Keywords

JEL Classifications

While defining itself as the study of genetic influences on behaviour, behavioural genetics has been mainly concerned with demonstrating and quantifying the contribution of genetic variation to variation in human behavioural traits. As such, it contrasts with the related field of evolutionary psychology that attempts to understand how some behavioural traits common to all humans have been shaped by evolution.

The large and growing literature on the impact of genetic variation on behaviour leaves no room for doubt that genetic endowment is an important influence on a surprisingly wide range of behaviours. Behavioural genetics has relied mainly on the study of relatives with different degrees of relatedness or adoption to estimate the contributions of genetic variation and shared family environment to explaining crosssectional variation in behavioural characteristics. More recently, behavioural geneticists have been extending their methodology to use relational studies to examine the covariation of different behavioural traits, and molecular genetic methodologies to trace the sources and causes of genetically induced differences in behaviour.

Below I give a brief introduction to the mechanics of heredity. This is a necessary introduction to the methods of behavioural genetics, which I explain next.

Mechanics of Heredity

The human genetic code is contained in 23 pairs of chromosomes made up of deoxyribonucleic acid or DNA. A DNA molecule consists of two backbone strands that are held apart by molecular pairs of four bases. A sequence of these four chemicals along one of the backbone strands encodes the plans for the different proteins from which our bodies are made. Other parts of the code are thought to control when proteins are created and in what quantities. There are about three billion base pairs on just one set of 23 chromosomes. A sequence of base pairs that codes the information for a protein or some other function is called a ‘gene’.

Of the three billion base pairs all but about three million are the same in all humans. Where base pairs differ it is said that a polymorphism exists. When a gene contains one or more polymorphic base pairs there will be different versions of the gene. Different versions of the same gene are referred to as alleles.

A person’s genotype is determined by what alleles that person has, while the physiological characteristics or behaviours that geneticists study are referred to as the phenotype. Any given phenotypic behaviour can be the result of having a particular genotype, a particular environmental influence, or some combination of the two. Phenotypic traits are said to be qualitative if they take a limited number of discrete forms and quantitative if they vary continuously. So the presence of the symptoms of Huntington’s disease, a degenerative neurological disorder that affects older people, is a qualitative trait while one’s score on an IQ test is a quantitative trait.

Genetic influence on a phenotype can involve one or more genes. For example, people who have the allele for Huntington’s disease in the single gene encoding the huntingtin protein will contract it. Those who don’t won’t. Contrast that with the genetic influence on measured cognitive ability, which is thought to involve many genes, each of which has a very small effect on scores on tests of mental ability. When many genes influence a phenotypic trait, it is said to be polygenic.

Both qualitative and quantitative traits can be polygenic. A trivial example of a qualitative trait that is polygenic would be having an IQ score over 130. Other than some psychopathologies, most of the behaviours studied are thought to be polygenic with differences in each gene, making only a small contribution to differences in behaviour. In theory a quantitative trait could be influenced by a single gene that influenced the mean of the trait while environment determined the variance around the mean, but no examples of this have been identified.

Normally people inherit 46 chromosomes – 23 from their mothers and 23 from their fathers. Since there are many genes on any one chromosome, the inheritance of different traits can be linked if genes on the same chromosome influence the traits. However, the linkage is not perfect. In the process of creating the chromosomes that will be passed on to one’s children in gamete cells (ova and sperm), contiguous parts of each pair of chromosomes can be swapped so that the chromosome that is passed onto one’s child is a combination of parts from both of one’s parents. This happens on average about once per chromosome in humans. Thus, traits that are influenced by genes located close together on the same chromosome are more likely to be inherited together than genes on the same chromosome that are at distant loci. As will be described later, this fact can be used to identify the location of the genes that affect a particular trait.

If one has different alleles for the same gene on each of a pair of chromosomes there are different possible impacts. In some cases, certain alleles will always be expressed (influence phenotype) if they are present. Such alleles are termed ‘dominant’. Other alleles for the same gene are called ‘recessive’ and will be expressed only if they are not paired with a dominant allele. In other cases, having two different alleles will have an effect on phenotype halfway between the effect of having two of the one allele and the effect of having two of the other. In this case genetic effects are termed ‘linear and additive’.

