of this model are given in T a b l e 1 0 . 1 8 . T h e standard errors of the estimates are also tabulated where appropriate.

1 0 . 3 . 7 Discussion T h e analysis suggests that the c o v a r i a n c e structure of impulsiveness is due to a single underlying factor, which is affected jointly b y genetic and environmental effects. B y showing that the genetic and environmental loadings are proportional to one another, we have in effect showed that the ratio o f variation due to c o m m o n genetic factors to that due to the c o m m o n environmental factor is consistent over all variables. T h u s there is a c o m m o n factor, which we m a y call impulsiveness in the b r o a d sense, whose heritability is a simple function of the ratio b o f the genetic and environmental loadings. T h e fact that b differs significantly from zero (6 = 0 . 8 0 ) indicates that the genetic loadings are jointly significant and justifies our attempts to estimate the p r o portion of the c o m m o n - f a c t o r variance that is due to genetic factors. If we write g for a n y one of the four genetic factor loadings and e for the

10.

Genetic and Environmental Covariance Between Traits

271

corresponding environmental loading then we have g = eb. Since the model fits, we may estimate the "heritability" of the common factor from 2

h

2

= b /(l

2

+ b ).

common-factor Substitution for b gives the estimate of the proportion of the variance that is genetically determined as 0.39. This value does not require any correction for unreliability, since we presume that sampling error will contribute only to the specific components of variation in the four measurements. Similarly, we may use our parameter estimates tö asses the relative contribution of genetic and environmental differences to the specific variation of the four measurements. We have established that the sexes differ in the genetic mechanism responsible for specific variation, so we are compelled to give separate estimates of the "specific heritabilities" for each sex. 2 2 Writing d { for a typical genetic specific variance and s i for the corresponding environmental specific, we have 2

2

hj = d /(d

2

+ s ).

The estimated heritabilities of the trait-specific variances are given in Table 10.19. With the exception of the risk factor (for which males show no specific genetic variation — in contrast with the females, who show significant genetic specific variation), the values are comparable to those obtained for the common factor. However, errors of measurement do not contribute to environmental variation in the common factor (given independent errors), but are expected to contribute to the specific environmental variation. Since the analysis is based on scores transformed to angles, we may approximate the error variance for each scale from the theoretical error of transformed proportions (see e.g. Snedecor and Cochran, 1980). These values will enable us to estimate how much of the trait-specific environmental variance is due to measurement error. For a scale consisting of items of equal difficulty and given local independence the theoretical error variance takes the value (4n) " If the items of a scale are not all equally difficult then this estimate of error will be larger than the true value. Table 10.19 Proportion of variation specific to four impulsiveness scales attributable to genetic factors. Trait

IMPN RISK NONP LIVE

Females

Males

0.350 0.329 0.332 0.144

0.380 0.000 0.332 0.146

272

Genes, Culture and Personality

Table 10.20 Analysis of the contribution of measurement error to specific variation.

Estimated contribution to specific environmental variance

Träte IMPN RISK NONP LIVE a b

Number öf items 12 10 12 6

"True" specific "Error" environmental 0 variance variance

Proportion due to "error"

Females

Males

0.63 0.63 0.95 0.56

0.60 0.54 0.91 0.48

0.57 0.00 0.88 0.55

0.012 0.015 0.001 0.033

0.021 0.025 0.021 0.042

The error variance is estimated as (4M) Cf. uncorrected values in Table 10.19

1

"Heritability" of specific variance corrected for measurement b error

(see text).

Table 10.21 Summary of the relative contributions of common and specific genetic and environmental effects to the total variation in each of the components of impulsiveness. Proportion due to genetic effects Sex

Trait

Female

IMPN RISK NONP LIVE IMPN RISK NONP LIVE

Male

a

Proportion due to environmental effects

Common

Specific"

Total

Common

Specific

Total

0.155 0.139 0.138 0.064 0.158 0.173 0.146 0.059

0.211 0.218 0.219 0.101 0.245 0.000 0.231 0.094

0.366 0.357 0.357 0.165 0.404 0.173 0.377 0.153

0.241 0.199 0.203 0.238 0.197 0.272 0.158 0.296

0.392 0.444 0.440 0.596 0.399 0.554 0.465 0.551

0.633 0.643 0.643 0.835 0.596 0.827 0.623 0.847

Error variation has not been deducted from the contribution of specific environmental factors.

