Genetic similarity, human altruism, and group selection
Genetic similarity, human altruism, and group selection
J. Philippe Rushton (1989)
Abstract: A new theory of attraction and liking based on kin selection suggests that people detect genetic similarity in others in order to give preferential treatment to those who are most similar to themselves. There are many sources of empirical and theoretical support for this view, including (1) the inclusive fitness theory of altruism, (2) kin recognition studies of animals raised apart, (3) assortative mating studies, (4) favoritism in families, (5) selective similarity among friends, and (6) ethnocentrism. Specific tests of the theory show that (1) sexually interacting couples who produce a child are genetically more similar to each other in blood antigens than they are either to sexually interacting couples who fail to produce a child or to randomly paired couples from the same sample; (2) similarity between marriage partners is most marked in the more genetically influenced of sets of anthropometric, cognitive, and personality characteristics; (3) after the death of a child, parental grief intensity is correlated with the child's similarity to the parent; (4) long-term male friendship pairs are more similar to each other in blood antigens than they are to random dyads from the same sample; and (5) similarity among best friends is most marked in the more genetically influenced of sets of attitudinal, personality, and anthropometric characteristics. The mechanisms underlying these findings may constitute a biological substrate of ethnocentrism, enabling group selection to occur.
1. Introduction
Resemblance in a variety of demographic, physical, and psychological attributes – including religious affiliation, socioeconomic status, appearance, abilities, attitudes, and personality – has long been considered an important factor in marriage, attraction, friendship, altruism, and group cohesion. Most explanations of the role of similarity in human relationships focus on immediate, environmental effects, for example, their reinforcement value (Byrne 1971). Recent analyses, however, suggest that genetic influences may also be involved, According to Rushton, Russell, and Wells's (1984) "genetic similarity theory," genetic likeness exerts subtle effects on a variety of relationships and has implications for the study of social behavior in small groups and even in large ones, both national and international. The ability to detect genetic similarity may mediate many aspects of interpersonal behavior, including the avoidance of inbreeding and the optimization of mate choice. In this paper, genetic similarity theory will be introduced in connection with altruism. It is proposed that genetically similar people tend to seek one another out and to provide mutually supportive environments such as marriage, friendship, and social groups. This may represent a biological factor underlying ethnocentrism and group selection.
2. The paradox of altruism
Altruism has long posed a serious dilemma for theories of human nature. Most social scientists studying altruism have focused on environmental explanations, although it is known that human differences are influenced by genes too (Rushton, Fulker, Neale, Nias Eysenck 1986), that altruism is found in many animal species, and that altruism's roots lie deep in evolutionary history (E. 0. Wilson 1975). Defined as behavior carried out to benefit others, in extreme form altruism involves self-sacrifice. In humans altruistic behavior ranges from everyday kindnesses, through sharing scarce resources, to giving up one's life to save others (Rushton 1980). In nonhuman animals, altruism includes parental care, warning calls, cooperative defense, rescue behavior, and food sharing; it may also involve self-sacrifice (E. 0. Wilson 1975). For example, the poisonous sting of a honeybee is an adaptation against hive robbers. The recurved barbs facing backward from the sharp tip cause the whole sting to be wrenched out of the bee's body, along with some of the bee's vital internal organs. These barbs have been described as instruments of altruistic self-sacrifice (Ridley & Dawkins 1981).
As recognized by Darwin (1871), however, a genetic basis for altruism would represent a paradox for theories of evolution: How could altruism evolve through "survival of the fittest" when, on the face of it, altruistic behavior diminishes personal fitness? If the most altruistic members of a group sacrifice themselves for others, they run the risk of leaving fewer offspring to pass on the very genes that govern the altruistic behavior. Hence altruism would be selected against, and selfishness would be selected for.
The resolution of the paradox of altruism is one of the triumphs that led to the new synthesis called sociobiology: By a process known as kin selection, individuals maximize their inclusive fitness rather than only their individual fitness by increasing the production of successful offspring by both themselves and their genetic relatives (Hamilton 1964; Maynard Smith 1964). According to this view, the unit of analysis for evolutionary selection is not the individual organism but its genes: Genes are what survive and are passed on, and some of the same genes will be found not only in direct offspring but in siblings, cousins, nephews/nieces, and grandchildren. If an animal sacrifices its life for its siblings' offspring, it ensures the survival of common genes because, by common descent, it shares 50% of its genes with each sibling and 25% with each sibling's offspring.
