Next: Current Theories of the Genesis of Homosexuality; Previous: Twin Studies of Homosexuality; Up: Contents

CRITICAL ANALYSIS OF THE STUDIES

Explaining the observed variability of a trait: The theory behind twin studies

Before an analysis of the studies mentioned in the previous section is attempted, a brief introduction to the theory behind twin studies is required.

The heritability of a given trait in a given population is that proportion of the total variance observed in that trait in that population which is attributable to inherited, genetic differences between members of that population. The remainder of the observed variation may be attributed to environmental differences, namely, differences in common environment (i.e. between families) and differences in specific environment (i.e. within families).

Thus, the simple equation for explaining phenotypic variance is:-

V(P) = V(G) + V(CE) + V(SE) (1)

[where V(P) is the observed phenotypic variance,

V(G) is the variance attributable to differences in genetic factors,

V(CE) is the variance attributable to differences in common environment, &

V(SE) is the variance attributable to differences in specific environment ]

Analysing the relative contributions of genetic and environmental factors in terms of variance has the advantages of simplicity and generalizability; trait variation so expressed may be divided into additive components that sum to the whole. Also, the use of variance allows us to generalise the results of an individual study onto the given population as a whole, while still retaining the additive nature of the model. However, a better indication of the relative importance of the respective influences on individual differences in a trait is given by the square root of each expression of variance (this is Wright's formulation - see Eysenck 1979, p107).

Equation 1 requires that the components due to genetic factors, common environmental factors and specific environmental factors are independent. The true situation is complicated by the possibility of genotype-environment interaction and of covariance of genotype and common environment. An interaction occurs when particular combinations of genes and environments have unexpected effects: for example, although a given environment might be beneficial for most genotypes, it may be detrimental to one specific genotype. Such effects can alter the apparent differences in the amount that genetic and environmental factors contribute towards the observed variation of a trait in either direction. Covariance can occur when a particular genotype tends to be associated with a particular common environment. This would have the effect of reinforcing genetic and common environmental influences, whilst accentuating individual differences and artificially lowering the true degree of variance from all sources.

The simple model of heritability groups all variability in a trait due to genetic factors under the label V(G). In fact, there may be several components to genetic variance. The additive component is that attributable to all individual genes whose action does not depend on their genetic context. The dominance component is the proportion of genetic variance attributable to dominant genes, that is, where the effect of a gene may depend on what other allele is present at that chromosomal locus. Finally, an epistatic component may exist, where the effect of a particular gene may depend on what other genes are simultaneously present.

The distinction is therefore often made between broad heritability, which refers to all of the genetic variance, and narrow heritability, which refers to just that portion of the genetic variance which is additive in nature.

There are several situations in which the additive genetic component of variance may deviate from expected values. One example is that of assortative mating (when mating tends to occur between individuals who are similar in the particular trait in question). Almost by definition, assortative mating must be a significant factor in the genetics of sexual orientation. Such a situation tends to increase the amount of additive genetic variance from that expected from random mating, but the average level of the trait in the population is not affected (see Fuller & Thompson 1978, or Loehlin et al. 1975 for further details).

The simple model of heritability is further complicated by the fact that the environment in which a child grows up is, in part, determined by the genetic makeup of its parents. Such cultural influence means that many aspects of what we consider to be environment (e.g. social class differences, cultural quality of the home) may themselves be partly genetic in origin.

Bearing the above points in mind, it is hardly surprising that studies attempting to investigate the relative importance of genetic factors versus environmental factors in determining the phenotypic expression of a trait are often easily criticised on theoretical, as well as methodological, grounds. The use of twins to study such factors also introduces problems. All such criticisms are discussed in the next section.

Problems with the twin study method

One of the fundamental assumptions of the twin study method is that closer resemblance in monozygotic twins is due to closer genetic similarity rather than more shared environment, i.e. that monozygotic and dizygotic twins share environment to the same extent. However, there is considerable evidence that the common environment is not the same for the two types of twins; the 'twinning reaction', or mutual dependence between twins, is greater between identical twins than between fraternal twins (e.g. Farber 1981, Eysenck 1979). Kamin notes that identical twins tend to be treated very similarly by parents, teachers etc., spend more time together, have more mutual friends, and so on, than do fraternal twins. Conversely, identical twins may, possibly unconsciously, try to make themselves slightly different from their cotwins in appearance, manner etc., in order to achieve a greater degree of individuality. Farber suggests that such "minute interpersonal and intrapsychic events may be the most potent in altering at least some traits fundamentally linked to the somatic blueprint of the individual."