T a b l e 1 0 . 2 0 shows estimates o f the specific environmental variation for each trait, with the appropriate theoretical errors for scales o f the corresponding length. T h e difference between the two sets o f estimates is an estimate of the "true" specific environmental variation, which is due to factors other than errors o f measurement. In every case rather m o r e than half o f the measureable specific environmental variation within families seems to b e attributable to errors o f measurement. Indeed, in the case o f the non-

10.

Genetic and Environmental Covariance Between Traits

273

planning factor we conclude that virtually all the detectable specific environmental variation is due to sampling error in the scores. Finally, T a b l e 1 0 . 2 1 gives a s u m m a r y o f the contributions o f the different sources o f variation to the four measurements in each sex derived from the parameters in T a b l e s 1 0 . 1 9 and 1 0 . 2 0 . For each sex, the phenotypic variance VPi for the zth trait was calculated b y substituting the appropriate parameter estimates in V „ - (g? + df) + ef + sf. 2

T h e contributions o f each o f the c o m p o n e n t s gf, d if ej and sj to the phenotypic v a r i a n c e VPi are then c o m p u t e d as ratios for each sex and variable separately. M a n y other s u m m a r y statistics could b e derived from the estimates in T a b l e 1 0 . 1 8 , including traditional heritability estimates for the individual variables. Adding together the contributions from the genetic and environmental c o m m o n factor yields the familiar c o m m u n a l i t y estimate for each v a r i a b l e . Adding the genetic contributions due to c o m m o n and specific variance for each variable in turn, w e h a v e the usual heritability estimate applicable to each variable as it would b e derived in a n y equivalent univariate analysis o f the individual scales. A p a r t from the substantive findings, the analysis o f impulsiveness illustrates h o w the maximum-likelihood m e t h o d can b e applied to answer questions about w h y multiple variables are correlated. T h e analysis shows that the phenotypic correlations between measures o f impulsiveness are due to genetic and environmental factors. M o r e than that, however, it seems as if the effects o f genes and environment are "channeled" through s o m e c o m m o n p a t h w a y before they affect the different facets o f impulsiveness. A l t h o u g h the correlation between traits appears to b e caused b y the same genes in b o t h sexes, there is a suggestion that the trait-specific c o m p o n e n t s o f impulsiveness m a y be due to different genetic effects in males and females. Even this simple example therefore illustrates s o m e o f the power and flexibility o f the method.

10.4 T H E C O V A R I A N C E S T R U C T U R E O F N E U R O T I C I S M In the same questionnaire used to develop subscales o f impulsiveness, 7 6 items were embedded that were drawn from scales that had been used at some time or other in the measurement o f neuroticism and related traits. T h e responses o f the 1 1 7 4 individual twins to these items were correlated and factored using iterative principal-axis factor analysis. M a l e s and females were c o m b i n e d in the analysis. W h e n a single c o m m o n factor w a s fitted it was clear that a general factor o f "neuroticism" could be identified, but that a

274

Genes, Culture and Personality

multiple-factor solution w a s needed to account adequately for the pattern o f interitem correlations. Seven correlated factors could be identified, covering the spectrum o f neurotic responses, s o m e o f which might be better described as "psychotic" since they included items relating to depression and perceived threat from others. T h e seven factors m a y b e tentatively identified in terms of their item content as "depression", "worry", "insomnia", "paranoia", "inattention", "shyness" and "psychosomatic s y m p t o m s " . Factor scores were derived for each subject and employed as the basis for subsequent genetic model fitting.

10.4.1 Data summary T a b l e 1 0 . 2 2 gives the matrices o f mean squares and m e a n products between pairs and within pairs for each group o f twins in the sample. T h e linear trend of factor scores o n age w a s partialled out o f the mean products between pairs.