It is accordingly predicted that the percentage of shared genes is an important determinant of the amount of altruism displayed. This is borne out in a number of species. Social ants, for example, are among the most altruistic species so far discovered, and because of a special feature of their reproductive systems, they also turn out to share 75% of their genes with their sisters (E. 0. Wilson 1975). Thus, in working for others, and sacrificing their lives if need be, they help to propagate their own genes. Extreme forms of altruism may also occur in clones (e.g., aphids), individuals that are 100% related (Ridley & Dawkins 1981).
Kin selection theory is central to contemporary sociobiological theorizing (Dawkins 1976; E. 0. Wilson 1975), although only recently have serious attempts been made to apply it to human relationships (Alexander 1979; 1987; Chagnon 1988; Chagnon & Irons 1979; Daly & Wilson 1988; Freedman 1979; Glassman et al. 1986; Reynolds et al. 1987; van den Berghe 1981; E. 0. Wilson 1978). One might expect kin selection theory's emphasis on altruism between relatives to have limited applicability to human beings, whose altruism is frequently directed' to nonkin and can usually be explained by such culturally influenced mechanisms as empathy, reciprocity, and social rules.
Adopting the mechanistic viewpoint of the "selfish gene" (Dawkins 1976), Rushton et al. (1984) extended kin selection theory to the human case by applying genetic similarity theory. We argued that if a gene can better ensure its own survival by acting so as to bring about the reproduction of family members with whom it shares copies, then it can also do so by bringing about the reproduction of any organism in which copies can be found. This would be an alternative way for genes to propagate themselves. Rather than merely protecting kin at the expense of strangers, if organisms could identify genetically similar organisms, they could exhibit altruism toward these "strangers" as well as toward kin. Kin recognition might be just one form of genetic similarity detection.
Much of this had been proposed earlier by others (e.g., Dawkins 1976; 1982; Hamilton 1964; Thiessen & Gregg 1980). Dawkins (1976; 1982), for example, building on the ideas of Hamilton (1964), suggested a thought experiment in which a gene has two effects: It causes individuals who have it (1) to grow a green beard and (2) to behave altruistically toward green-bearded individuals. The green beard serves as a recognition cue for the altruism gene. Altruism could therefore occur without the need for the individuals to be directly related. Similarly, Thiessen and Gregg (1980) suggested that "the flow of altruistic behaviors, the ease of information transfer, and the genetic benefits of positive assortative mating are linked to the degree with which interacting individuals share homologous genes" (p. 111). They reviewed data from both animals and humans to support their contention that "friends of either sex, as well as mates, resemble each other in many ways, suggesting that genetic assortment operates at all levels of social affiliation" (p. 117).
Several researchers have used mathematical models to show that under a variety of conditions, selection may favor genetic mutations that incline organisms to aid other organisms that share copies of genes but are not necessarily kin (Samuelson 1983; but see Hamilton 1987; Glassman et al. 1986; Russell 1987). For example, Russell (1987) showed that if such a mutation occurred, and if the benefit to the recipient was one-and-a-half times as great as the cost to the donor, a gene for directing altruism toward siblings disappears, but a gene for like-gene detection evolves. As discussed in section 10, mathematical models of the coevolution of genes and culture can amplify this process enormously by taking into account that people adopt ideologies and behaviors that benefit large populations.
Another way that sociobiologists have suggested that altruism could evolve is through reciprocity. Here there is no need for genetic relatedness; performing an altruistic act need only lead to an altruistic act in return (Trivers 1971). Genetic similarity and reciprocal altruism may interact: The more genes are shared by organisms, the more readily reciprocal altruism and cooperation should develop because this eliminates the need for strict reciprocity. Axelrod and Hamilton (1981), Rothstein (1980), and Thiessen and Gregg (1980) make the same point. Thiessen and Gregg state that "cooperation among 'nonrelatives' ('reciprocal altruism') may be based in large part on genetic and phenotypic similarity" (p. 133).