Such tendencies of monozygotic twins to differentiate themselves from each other are, of course, related to their initial degree of similarity. However, a small proportion of monozygotic twins are not strikingly similar in appearance (see, e.g. Eysenck & Kamin 1981). If a twin study relies purely on physical similarity to determine the zygosity of twins, then it is likely that a number of monozygotic twins will therefore be misclassified as dizygotic. This criticism can be applied to some of the earlier twin studies in many fields, but more recent research has utilised more reliable techniques for zygosity determination (e.g. blood-group matching, fingerprint ridge counts etc.).

Roughly two-thirds of identical twins are monochorionic, that is, they shared the same chorion, and hence the same blood supply, during prenatal development. In such an arrangement, one twin receives the maternal blood supply after it has passed through the other, and is therefore at a severe competitive disadvantage regarding oxygen supply, hormone supply, etc.. The consequences of this on the later development of the twins can include gross phenotypic differences, even before postnatal influences of the family are considered (Mesnikoff et al. 1963). Such a process can lead to an over-estimation of the proportion of variance attributable to the specific postnatal environment of the twins.

Conversely, if we consider the production of hormones by the foetus as opposed to those supplied by the mother, the quantity of hormones produced, along with the absolute and relative timing of production, is controlled by the genetic code of the developing individual. Therefore, monozygotic twins experience higher similarity in foetal hormone production, both in timing and in amount, than do dizygotic twins. This has the effect of increasing the similarity for prenatal-hormonally determined traits in monozygotic twins relative to dizygotic twins.

Considering the penultimate point, even monozygotic twins, sharing exactly the same genes, may display phenotypic differences due to their different prenatal environment. Things becomes even more complicated, however, when one considers that not all of an individual's genes are active at any point in his or her life. Gottesman (1974) states that "it cannot be over-emphasized that it is environmental factors through such extracellular metabolic intermediates as hormones, vitamins and toxins that determine which genes get switched on and how long they function . . . Since only a small portion of the genome (perhaps 5-20%) is activated at any one time, the effective genotype upon which environmental factors are acting is constantly changing." Farber (1981) points out that such genetic 'timetables' of vulnerability to environmental influence may be more similar among identical twins than fraternal twins, and more similar among fraternal twins than non-twins. She therefore suggests that "some of the similarity in specific traits is not so much because the trait itself is strongly predetermined, but because the twins were susceptible to environmental influence when they were in similar stages of psychological and maturational organisation." Such factors would lead to an overestimate of heritability estimated from twin studies.

It is, in fact, generally found that twin studies of a particular trait suggest higher estimates of heritability than do adoption studies (Plomin, 1990). In addition to Farber's proposed process, this may also be explained by nonadditive genetic variance, such as epistasis, which covaries completely for identical twins, but contributes little to the resemblance of first-degree relatives.

In many twin studies it is likely that at least two types of bias operate in the selection of twins pairs for inclusion in the sample from all possible twins in the population who meet the criteria for the study. One such bias is concordance dependent ascertainment, where the probability of twin pairs being included in a study of a particular trait is dependent on whether they are concordant or discordant for that trait. Such a bias can occur in a number of ways, even when a voluntary recruitment procedure is adopted. Another bias that may occur is that of non-independent ascertainment, where ascertainment probability depends on the combination of within-pair similarity and the type of relative (e.g. monozygotic or dizygotic twins); for example, it may happen that concordant monozygotic twins are more likely to be included in a particular study than are concordant dizygotic twins.

If a twin sample is obtained which has avoided all such biases, then we still have to ask whether any sample of twins can be representative of the population from which they were drawn. For example, the probability of a twin birth increases with the age of the mother until about the age of 39 (see Farber 1981, chapter 1). This leads to an increased chance of chromosomal anomalies in twins which could affect the concordance rate for traits associated with that chromosomal locus. Farber also points out that there is a high frequency of premature births in multiple deliveries. She suggests the possibility that prematurity can make an individual differently susceptible to the environment than a full-term individual is.