1 0 . 4 . 2 Univariate analysis Before attempting a multivariate analysis o f the c o v a r i a t i o n between factors, we summarize the results o f a univariate analysis, which m a y help us in building an appropriate multivariate model. T h e results o f fitting the V A , £ w model to the mean squares for each o f the variables are given in T a b l e 1 0 . 2 3 . T h e fit is g o o d for t w o factors ("insomnia" and "paranoia"), barely adequate for "depression" and "inattention", and b a d for the remaining three factors. Allowing for the effects o f sex-limited gene expression and sex differences in the within-family environmental c o m p o n e n t improved the fit significantly for "depression" and "paranoia", but the overall fit of the model w a s still p o o r for the "worry", "shyness" and "psychosomatic s y m p t o m s " factors (Table 1 0 . 2 4 ) . W e attempted to explain the failure of the additive model for the t w o worst cases ("shyness" and "psychosomatic symptoms") in terms o f competitive interactions between twins. In view o f the evidence for sex differences, w e excluded the unlike-sex pairs and tested for the improvement in fit attributable to competition for males and females separately. In three out o f four cases ( T a b l e 1 0 . 2 5 ) the inclusion o f competitive effects in the model led to a significant improvement in fit to the observed mean squares and resulted in non-significant residual effects. T h e results in T a b l e 1 0 . 2 5 have to be treated with certain skepticism for t w o reasons. First, we k n o w (see Chapter 4) that the resolution o f

Table 10.22

Mean squares and mean products for seven neuroticism factors.

Between MZ female pairs, df = 231: Depression 1.190 Worry 0.464 1.094 Insomnia 0.522 0.395 Paranoid 0.469 0.227 Inattention 0.631 0.416 Shyness 0.172 0.293 Psychosomatic 0.711 0.546 >, df = 233: Depression 0.498 Worry 0.112 0.481 Insomnia 0.162 0.116 Paranoid 0.122 0.070 Inattention 0.172 0.108 Shyness 0.075 0.105 Psychosomatic 0.198 0.131 Between MZ male pairs, df = 81: Depression 1.116 Worry 0.716 Insomnia 0.404 Paranoid 0.556 Inattention 0.761 Shyness 0.203 Psychosomatic 0.356 Within MZ male pairs, df = 83: Depression 0.307 Worry 0.136 Insomnia 0.868 Paranoid 0.096 Inattention 0.141 Shyness -0.026 Psychosomatic 0.357

1.152 0.343 0.445 0.126 0.755

0.886 0.380 0.136 0.289

1.078 0.093 0.651

0.945 0.140

1.509

0.521 0.902 0.056 0.043 0.225

0.399 0.102 0.230 0.084

0.403 -0.004 0.088

0.412 0.064

0.479

1.432 0.209 0.391 0.723 0.291 0.428

0.792 0.388 0.476 0.044 0.323

0.891 0.663 0.119 0.286

1.424 0.159 0.329

0.630 0.171

0.602

0.297 0.772 0.012 0.078 0.050 0.074

0.418 0.013 0.074 0.063 0.130

0.282 0.087 0.008 0.038

0.354 0.159 0.141

0.281 -0.010

0.304

0.944 0.304 0.228 0.059 0.452

0.951 0.315 0.131 0.296

0.745 0.009 0.270

0.857 0.167

0.850

0.751 0.273 0.155 0.064 0.306

0.570 0.118 0.115 0.228

0.597 0.040 0.253

0.774 0.247

0.749

0.891 0.308 0.351 -0.017 0.321

0.566 0.267 0.004 0.240

0.800 -0.099 0.200

0.594 0.082

0.493

0.675 0.108 0.258 0.123 0.264

0.392 0.196 0.172 0.081

0.879 0.027 0.277

Between DZ female pairs, df = 142: Depression 0.869 Worry 0.285 0.864 Insomnia 0.292 0.200 Paranoid 0.420 0.223 Inattention 0.350 0.171 Shyness 0.113 0.297 Psychosomatic 0.373 0.386 Within DZ female pairs, df = 144: Depression 0.803 Worry 0.220 0.573 Insomnia 0.156 0.773 Paranoid 0.202 0.033 Inattention 0.296 0.141 Shyness 0.114 0.248 Psychosomatic 0.315 0.198 >, df = 50 Between DZ male Depression 0.806 Worry 0.228 0.749 Insomnia 0.503 0.176 Paranoid 0.384 0.245 Inattention 0.447 0.122 Shyness -0.114 0.109 Psychosomatic 0.421 0.188 Within DZ male pairs. df = 52: Depression 0.628 Worry 0.315 0.842 Insomnia 0.227 0.229 Paranoid 0.152 0.239 Inattention 0.366 0.451 Shyness 0.065 0.200 Psychosomatic 0.236 0.412