3. Detecting genetic similarity
In order to pursue a strategy of directing altruism toward similar genes, the organism must be able to detect genetic similarity in others. There is clearly no such thing as "genetic ESP." For individuals to direct altruism selectively to genetically similar individuals, they must respond to phenotypic cues. The importance of kin recognition mechanisms was noted by Hamilton (1964). Four such mechanisms that have been considered in the literature (Fletcher & Michener 1987; Holmes & Sherman 1983) are discussed in the following sections.
3.1. Innate feature detectors
Hamilton (1964) suggests that individuals have "recognition alleles" that control the development of mechanisms allowing them to detect genetic similarity in strangers. Most reviewers have considered the existence of genetic similarity detectors (e.g., the "green beard effect" described earlier) to be improbable. Their existence should not yet be discounted, however, because innate pattern recognition does occur in other areas; several studies will be discussed later that suggest we may indeed have such an ability.
3.2. Phenotype matching
The individual may be genetically guided to learn its own phenotype, or those of its close kin, and then to match new, unfamiliar phenotypes to the template it has learned – for example, Dawkins's (1982) "armpit effect." Individuals that smell (or look or behave) like oneself or one's close kin could be distinguished from those that smell (or look or behave) differently. This mechanism would depend on the existence of a strong correlation between genotype and phenotype.
3.3. Familiarity or association
Preferences may also depend on learning through social interaction. This may be the most common means of kin recognition in nature. Individuals that are reared together are more likely to be kin than nonkin. This may also involve a more general mechanism of short-term preference formation. Zajonc (1980) has shown experimentally that the more one is exposed to a stimulus, the more one prefers it. Based on studies of Japanese quail and of humans, respectively, Bateson (1983) and van den Berghe (1983) have suggested that sexual preferences may be established early in life through an imprinting-like process.
3.4. Location
The fourth kin recognition mechanism depends on a high correlation between an individual's location and kinship. The rule states: "If it's in your nest, it's yours." Where a person is and whom the person encounters can also be based on similar genes – for example, if parents exert discriminatory influence on where and with whom their children interact. Many factors that influence one's locus and contacts, such as intelligence, personality, values, and vocational interests, turn out to have some genetic basis (Loehlin et al. 1988.) Physical proximity has been widely observed to be predictive of friendship formation and spouse choice (Burley 1983).
Currently one can only speculate about the extent to which these four mechanisms operate in humans. They are not mutually exclusive. If there are evolutionary advantages to be gained from the ability to detect genetic similarity, all the mechanisms may be operative. If "stronger" mechanisms (innate feature detectors, phenotype matching) operate, it should be possible to demonstrate that interpersonal relationships are mediated by genetic similarity without the help of learning from familiarity or location. Several animal and human studies are described in the next section that suggest this occurs.
4. Kin recognition in animals
There is dramatic experimental evidence that many animal species recognize genetic similarity. Greenberg (1979) showed that the sweat bee, Lasioglossum zephyrum, can discriminate between unfamiliar conspecifics of varying degrees of relatedness. Guard bees of this species block the nest to prevent intruders from entering. In this study bees were first bred for 14 different degrees of genealogical relationship with each other. They were then introduced near nests that contained sisters, aunts, nieces, first cousins, or more distantly related bees. In each case the guard was expected to make a binary decision – either permitting the bee that was introduced to pass or actively preventing it from doing so. There was a strong linear relationship (r = 0.93) between the ability to pass the guard bee and the degree of genetic relatedness. The greater the degree of genetic similarity, the greater the proportion of bees that were allowed to enter the hive. The guard bees appear to be able to detect the degree of genetic similarity between themselves and the intruder. In subsequent kin recognition studies, Breed (1983) and Getz and Smith (1983) have shown that the honeybee, Apis mellifera, is able to discriminate between full and half sisters raised in neighboring cells.