There can be interactions between multiple genes in creating effects on phenotype. The phenomenon is called ‘epistasis’. For example, there is epistasis if two different alleles of two different genes must be present for a phenotypic trait to be present. In this case, genetic effects on this trait will not be linear and additive.

Relational Studies

Arguably the first behavioural genetics study was Galton’s Hereditary Genius (1869) in which he looked at patterns of career success in English families. He showed that close relatives of prominent men were also likely to achieve distinction, but that the probability fell with more and more distant relatives. While a genetic basis for ability would explain this pattern, so would family connections and a host of other environmental factors. Modern behavioural genetics research uses relational data, but in a way that attempts to control for family environment.

The simplest version of this type of study looks at the behavioural similarity of identical (or monozygote) twins who are raised apart. Such twins are genetic copies of each other as they grew from the same fertilized egg, but, if they are reared apart, then environmental similarities can’t explain any behavioural similarities. If one assumes that genetic and environmental influences on a trait are linear and additive, then one can write

$$ P= hG+ cS+ eN $$
(1)

where P is a measure of the phenotypic behaviour, G is genetic endowment, S is an index of the influence of shared family environment, and N is an index of the influence of environmental factors not shared by family members. The variables G, S and N are not observed, but the parameters h, c and e can still be estimates. If all variables are measured as standard deviations from their means, and G, S and N are uncorrelated, then h, c and e will be the correlations of the respective variable with P and their squares will be the fraction of variance in P that is explained by each. The fractions of variance in P explained by genetic endowment, shared family environment, and non-shared environment are commonly denoted h2, c2 and e2. The sum of the squared coefficients will be one. Under the assumptions that the S’s and N’s of identical twins raised apart are uncorrelated, the expected correlation of P for pairs of twins is h2 or the fraction of variation in the population explained by differences in genetic endowments. This statistic is referred to as the heritability of the trait P.

If one also has data on the correlation of the behaviour for identical twins raised together, one can construct an estimate of the fraction explained by the two environmental components as well. Under the assumption that identical twins raised together have both the same G and the same value for S, the correlation of P across pairs of identical twins raised together will be h2 + c2. So the difference between the correlation of P for identical twins raised apart and those raised together will be the fraction of variance explained by shared family environment, and 1 minus that correlation will equal the share explained by non-shared environment.

With one additional assumption it is not necessary for the adopted siblings to be identical twins. Since natural siblings receive half of their genes from each parent and the genes received from each parent are in some sense a random subset of the parents’ genes, it is not unreasonable to assume that the correlation of G for siblings who are not identical twins will be .5. In that case the expected correlation of a phenotype behaviour for siblings raised apart will be .5 h2, and multiplying that value by 2 yields an estimate of the fraction of variance in the population explained by variation in genetic endowments. Once again, the difference between the correlation for siblings raised apart and those raised together will provide an estimate of the fraction of variance explained by shared family environment. The share attributable to non-shared environment can be computed as 1 minus the sum of the shares of genetic endowment and family environment.

If the effects of genetic endowment are not linear, then heritability estimates derived from studying twins adopted apart will be larger than those for siblings raised separately. Since monozygote twins are genetically identical, they will be affected by dominant genes and interaction effects between genes (epistasis) in exactly the same way. Thus, studies of identical twins measure what is called ‘broad- sense heritability’ (denoted H2) unless dominance and epistasis effects are absent. In the presence of dominance and epistasis effects the correlation of phenotypes between normal sibling pairs raised apart will be less than half of that of identical twins raised apart. Twice the correlation for normal siblings raised apart is said to measure narrow-sense heritability since it doesn’t reflect the contribution of nonlinear genetic effects.

Estimated variance shares from adoption studies can be criticized on a number of grounds. Siblings raised apart, and particularly twins, will share aspects of their prenatal environment at least. They may also share their post-natal environment if they are not adopted away immediately. Also, siblings who are put up for adoption may end up in similar environments for a number of reasons. They may be adopted by relatives, or they may be adopted through the same agency that places children with parents of a particular social class in a particular geographic area. Adopting families may be matched to the socio-economic status of the biological mother. Similar environments will cause adoptees to resemble each other even if there is no effect of genetic endowment and will bias estimates of heritability upward. Adoption itself may affect the trait, leading to an overestimate of heritability and an underestimate of the role of shared environment.