0.642 0.043

0.642

P%

xi

h2

VA EW

Factor

Table 10.23

5-10

15.2

2-5

17.6 30-50

8.2

0.35

0.516

0.434

0.466

0.43

0.272

0.0324

0.324

0.41

Insomnia

Worry

Depression

10-20

12.6

0.45

0.362

0.291

Paranoia

5-10

14.5

0.45

0.405

0.336

Inattention

V

EB> A

E W, V A,

57.67***

75.35***

—

—

6

47.42***

—

83.99***

—

6

97.73***

49.58***

40.95***

42.50***

—

5

60.97***

—

165.35***

5

60.97***

V D 49.58***

-81.90

108.31***

Female and male and opposite sex: E W, E B E W, E B

E

£w>

*>

V

A

62.04***

69.78***

—

—

8

135.90***

62.04***

69.78***

—

—

8

110.94***

7

64.41***

49.45***

E W, V A, V D 4 9 . 4 5 * * *

34.33*** —

47.97*** 150.95***

— -68.66

7

64.41***

Significance levels: ** < 0 . 0 1 , *** < 0 . 0 0 1 .

Table 14.5 Parameter

Parameter estimates from Australian conservatism data. Estimate ± s.e.

E BF

41.5 62.1 43.4 49.4 34.6

±6.3 ±3.3 ±1.7 ±7.5 ±6.1

E

37.1

±5.0

VA E

WM

E

WF

E

BM

B FM

kLies females

4.40 0.27 ± 0 . 0 4 0.35 ± 0 . 0 5

14.

Twin Studies of Social-A ttitude Dimensions

363

In a n o t h e r m a j o r respect, the results for the larger Australian sample are different. W h e t h e r unlike-sex pairs are included in the analysis or omitted from it, there is evidence o f heterogeneity between the parameters in the t w o sexes. A single estimate o f VA c a n explain the genetic c o m p o n e n t o f variation in males and females, but the estimates o f the t w o environmental c o m p o n e n t s differ significantly between sexes. T h e final parameter estimates are given in T a b l e 1 4 . 5 , and the estimated proportional contributions o f the three sources to variation in conservatism scores are summarized, for b o t h sexes, in T a b l e 1 4 . 6 . T h e environmental variation within families is divided into long-term effects and the effects o f short-term (three-month) unreliability. M o s t o f the environmental variation within pairs is due to long-term effects. T h e proportions o f variance attributable to the main sources o n l y differ slightly from those obtained in L o n d o n . T h e contribution o f £ B is about the same, the average contribution o f genetic effects is about 3 0 % rather than 4 0 % , and the overall contribution o f the environment within families is increased from a b o u t 3 0 % in the L o n d o n sample to approximately 4 0 % in the Australian data. In m a n y important resjpects the results from the m u c h larger Australian study replicate the essential features o f the smaller British study. A s with personality, so with conservatism, the larger study confirms most aspects o f the smaller o n e , but adds certain glosses. In the case o f the conservatism dimension o f social attitudes b o t h studies c o n c u r in showing a m a j o r c o m p o n e n t o f the family environment (or assortative mating) in twin resemblance. S u c h a finding is gratifying because it confirms that our models and methods are perfectly capable o f detecting important environmental effects. Furthermore, given large enough samples, the results o f our analyses replicate v e r y well across different Western populations. Consistent similarities appear for the same trait. Consistent differences appear for different traits.

Table 14.6 Sources of variance (%) for age-corrected conservatism scores. Females error

Males 18

9

36 VA

individual environment 35

41 18

"I total genetic

27 | 49 14

. assortative mating J 29^ - family environment

- 32

27

| 38

J

_11 32

15

""""

21

.