There is also evidence that the ability to detect genetic similarity exists in various species of plants, tadpoles, birds, rodents, and rhesus monkeys. In studies of the frog Rana cascadae, by Blaustein and O'Hara (1981; 1982), tadpoles were separated before hatching and reared inisolation. The individual tadpoles were then placed in a rectangular tank with two end compartments created by plastic mesh. Siblings were placed in one compartment and nonsiblings in the other. The separated tadpoles spent more time at the siblings' end of the tank. Because the tadpoles were separated as embryos and raised in complete isolation, an ability to detect genetic similarity is implicated. Similar findings have been reported for Bufo americanus toad tadpoles (Waldman 1982). Kin recognition has been reported in Japanese quail by Bateson (1983), and in Canada geese by Radesater (1976).
Mammals are also able to detect degrees of genetic relatedness (Fletcher & Michener 1987; Holmes & Sherman 1983). For example, Belding's ground squirrels produce litters that contain both sisters and half sisters. Despite the fact that they shared the same womb and inhabit the same nest, full sisters fight less often than half sisters, come to each other's aid more, and are less prone to chase one another out of their home territory. Similar findings have been noted among captive multimale, multifemale groups of rhesus monkeys growing up outdoors in large social troops. Adults of both sexes are promiscuous, but mothers appear to chase paternal half siblings away from their infants less often than they do unrelated juveniles, and males (despite promiscuity) appear to "recognize" their own offspring, for they treat them better (Suomi 1982). In the preceding examples, the degree of genetic relatedness was established by blood tests. Walters (1987) has reviewed well-replicated data from several primate species indicating that grooming, alliance formation, cooperative defense, and food sharing occur more readily in kinship groups.
5. Kin recognition in humans
Because language represents a powerful new way to distinguish kin, it is more difficult to demonstrate that humans can recognize kin in a way that parallels kin recognition in nonhuman animals. Some indirect evidence is nonetheless discussed by Wells (1987). One approach is to show that humans have perceptual abilities that enable them to discriminate between kin and nonkin; another is to consider whether there are aspects of human social interactions that reflect differing degrees of kinship.
Humans are capable of learning to distinguish kin from nonkin at an early age. Infants can distinguish their mothers from other women by voice alone at 24 hours of age, know the smell of their mother's breast before they are 6 days of age, and recognize a photograph of their mother when they are 2 weeks old (see Wells 1987, for review). Mothers are also able to identify their infants by smell alone after a single exposure at 6 hours of age, and to recognize their infant's cry within 48 hours of birth.
Human behavior also seems to follow lines of genetic similarity with respect to kin preference. For example, among the Ye'Kwana Indians of South America, the words "brother" and "sister" cover four different categories ranging from individuals who share 50% of their genes (identical by descent) to individuals who share only 12.5% of their genes. Hames (1979) has shown that the amount of time the Ye'Kwana spend interacting with their biological relatives increases with their degree of relatedness, even though their kinship terminology does not reflect this correspondence. Anthropological data also show that in societies where certainty of paternity is relatively low, males direct material resources to their sisters' offspring (to whom their relatedness is certain) rather than to their wives' offspring (Kurland 1979). [See also Hartung: "Matrilineal Inheritance" BBS 8 (4) 1985.] An analysis of the contents of 1,000 probated wills revealed that after husbands and wives, kin received about 55% of the total amount bequeathed whereas nonkin received only about 7%; offspring received more than nephews and nieces (Smith et al. 1987).
When the level of genetic similarity within a family is low, the consequences can be serious. Children who are unrelated to a parent are at risk; a disproportionate number of battered babies are stepchildren (Lightcap et al. 1982). Also, unrelated people living together are more likely to kill each other than are related people living together (Daly & Wilson 1988). Converging evidence shows that adoptions are more likely to be successful when the parents perceive the child as similar to themselves (Jaffee & Fanshel 1970).
6. Testing genetic similarity theory
Almost all the studies reviewed so far were carried out to test kin selection theory. However, genetic similarity theory goes further and makes predictions about nonkin and sibling relationships that kin selection theory does not make. Several recent studies have found evidence supporting genetic similarity theory from data on mate choice, sibling favoritism, and same-sex friendships.