Gottesman and Carey (1983) suggest several internal checks that may be performed on twins sample data so that some confidence may be felt regarding their representativeness. These include checking the proportion of the two sexes, the proportion of various zygosities, and whether twins are over-represented in the reference population of the trait in question. Of course, such checks require knowledge of the corresponding figures for the reference population, which are often uncertain.

All of the above represent formidable, if not insurmountable, problems for the experimental design of a twin study from which we can hope to obtain any meaningful, generalizable results. This has not, as we have seen, prevented a substantial number of researchers from conducting such studies. But even if we assume that these problems have been overcome, the interpretation of the results of twin studies, usually given in terms of concordance rates for monozygotic and dizygotic twins, is problematic.

The ratio of concordance in identical twins to that in fraternal twins may seem like a promising statistic, but is, in fact, not very informative. For example, such a ratio is sensitive to the base rate of the trait in the given population, and will usually have a considerable associated standard error (Kendler, 1989). Also, Gottesman and Carey (1983) demonstrated that quite different concordance rates between the sexes can reduce to the same estimates of underlying heritability.

For a better understanding of the results of twin studies, Gottesman and Carey recommend "that appropriate population risks be determined and that the concordance rates be converted into correlations in the liability toward developing the disorder." They end their report on twin studies on an optimistic note, by listing recent innovations and developments which should assist twin research. Such developments include multivariate analysis, longitudinal twin study analysis, brain scan differences, the use of data on other relatives to check on the assumptions of twin strategies, and a renewal of interest in identical twins reared apart.

Specific problems with the twin studies of homosexuality

Many of the criticisms of twin studies mentioned in the previous section apply to the twin studies of homosexuality.

Rosenthal (1970) has severely criticised Kallman's 1952 study for the high incidence of psychiatric disorders among the probands and their cotwins. He comments that "only 10 of the 80 monozygotic twins and 18 of the entire sample (170 individuals) were thought to be 'sufficiently adjusted' emotionally and socially." There is a possibility that the homosexuality shown by some of the twins is secondary and reactive to their psychopathology, or vice versa. Kallman's study could also be subject to non-independent ascertainment and concordance-dependent ascertainment, although the exact method of recruitment is not explained in his report. As for the 100% concordance rate reported among monozygotic twins, Kallmann himself regarded this as a "statistical artifact" (see his discussion at the end of Rainer et al.'s 1960 report), and was not surprised when monozygotic twins discordant for sexual orientation were reported. Lykken et al. (1987) point out that many twin studies have a disproportionate number of monozygotic probands compared to the given population. This criticism applies to Kallman's report, which included 40 monozygotic pairs and 45 dizygotic pairs. The proportion of monozygotic to dizygotic twins in American and European populations is roughly 1:2 (see Lykken et al. 1987). For these reasons, Kallman's study cannot be considered as representative of the American population as a whole, but is best looked upon as a useful and provocative preliminary publication that has prompted much subsequent research.

Heston and Shields (1968) explain their recruitment and interview techniques in more detail. They emphasize that twins recruited through the Maudsley Twin Register are unselected as regards concordance and zygosity, and attempt to show that monozygosity per se is not associated with homosexuality, and that the incidence of homosexuality in members of same-sexed male twins is no greater than in the parent Maudsley population. Their data may therefore reasonably be considered as representative of this population (but note the small numbers of probands involved), but, again, we see a high incidence of psychiatric disorders in probands and their cotwins. Hence, there is difficulty in generalizing the results of this study onto a larger population.

As Bailey and Pillard point out, their 1991 investigation falls short of the ideal recruitment procedure of systematic sampling from a well-specified population. They admit that concordance-dependent ascertainment (which they term type 1 bias) might have occurred, but note that "concluding that sexual orientation is partially heritable based on different patterns of monozygotic and dizygotic twin concordance is equally valid whether or not type 1 bias occurred." Non-independent bias may also have affected the results; probands with heterosexual adoptive brothers were significantly less likely to consent to have their relative contacted than probands with heterosexual twins, whereas cooperation did not differ notably if relatives were homosexual. Bailey and Pillard suggest that this could lead to an underestimation of the proportion of heterosexual relatives in the adoptive brothers, compared with the twin subsamples, resulting in an underestimation of heritability. They also found that, contrary to the predictions of a simple genetic hypothesis, the rate of homosexuality in nontwin brothers was lower than that of dizygotic cotwins, and roughly equal to that of adoptive brothers. Two possible explanations for this finding were suggested, as described in the previous section, one being merely fluctuations in sampling. Hence, the desirability of replicating the findings is emphasized. Despite these shortcomings, this study clearly represents the most significant research in twin studies of homosexuality conducted to date, and its findings suggest many questions which should be addressed by future studies.