Genes, Culture and Personality

364

T h e detection o f a genetic c o m p o n e n t of variation in social attitudes surprised us at first, and it will p r o b a b l y surprise others t o o . T h e fact that the social-attitudes data reveal a significant E B c o m p o n e n t , while the extraversion and neuroticism scales do not, vindicates the marriage between large genetically informative samples and a few well-defined variables in replicated studies. T h e variables for which the a priori expectation of cultural determination is greatest are those for which the twin study detects apparent cultural inheritance. T r a i t s for which it is less easy to visualize mechanisms of social interaction between relatives (neuroticism, for example) are those for which the apparent effects of social interaction are smallest. T h e genetic models that result represent an important heuristic for the interpretation of h u m a n differences in personality and social attitudes. T h e results lead to further predictions. Because o f the large samples involved, we h a v e a fairly g o o d idea of h o w results should l o o k in other types of study that would provide further tests of our models. For example, if we are correct about the lack of a social-environmental effect on personality measures then we should predict that the correlations between unrelated individuals reared together should be approximately zero. T h e correlation for separated m o n o z y g o t i c twins should be not significantly less than the correlations for twins reared together. In contrast, for conservatism scores we might expect correlations of 0 . 3 - 0 . 4 between unrelated individuals reared together under the environmental model, and the correlations o f separated m o n o z y g o t i c twins to b e about half those o f twins reared together.

14.3

THE CAUSES OF LONGITUDINAL CONSISTENCY

IN

CONSERVATISM

T w o of the L o n d o n studies involving the P O I and the W i l s o n - P a t t e r s o n conservatism scale yield a subset o f twin pairs for w h o m b o t h measures o f conservatism can b e derived. T h e interval between tests was approximately three years, so we m a y analyze the genetic and environmental basis of consistency over instruments and occasions of measurement. T h e study will provide a direct test o f whether or not the t w o instruments are really identifying the same genetic and cultural effects and will determine whether individual changes in the environment over a three-year period, and their interaction with genotype, are a m a j o r c o m p o n e n t of individual profiles o f attitude change. O u r analysis focuses o n l y on those twin pairs for w h o m complete data are available on b o t h o c c a s i o n s . T h e measurements made on the two o c c a s i o n s are regarded as two distinct variables, which m a y be correlated genetically or environmentally. T h e data are summarized, for each group o f twins, b y a

14.

Twin Studies of Social-Attitude Dimensions

365

Table 14.7 Twin covariances and correlations for repeated measures of 0 conservatism. Twin 1 Occasion

Twin 2 Occasion

df

1

2

1

2

MZ f

181

MZm

53

DZ f

97

DZm

21

D Z mf

39

69.90 0.55 0.59 0.43 61.92 0.63 0.74 0.50 119.27 0.64 0.58 0.47 123.21 0.61 0.31 0.26 86.13 0.53 0.18 0.29

3.91 0.72 0.49 0.64 4.17 0.71 0.48 0.55 6.33 0.81 0.45 0.47 6.52 0.93 0.21 0.10 4.52 0.83 0.28 0.34

45.27 3.80 84.42 0.60 50.63 3.50 76.12 0.50 60.87 3.89 91.96 0.74 32.69 1.95 89.54 0.53 15.23 2.32 84.64 0.57

3.03 0.46 4.64 0.71 3.40 0.40 3.79 0.74 5.01 0.42 6.96 0.97 2.33 0.08 4.07 0.65 2.18 0.25 4.27 0.66

Twin type

a

Correlations are given in the lower triangle.

4 x 4 c o v a r i a n c e matrix in which the variables are occasions classified b y first and second twin. T h e r a w c o v a r i a n c e matrices, corrected for age, are given in T a b l e 1 4 . 7 . T h a t the numbers are smaller than in previous analyses of conservatism is due to the fact that m a n y twins on the registry did not return b o t h questionnaires. T h e differences in variance between first and second occasions are merely a reflection o f the fact that different tests and transformations o f the data were used on the two o c c a s i o n s . T h e basic model for the c o v a r i a n c e structure o f the repeated measures assumes that differences in conservatism are ultimately caused b y additive genetic differences, within-family environmental effects and between-family environmental effects. W e denote the additive genetic variance on the first and second occasions b y V A 1 and VA2 respectively. T h e genetic c o v a r i a n c e between o c c a s i o n s is V A 1 2. W e m a y define similar parameters for the environmental variances and c o v a r i a n c e s within families, E W 1, E W 2, E W 1 2, and between families, Em, E B 2, E B 1 2.