6.1. Spouse selection
A well-known phenomenon that is readily explained by genetic similarity theory (but not by most versions of kin selection theory) is positive assortative mating, that is, the tendency of spouses to be nonrandomly paired in the direction of resembling each other in one or more traits more than would be expected by chance. It is widely accepted that there is similarity between human spouses in such characteristics as race, socioeconomic status, physical attractiveness, ethnic background, religion, social attitudes, level of education, family size and structure, IQ, and longevity (Buss 1985; Epstein & Guttman 1984; Thiessen & Gregg 1980). [See Buss: "Sex Differences in Human Mate Preferences" BBS 12 (1) 1989]. For example, the median assortative mating coefficient for IQ in one review, averaged over 16 studies of 3,817 pairings, was 0.37 (Bouchard & McGue 1981). Spouse correlations tend to be high for opinions, attitudes, and values (0.40 to 0.70) and low for personality traits and personal habits (0.02 to 0.30). Spouses also resemble each other in a variety of physical features. Rushton, Russell, and Wells (1985) combined anthropometric data from a wide range of studies and found low but positive correlations for more than 60 different measures, including height (.21), weight (.25), hair color (.28), eye color (.21), chest breadth (.20), wrist circumference (.55), and interpupillary breadth (.20). If mating had been random, the correlations between spouses would have been zero.
Less well known is the fact that spouses also resemble each other in socially undesirable characteristics, [1] including aggressiveness, criminality, alcoholism, and psychiatric disorders such as schizophrenia and the affective disorders. Guze et al. (1970) found that both the wives and the sisters of criminals tended to exhibit the same psychopathology. Gershon et al. (1973) reported that both the wives and the relatives of males suffering from affective disorders exhibited a high prevalence of the same problem. Although alternative reasons can be proposed for this finding, such as the consequences of competition for the fittest mates (Burley 1983), it does raise the possibility that the tendency to seek a similar partner may sometimes override considerations such as mate quality and individual fitness.
A study of cross-racial marriages in Hawaii found more similarity in personality test scores among males and females who married across ethnic groups than among those marrying within them (Ahern et al. 1981). The researchers posit that, given the general tendency toward homogamy, couples marrying heterogamously with regard to racial/ethnic group tend to "make up" for this dissimilarity by choosing spouses more similar to themselves in other respects than do persons marrying within their own racial/ethnic group.
One could argue that positive assortative mating has nothing to do with questions about genetic similarity, that it results only from common environmental influences. This view cannot easily account for the incidence of assortative mating for similarity in species ranging from insects to birds to primates, in laboratory as well as in natural settings (Bateson 1983; Fletcher & Michener 1987; Thiessen & Gregg 1980). To have evolved independently in such a wide variety of species, positive assortative mating must confer substantial advantage. Advantages thought to accrue to choosing an "optimal" degree of genetic similarity in human mates include (1) increased marital stability, (2) increased relatedness to offspring, (3) increased within-family altruism, and (4) greater fecundity. Evidence for the first three putative advantages is presented in section 6.2. With respect to the fourth, Thiessen and Gregg (1980) reviewed literature showing positive correlations between number of children and degree of between-spouse similarity in anthropometric variables, intelligence test scores, educational attainment, and family size in the parental generation. Bresler (1970) found that fetal loss increased with the distance between the birthplaces of parents and with each additional country of birth among great-grandparents. Additional evidence for a relationship between similarity and fecundity is presented in section 6.1.1.
The upper limit on the fitness-enhancing effect of assortative mating for similarity occurs with incest. Too much genetic similarity between mates increases the chances that harmful recessive genes may combine. The negative effects of "inbreeding depression" have been demonstrated in many species, including humans (Jensen, 1983; Thiessen & Gregg 1980). As a result, many have hypothesized that the "incest taboo" has an evolutionary basis, possibly mediated through negative imprinting on intimate associates at an early age. [See van den Berghe; "Inbreeding Avoidance" BBS 6 (1) 1983.] Optimal fitness, then, may consist in selecting a mate who is genetically similar but not actually a relative. Van den Berghe (1983) speculates that the ideal percentage of relatedness is 12.5% identical by descent, or the same as that between first cousins. Other animal species also avoid inbreeding. For example, several experiments have been carried out with Japanese quail, birds that, although promiscuous, proved particularly sophisticated. They preferred first cousins to third cousins, and both of these relatives to either unrelated birds or siblings, thus avoiding the dangers of too much or too little inbreeding (Bateson 1983).