The smaller case studies of homosexuality in twins may be useful in highlighting similarities or differences in the environment experienced by each twin of a pair that have lead to their concordance or discordance for sexual orientation. However, by their very nature (e.g. small sample sizes, bias in recruitment etc.) they cannot produce results which can be generalised to a wide population. It should be noted that many of these reports are of a psychoanalytic nature, and the combined results of many such cases have lead to the development of psychoanalytic theories of homosexuality (see Freud 1953, Friedman 1988, Lewes 1989, and next section). However, such theories can, at best, only be generalised onto the population of individuals receiving psychoanalytic treatment.

The only study of homosexuality in identical twins reared apart is that of Eckert et al. (1986). Some of the problems associated with the data were discussed in the previous section. Although the Minnesota study maintains notably stringent criteria for inclusion of twins, such studies can never fulfil all of the theoretical assumptions upon which heritability calculations are based. It has been suggested (see, e.g. Farber 1981) that even an awareness of twinship in separated twins can affect their development. Even twins separated at birth have shared the prenatal environment of the uterus, which, according to recent theories, may have a critical role in the development of sexual orientation (see next section). Therefore, if separated identical twins show concordance for a particular trait, this cannot, in practice, be directly attributed entirely to their shared genes.

Conclusions from the studies conducted to date

For the many reasons mentioned in the previous section, it is inappropriate to pool the data from the studies to perform some type of meta-analysis (e.g. Rosenthal 1984, Oakes 1986). In their paper on the analysis of twin data, Gottesman and Carey (1983) emphasize the importance of only comparing data pertaining to common population risk and diagnostic criteria.

The studies summarised are of inconsistent quality, with biased and limited samples. Lykken et al. (1987) suggest that the "only wholly dependable method of avoiding errors of recruitment bias may be to employ an incentive or method of recruitment that is about equally effective with dizygotic as with monozygotic twins, preferably a method that is successful with most (>80%) of pairs solicited." Unfortunately, such methods are seldom, if ever, available.

The greatest problem for studies of this type remains that of recruiting large numbers of non-institutionalised probands due to the social ostracism of homosexuals. This problem is gradually diminishing, as evidenced by Bailey and Pillard's 1991 study, but is still far from being negligible.

As for the theory behind twin studies, it has been shown that there are flaws in many people's understanding of the concept of heritability. For example, Bouchard et al. (1990) point out that heritability must increase as V(E), the variance affected by the environment, decreases. Hence, the heritability of a psychological trait reveals as much about the culture as it does about human nature.

Bailey and Pillard's 1991 study is clearly superior to any of its predecessors in terms of experimental design and analysis of results. Although it, too, has its weak points, it is notable that the estimates of heritability derived from the data were significant under a wide range of assumptions concerning the base rate of homosexuality and the degree of ascertainment bias. In contrast, estimates of the proportion of phenotypic variance explained by shared environmental differences were not significant under the same range of assumptions.

These results give reason to believe that there is some constitutional component to male homosexuality. However, the twin data are consistent not only with a purely genetic explanation, but also with one involving possible differences in the degree of shared prenatal environment between monozygotic and dizygotic twins (as explained earlier, monozygotic twins experience higher similarity in foetal hormone production, both in timing and in amount, than do dizygotic twins). Some recent theories of the genesis of homosexuality, to be mentioned in the next section, place critical importance on hormone levels in the prenatal environment of an individual. If such theories are true, then the difference in concordance rates between monozygotic and dizygotic twins could be explained largely in these terms (see next section). It should be noted that such an explanation still relies on genetically controlled prenatal hormone production to account for observed differences in concordance between monozygotic and dizygotic twins.


Next: Current Theories of the Genesis of Homosexuality; Previous: Twin Studies of Homosexuality; Up: Contents
This document was converted to HTML on 16th December 1997 by Tim Taylor http://archive.tim-taylor.com/papers/twin_studies/analysis.html