Genes, Culture and Personality

366

Table 14.8 Expected covariances for MZ (upper triangle) and DZ (lower triangle) twins measured on two occasions. Occasion 1 Twin 1 Twinl Occasion Twin 2 Twin 1 Occasion Twin 2

V A1 + E W1 + E B1 1 ± V A1 + E B1 V A 12 + E W 21 + E B 2 J VA12 + £ ß l 2

12

Occasion 2

Twin 2

Twin 1

Twin 2

V A1 + E B1

V A 12 + E W 21 + E B

V A1 + E W1 + E B1 ^ V A 12 + E B 12

V A 12 + E B 12 V A2 + E W2 + E B2

YA12 + ^W12 + £ ß l 2

2^A2 + ^B22 +

12

V A 12 + E B

12

V A 12 + E W 21 + E B V A2 + E B2

12

E

B2 + ^W2

T h e expected variances and covariances for M Z and D Z twins for the t w o occasions are given in T a b l e 1 4 . 8 . T h e maximum-likelihood m e t h o d (see Chapter 10) w a s used to obtain estimates o f the c o m p o n e n t s o f variance and c o v a r i a n c e from the observed c o v a r i a n c e matrices. T h e correlation in the between-family environment over occasions approaches the upper b o u n d of unity for these data, so we substituted the term ( E B 1, EB2)V2 for E B 12 in the expectations for the c o v a r i a n c e s . T h e chisquare for testing the goodness o f fit o f the model was 4 4 . 8 for 4 2 df ( 0 . 3 < P < 0 . 4 ) , confirming that the model that fitted each of the separate twin studies can also account adequately for the joint pattern o f c o v a r i a t i o n in the analysis of repeated measures. T h e maximum-likelihood parameter estimates are given in T a b l e 1 4 . 9 . T h e cross-temporal stability of the causes of variation in conservatism is greatest for the between-families environmental c o m p o n e n t o f variance ( r B 12 = 1 ) , and lowest for the within-families environmental c o m p o n e n t ( r w i2 = 0 . 2 7 ) . T h e correlation in genetic effects over time, r G 1 2, is 0 . 7 2 . T h u s , in spite of the fact that attitudes are c o m m o n l y assumed to be v e r y labile, certain crucial determinants seem to exercise a consistent effect o v e r

Table 14.9 Parameter estimates for repeated measures of conservatism different scales. Variances Occasion 1

Occasion 2

Cross-temporal correlation

VA EW

41 ± 1 1

0.39

±0.09

0.72

30±3

0.28

±0.03

0.27

EB

16 ± 1 0

a

Parameter fixed on upper bound.

0.10 ± 0 . 0 8

14.