6.1.1. Blood tests. To directly test the hypothesis that human mating follow lines of genetic similarity, Rushton (1988a) examined blood antigen analyses from nearly 1,000 cases of disputed paternity. Seven polymorphic marker systems - ABO, Rhesus (Rh), MNSs, Kell, Duffy (Fy), Kidd (Jk), and HLA - at 10 loci across 6 chromosomes were examined in a sample limited to people of North European appearance (judged by photographs kept for legal identifications). Such blood group differences provide a biological criterion sufficient to identify more than 95% of true relatedness in situations of paternal dispute (Bryant 1980) and to reliably distinguish even between fraternal twins raised in the same family (Pakstis et al. 1972). They also provide a less precise but still useful estimate of genetic distance among unrelated individuals. The method of calculating genetic similarity is explained in Table 1.
The results (Table 2) showed that the degree of genetic similarity within pairs related to (1) whether the pair was sexually interacting or randomly generated from the same sample and (2) whether the pair produced a child. Sexually interacting couples were found to share about 50% of measured genetic markers, part way between mothers and their offspring, who shared 73%, and randomly paired individuals from the same sample, who shared 43% (all comparisons were significantly different, p < .001). In the cases of disputed paternity, genetic similarity predicted whether the male was the true father of the child. Males not excluded from paternity were 52% similar to their partners whereas those excluded were only 44% similar (p < .001).
6.1.2. Genetic similarity detection between marriage partners. If people choose each other on the basis of shared genes, it should be possible to demonstrate that interpersonal relationships are influenced more by genetic similarity attributable to a similar environment. Positive assortative mating might be expected to occur on the basis of the more heritable rather than the less heritable traits because the more genetically influenced traits reflect the underlying genotype better and provide a more accurate cue for matching. To control for the effects of other variables, this hypothesis must be tested on sets of homogenous traits (e.g., anthropometric versus attitudinal variables; see section 7.2).
The issue of differential heritabilities has not yet been resolved in the behavior genetic literature (Loehlin et al. 1988). One result of our work, however, has been the finding that differential estimates of genetic influence on anthropometric, attitudinal, cognitive, and personality variables are considerably more generalizable, even across distinct ethnic and national groups, than might have been expected (Rushton 1989a). Numerous studies have shown that these estimates do indeed predict the degree of similarity between marriage partners. Several of these correlations are summarized in Table 3. Note that many of the estimates of genetic influence in this table are based on calculations of midparent-offspring regressions using data from intact families; this combines genetic and shared-family environmental influences. The latter source of variance, however, is surprisingly small (Plomin & Daniels 1987) and has not been found to add systematic bias. Nonetheless, it should be borne in mind that genetic influence has often been calculated in this way.
Using a within-subjects design, Russell et al. (1985) examined data from three studies reporting independent estimates of genetic influence and assortative mating and found positive correlations between the two sets of measures (r = 0.36, p < 0.05, for 36 anthropometric variables; r = 0.73, p < 0.10, for 5 perceptual judgment variables; and r = 0.44, p < 0.01, for 11 personality variables). In the case of the personality measures, test-retest reliabilities over a three-year period were available and were not found to influence the results.
Another test of the hypothesis was made by Rushton and Russell (1985) using data on 54 personality traits. It was found that both component and aggregate estimates of genetic influence predicted similarity between spouses (rs = 0.44 and 0.55, ps < 0.001). Rushton and Russell (1985) reviewed other reports of similar correlations, including Kamin's (1978) calculation of r = 0.79 (p < 0.001) for 15 cognitive tests and the DeFries et al. (1978) calculation of r = 0.62 (p < 0.001) for 13 anthropometric variables. Cattell (1982) too had noted that between spouse correlations tended to be lower for the less heritable, more specific cognitive abilities (tests of vocabulary and arithmetic) than for the more heritable general abilities (g, from Progressive Matrices). Differential test reliability is unlikely to have been the cause of the findings concerning the anthropometric variables reported by DeFries et al. (1978) or those reported by Russell et al. (1985) because Rushton (1989b, see section 6.3) found that these estimates can be made with very high degrees of precision (e.g., inter-rater reliability > 0.90).