Even if adoption doesn’t place siblings in similar environments, it almost certainly restricts the range of environments compared with those occupied by children living with their natural parents, as adoption agencies rigorously screen parents wishing to adopt. Stoolmiller (1999) argues that this restriction of range leads adoption studies to underestimate the role of shared family environment and overestimate the importance of genetic differences in explaining variance in the general population, since there is much more variation in family environment in the general population than in adopting families. This illustrates an important characteristic of heritability estimates – they apply only to the population in which they are estimated. Populations with different amounts of variation in environment or genetic endowment would exhibit different heritabilities. Finally, the assumption that the correlation of normal siblings with no environment in common will be exactly .5 h2 is probably wrong for another reason. It assumes that each parent’s genes for a trait are a random draw from the population – that is, that men and women don’t choose each other as mates on the basis of the characteristic being studied or anything related to it. If parents are likely to have genes for the trait in common, then the expected correlation will be higher and multiplying it by 2 will overestimate heritability. If opposites attract, then multiplying the sibling correlation by 2 will understate heritability. Estimates of the variance explained by shared family environment will be affected and biased in the opposite direction to heritability.

An alternative to adoption studies are those that contrast the similarity of identical twins with that of fraternal twins. Identical twins are genetic copies of each other while fraternal twins are no more alike genetically than brothers and sisters. Thus we would expect identical twins to be more similar for traits that are subject to genetic influence. Again, under the standard assumptions, the correlation of identical twins in a population will be h2 + c2. If one assumes that fraternal twins’ genetic endowments have a correlation of .5, then their correlation will be .5 h2 + c2. Thus, twice the difference between the correlation for identical and fraternal twins is an estimate of heritability. The fraction of variance explained by shared environment will be equal to the identical twin correlation minus the estimate of heritability, and that of non-shared environment will equal 1 minus the identical twin correlation.

Twin studies, too, can be criticized on a number of grounds. The assumption that the correlation of genetic endowment for fraternal twins will be .5 rests on random mating. If husbands and wives tend to have similar genetic endowments for the characteristic being studied, then the fraternal twin correlation will be greater than .5, and doubling the difference between fraternal and identical twins will understate heritability and overstate the role of shared environment. On the other hand, if there are dominance and epistasis effects, doubling the difference will overstate both broad and narrow sense heritability.

A common criticism of twin studies is that identical twins have more similar environments than fraternal twins and that accounts for some of their greater similarity. Whether or not this is a valid criticism, it certainly illustrates a common misunderstanding about the meaning of heritability. If identical twins have more similar environments because they behave in more similar ways and create for themselves more similar environments, some would say that it is legitimate to attribute the influence of environment of this sort to genetic endowment. In the same sense, natural siblings may have more similar environments than adopted siblings – even if they are raised apart – because their more similar genes induce more similar behaviour which induces more similar responses from their environment. If two siblings are both genetically predisposed to be taller, they may both end up playing on the high-school basketball team, where they receive professional coaching which greatly improves their skills. The similarity of their basketball skill is a direct effect of similar environments, but it is also an indirect effect of genetic endowment. Both twin and adoption studies will attribute such induced environmental effects to genetic endowment.

A common error in the interpretation of heritability estimates is the assumption that, if heritability is high, the effects of environment must be small and the trait not easy to change through environmental intervention. However, if heritability estimates attribute to genetic endowment indirect effects that come through environment, it’s easy to see that this is not the case (see the discussion of malleability in the entry on cognitive ability). If a tall person is good at basketball mainly because he has received good coaching, then the skill of shorter people can probably be improved a great deal by coaching as well (even if they can never be quite as good as the tall person). When genetic endowment has both direct physiological effects on a trait and indirect effects through induced environment, there is gene x environment correlation. Relaxing the assumption that genetic endowment and environmental influences are correlated doesn’t invalidate heritability estimates, but it does change their interpretation as just explained. The fractions of variance explained by shared and non-shared environment in twin and adoption studies are not the full effect of environment, but the fractions explained by the residual environment – that part that can’t itself be explained by differences in genetic endowment.