Twin Studies of Social-Attitude Dimensions

367

time, even when the measurements taken on the t w o occasions differ greatly in their format and content. T h e items o f the P O I are c o m p l e x statements, requiring a response on a five-point scale. T h e items o f the W i l s o n - P a t t e r s o n scale are single words, to w h i c h response is on a simpler three-point scale (see Chapter 1 2 ) . Nevertheless, the o n l y effects for which the cross-temporal correlation is small are the environmental influences within families, with which errors o f measurement are confounded. In so far as individuals' overall conservatism scores change with time relative to their peers, therefore, the main causes o f change are the unique experiences o f the individual, which are not shared, even with a c o t w i n . T h e fact that the genetic correlation across o c c a s i o n s is large (not significantly less than unity but significantly greater than zero) implies that the expression o f genetic factors that affect conservatism is r e m a r k a b l y consistent across time and questionnaires. T h e t w o conservatism scales are assessing the same genetic effects. Similarly, the correlation in the between-family environmental effects w a s unity. T h e shared environmental experiences o f twins therefore exert a long-term effect on conservatism that is consistent o v e r the t w o types of measurement. T h e r e is n o evidence that shared experiences o v e r the threey e a r period in question caused some pairs o f twins to increase their conservatism scores and others to respond in a m o r e radical direction on the second o c c a s i o n . Furthermore, since the genetic effects are virtually the same on the t w o occasions, there is n o evidence that a n y cultural effects intervening between the t w o o c c a s i o n s had changed the ranking o f genotypes with respect to the trait. T h a t is, there is n o evidence of genetic control o f the direction o f attitude change o v e r the three-year period in question. If different genotypes had responded differently to the two instruments, o r had responded differently to the events o f the intervening years, then the genetic correlation w o u l d h a v e been significantly less than unity. T h e longitudinal study shows that genetic effects and environmental differences between families h a v e a consistent and lasting (i.e. over three years) effect on the overall tendency o f individuals to endorse conservative or radical attitudes. T h e s e effects are not altered or eradicated b y such normal cultural changes as occurred in the three-year period o f follow-up. T h i s finding does not m e a n that attitudes do not change, but rather that there are effects o f genotype and the family environment that persist in society in spite of short-term changes. A n alternative interpretation, equally consistent with the twin data, is that the genetic effects o f assortative mating affect those aspects o f social attitudes that display the greatest long-term consistency. A s far as individual environmental experiences are concerned, we find that they could explain a b o u t 3 5 % o f the total variation in c o n servatism on each o c c a s i o n . T h e long-term consistency o f these effects is r e m a r k a b l y l o w . T h u s the differences that we see, even within identical twins reared together, are substantial and extremely labile. T h e low

368

Genes, Culture and Personality

correlation o f the within-family environmental effects o v e r time implies that it is a matter o f c h a n c e and individual experience which identical twin is m o r e conservative than his cotwin on each o c c a s i o n . In so far as conservatism is a reflection o f unique experiences o f the individual, these effects are short-lived and c a n n o t b e shown to extend over a three-year period. If the unique experiences o f the individual had a lasting effect then w e should expect the correlation o f within-family environmental effects to b e m u c h greater than zero.

14.4 ANALYSIS OF PRIMARY SOCIAL-ATTITUDE FACTORS Although m u c h of the variation in attitudes is explained b y the first t w o principal c o m p o n e n t s , factor analysis reveals a number o f p r i m a r y factors that are stable over sexes and populations. Feingold (1984) identified five correlated p r i m a r y factors in the P O I in British and U S samples for which there was reasonable consistency o v e r sexes. T h e factors were identified b y item content to b e "authoritarianism", "religion", "socialism", "prejudice" and "permissiveness". Items having salient loadings on each o f the primary factors are listed in T a b l e 1 4 . 1 0 . T h e s u m m a r y statistics for the British twin sample for the five p r i m a r y factors are given in T a b l e 1 4 . 1 1 . M o d e l s were fitted b y the maximum-likelihood method, using the General Linear Interactive Modelling ( " G L I M " ) program (Neider, 1 9 7 5 ) . Results for models with n o sex-limited genetic or environmental effects are summarized in T a b l e 1 4 . 1 2 . M o d e l s that excluded genetic effects did not fit the data on a n y o f the variables. M o d e l s that included b o t h sources o f environmental variation and genetic effects gave a g o o d fit to four out of the five factors. N o n e o f the three models could explain the data on "prejudice". T h e "authoritarianism" and "religion" factors showed statistically significant estimates o f the betweenfamilies environmental c o m p o n e n t . P a r a m e t e r estimates under the bestfitting model, assuming no sex limitation, are given in T a b l e 1 4 . 1 3 . T h e results of allowing for sex differences in genetic and environmental effects (cf. C h a p t e r 4) are seen in T a b l e 1 4 . 1 4 . O n c e again, the purely environmental explanation failed to account for variation in four out of five factors. "Religion" formed the o n l y exception. A s before, the data on "authoritarianism" and "religion" were explained far better b y models that allowed for b o t h genetic and cultural c o m p o n e n t s of twin resemblance. Similarly, when allowance is made for sex differences in the expression o f genetic and environmental differences, the data o n "prejudice" require a joint genetic and cultural explanation. T h e parameter estimates for the t w o p r i m a r y factors for which sex-limited effects i m p r o v e d

14.