There is another reason why high heritability estimates do not mean that the effects of environment are necessarily weak. Recall that heritability estimates are valid only in the population in which they are estimated. If we were to study nearsightedness in a population of people who were not wearing corrective lenses, we would find it highly heritable. If we studied scores on an eye test allowing people to wear their corrective lenses, we would probably find very low heritability of test scores. The high heritability of nearsightedness in the first case certainly wouldn’t mean that we couldn’t treat it with corrective lenses.

Interaction of environment and genotype can create problems of interpretation similar to the just-described problems caused by the correlation between genotype and environment. Interaction is said to exist when environment has different effects depending on a person’s genotype. In this case genetic effects are not linear and additive and the variance shares computed using standard behavioural genetic methods do not provide a meaningful measure of effects of genetic endowment and environment on the trait. None the less, high estimates of heritability for a population still indicate a substantial role for genetic variation in causing variation in the trait.

Some of the shortcomings of twin studies and adoption studies can be overcome by combining data from the two. Since they are subject to different biases, if results for the two types of studies are very similar, one can have some confidence that the biases are not important. Data from the two types of studies can be formally combined and used to estimate more elaborate models of inheritance that relax one or more assumptions such as linearity, random mating, or similar treatment of identical and fraternal twins. Information on other types of relations and more distant relations can be added to model building studies as well.

Of all the behaviours to which relational methods have been applied, the one that has received the most attention is scores on tests of cognitive ability. These studies have been extremely controversial – at least in part because of the widespread misunderstanding that high heritability precluded an important role for environment. Today it is widely accepted that the heritability of cognitive test scores in adults is very high (0.6 or more; Neisser et al. 1996; Plomin et al. 2000, pp. 164–77), but it is understood that this does not imply a limited role for environment (as genetic endowment may be acting indirectly through the environment).

Besides cognitive ability, a wide range of other behaviours have been studied. The degree to which people display the symptoms of a number of psychopathologies has been shown to be subject to genetic influence (Plomin et al. 2000, chs 8 and 12). Major measurable aspects of personality (Loehlin 1992), religiosity (Waller et al. 1990), attitudes towards one’s job (Lykken et al. 1993), social attitudes (Martin et al. 1986) (including political conservatism; Eaves et al. 1997), education (Behrman and Taubman 1989), earnings (Taubman 1976), and even the amount of time spent watching television (Plomin et al. 1990), have all been shown to be subject to genetic influence. In most cases, studies find that the fraction of variance explained by variation in genetic endowment is large and greater than the fraction explained by family environment (Turkheimer 2000). Also interesting are the exceptions that have been found to this general pattern. For example, how often one attends church is influenced by one’s genetic endowment, but not the type of church one attends.

A relatively recent development in relational studies is their use to analyse the sources of covariance between different measures of behaviour. By using similar assumptions to those used to identify variance shares, it is possible to tell whether correlations between variables are due mainly to common genetic factors, common environmental factors or both. For example, tests of cognitive ability are strongly correlated with scores on achievement tests and both are highly heritable. Are the same genetic factors responsible for both (as would be the case if genetic influence on achievement came entirely through its effects on cognitive ability)? For the most part they are, though some genetic influence is specific to achievement (Plomin et al. 2000, p. 201).

Animal Models and Molecular Genetics Studies

Work with animals allows behavioural geneticists to do many things that are impossible with human subjects. For example, animals can be bred for certain behavioural traits and then the specially bred animals can be used in experiments. One of the most interesting demonstrations of gene x environment interaction comes from a study of two strains of rats that had been bred for their performance in solving mazes (Cooper and Zubek 1958). One strain was bred for superior performance and one for inferior performance. Rats raised in very sparse environments performed poorly in solving mazes no matter what their genetic endowment. Rats raised in enriched environments performed much better and there was little effect from their genetic endowment. However, rats raised in normal laboratory environments showed large differences consistent with their genetic endowments.

Animal studies can be particularly useful when combined with some of the new molecular genetic techniques. Certain genes can be turned off and the impact on behaviour studied. Genetic mutations can be created in experimental animals and the impact of the mutation on behaviour examined. Selectively bred animals can be compared for the frequency of different alleles to determine where genes that influence a trait are located.

Searches of this sort are facilitated by the previously described tendency for genes that are located close together on a chromosome to be inherited together. Suppose, for example, that animals that had been bred for an extreme form of some behaviour showed a much higher frequency of one allele on one chromosome than did the population from which they were bred. This would not mean that that allele played a role in the development of that trait, but it would make it more likely than not that one or more genes on the chromosome on which the gene was located played some role. The allele that is found to be associated with the trait being studied is said to be a marker for the trait, while the genes with the polymorphisms that matter for the trait are said to be trait loci. If the trait is a quantitative trait, each locus is referred to as a quantitative trait locus (QTL).

If several markers are studied on the same chromosome, some may be found to be more highly associated with the trait than others. The more highly associated markers are likely to be closer to one or more trait loci since, the closer two genes are together on the same chromosome the more likely it is that they will be inherited together.

This technique has been used to identify the location of genes with a large role in determining differences in fearfulness in mice. The same sequence of genes exists in the human genome and it is possible that variations in them may explain why some people develop anxiety disorders and some don’t. Understanding the role of these genes may lead to more effective treatment.

Association techniques can also be used in humans, but are subject to a number of problems. In the example just discussed, the mice studied were all bred from the same homogenous population. The breeding for the trait is likely to have induced any association found between a marker and a phenotype trait. However, in human populations markers and traits could be associated even if there was no genetic influence on the behaviour. This is referred to as the ‘chopstick’ problem, which is named after a commonly cited example of a spurious association. In a population that included native Chinese and Europeans, using chopsticks would be associated with any marker more common in Chinese. This problem can be partially overcome by studying more homogenous populations or contrasting sibling pairs, as differences in marker frequency are more likely to signal genetic causation in these cases. In the extreme, studies can be done on large extended families. The families can be studied for co-transmission of the trait and particular alleles. These are termed ‘linkage studies’. Linkage studies were used to identify the gene responsible for Huntington’s disease.

Linkage studies solve another problem of association studies in humans. Within a family, even markers fairly distant from a trait locus will have some degree of association with the trait. In the general population, markers are likely to be associated with traits only if they are trait loci themselves or are located very close to them, as recombination of chromosomes will eventually break down the association of any marker that is not a trait locus with the trait after a sufficient number of generations. A much smaller number of markers can be used to scan for the location of trait loci in a linkage study than in a study looking for association in the general population. However, linkage studies are not very good at finding QTLs when there are many genes contributing to a phenotype. Association studies in large populations are more promising, but only if the area of the genome to be examined can be narrowed on the basis of hypothesis about what systems might be involved. So far this approach has shown some promise. For example, associations have been found between a particular allele for a dopamine receptor gene and hyperactivity disorder in children (Thapar et al. 1999).

The Future

Relational studies have demonstrated that variation in a surprisingly wide range of behaviours is substantially influenced by genetic differences. Molecular genetics has begun to discover some of the mechanisms by which genetic differences cause differences in behaviour, but work of this sort has barely scratched the surface, and further development faces some difficult obstacles. Most of the behaviours that have been studied are thought to be affected by many different genes, each of which has a small effect. This will make identifying QTLs difficult without some theory of what physiological processes might be involved and where the genes affecting those processes are in the genome. But what theory might one have about the location of physiological processes affecting, for example, time spent watching television? When one begins to think about the many ways in which physiological differences could affect a wide range of behaviours, the task seems daunting. Suppose there was an allele that when present made people feel more discomfort when they were cold than others without the allele. Such people might be inclined to spend more time inside watching TV. They might also be less athletic and/or more likely to spend a lot of time reading. If they read more, they might have larger vocabularies and score better on IQ tests. If their reading made them more sceptical, they might be less likely to attend church. Depending on how myriad and diffuse such cascading effects are, it might be impossible to understand how more than a small fraction of genetically induced differences in behaviour comes about. Still, that doesn’t mean that valuable knowledge can’t be gained from studying the pathways that can be identified. Such knowledge might accumulate faster if those studying the genetic influences on behaviour concentrated less on refining estimates of heritability and more on analysing the role of genetic differences in explaining the covariance of different behaviours.

See Also