Twin Studies of Social-Attitude

Table 14.10

369

Dimensions

Items loading on the five primary attitude factors.

Item

Primary factor

6 10 13 16 18 28 29 34 42 47 51

Authoritarianism Peace, not national sovereignty Flog violent criminals My country right or wrong Obligation to family Abolish barbaric death penalty Compulsory military training Flog sex criminals Conscientious objectors traitors Retain independence in world organization Treatment of criminals too harsh Life short and to be enjoyed

-0.41 0.69 0.44 0.37 -0.59 0.53 0.74 0.34 0.57 -0.61 0.34

14 17 23 25 31 33 36 39 45 48 53 56 59

Religion Good enough life without religion No survival after death Sunday observance old-fashioned Acceptance of Church's teachings God an invention of human mind Church should increase influence Religious people hypocrites Religion civilization's only hope Compulsory religious education Church main bulwark against evil Christ divine Universe created by God Faith in supernatural power

-0.56 -0.61 -0.43 0.33 -0.65 0.64 -0.36 0.68 0.58 0.47 0.70 0.74 0.48

4 24 52

Socialism Introduce socialism Capitalism immoral Occupation better than war

2 15 38 44 54 55

Prejudice Coloureds innately inferior Coloureds shouldn't be foremen over whites Don't help Asian refugees Make discrimination illegal Racial segregation Punish homosexuals

0.50 0.46 0.52 -0.49 0.60 0.38

8 12 27 37 49 51

Permissiveness Make divorce easier Trial marriage Encourage free love Extramarital sex wrong Travelling without a ticket Life short and to be enjoyed

0.38 0.58 0.48 -0.47 0.32 0.33

Item loading

0.43 0.45 0.37

0.44

0.48

0.44

0.54

0.74

0.69

r 1.329 0.277 1.389 0.374 1.095 0.391 1.491 0.483 1.085 0.528 1.120 0.650

ms r

0.27

0.35

0.51

0.47

0.58

0.66

Religion

0.969 0.344 0.894 0.382 0.819 0.482 1.050 0.476 0.854 0.417 1.274 0.581

ms r

0.37

0.22

0.38

0.26

0.40

0.48

Socialism

1.122 0.301 0.931 0.353 1.139 0.398 0.818 0.526 1.010 0.647 1.173 0.802

ms

r

0.19

0.22

0.22

0.48

0.45

0.61

Prejudice

All mean squares are corrected for the effect of age, and the DZ opposite-sex pairs are corrected for the main effect of sex.

1.234 0.228 1.569 0.235 1.473 0.442 1.298 0.507 1.444 0.507 1.201 0.465

324 325 119 120 193 194 58 59 59 59 65 65

Between MZ f Within MZ f Between M Z m Within M Z m Between DZ f Within DZ f Between D Z m Within D Z m Between D Z f m Within D Z ^ Between D Z mf Within D Z mf

a

ms

df

Authoritarianism

Factor

Mean squares between and within twin pairs and correlations for primary-factor scores. 0

Relationship

Table 14.11

1.213 0.245 1.146 0.296 0.834 0.406 0.722 0.503 0.958 0.632 1.129 0.497

ms

r

0.39

0.21

0.18

0.35

0.59

0.66

Permissiveness

14.

371

Twin Studies of Social-Attitude Dimensions

Table 14.12 Chi-square values for models fitted to five primary factors. Factor df

Authoritarianism

Religion

Socialism

Prejudice

Permissiveness

10 E w, E B VA,EW 10 VA, E w , E B 9

55.68*** 17.29* 9.32

36.12*** 20.09* 14.07

18.78* 11.83 9.71

49.25*** 24.07** 23.31**

56.54*** 11.33 11.32

Model

* P < 0 . 0 5 ; ** P < 0 . 0 1 ; *** P < 0 . 0 0 1 .

Table 14.13 Sources of variance for best fitting model to five primary factors (%)• Factor Authoritarianism

Religion

51 27 22 9.32 9 0.3