Subsequently, these analyses were extended to include a between-subjects design and the phenomenon was found to be generalizable. Rushton and Nicholson (1988) analyzed data from studies using 15 subtests from the Hawaii Family Study of Cognition (HFSC) and 11 subtests from the Wechsler Adult Intelligence Scale (WAIS); positive correlations were calculated within and between samples. For example, in the HFSC, parent-offspring regressions (corrected for reliability) using data from Americans of European ancestry in Hawaii, Americans of Japanese ancestry in Hawaii, and Koreans in Korea correlated positively with between-spouse similarity scores taken from the same samples and with those taken from two other samples: Americans of mixed ancestry in California, and a group in Colorado. The overall mean, r, was 0.38 for the 15 tests. Aggregating across the numerous estimates to form the most reliable composite gave a substantially better prediction of mate similarity from the estimate of genetic influence (r = 0.74, p < .001). Similar results were found in the WAIS. Three estimates of genetic influence correlated positively with similarities between spouses based on different samples, and in the aggregate they predicted the composite of spouse similarity scores with r = 0.52 (p < 0.05).
Parenthetically, it is worth noting that partialling out g in both the HFSC and the WAIS analyses led to substantially lower correlations between estimates of genetic influence and assortative mating, thus offering support for the view that marital assortment in intelligence occurs primarily with the g factor (Cattell 1982; Eaves et al. 1984; Nagoshi & Johnson 1986). The g factor tends to be the most heritable component of cognitive performance measures (Vernon 1989). [See also Jensen: "Black-White Difference" BBS 8 (2) 1985]
6.2. Intrafamllial relationships
One consequence of genetic similarity between spouses is a concomitant increase of within-family altruism. Several studies have shown that not only the occurrence of relationships but also their degree of happiness and stability can be predicted by the degree of matching of personal attributes (Bentler & Newcomb 1978; Cattell & Nesselroade 1967; Eysenck & Wakefield 1981; Hill et al. 1976; Meyer & Pepper 1977; Terman & Buttenwieser 1935a; 1935b). Because many of the traits on the basis of which spouses choose each other are about 50% heritable (Loehlin et al. 1988), it follows that the matching results in genetic similarity. Whereas each trait may add only a tiny amount to the total genetic variance shared by spouses, the cumulative effects could be considerable.
A related prediction can be made about parental care of offspring that differ in similarity. Because kin selection theory emphasizes that all siblings having genes "identical by descent" with a 0.5 coefficient of relationship (Mealey 1985; Trivers 1985), sibling differences have been overlooked as a topic of research. Positive assortative mating for genetically based traits may combine with the vagaries of meiosis, however, to make some children genetically more similar to one parent or sibling than to others. This can be illustrated as follows. If a father gives his child 50% of his genes, 10% of them shared with the mother, and the mother gives the child 50% of her genes, 20% shares with the father, then the child will be 60% similar to the mother and 70% similar to the father. Although the predictions from kin selection theory are unclear because of its focus on genes that are identical by descent (in which all full siblings share coefficients of 0.5), genetic similarity theory predicts that parents and siblings will favor those who are most similar.
Littlefield and Rushton (1986) attempted to shed light on this hypothesis. In their study of bereavement following the death of a child, it was predicted that the more similar to the parent the child was perceived to be, the greater would be the intensity of that parent's grief experience. (We assume that the perceived similarity with offspring is correlated with genetic similarity, an assumption supported by data from blood tests; see Pakstis et al. 1972). All respondents had to pick which side of the family the child "took after" more, their own or their spouse's. Spouses agreed 74% of the time on this item, and both mothers and fathers, irrespective of the sex of the child, grieved more for the children they perceived as resembling their side of the family more. Other evidence of within-family preferences comes from a review by Segal (1988) of feelings of closeness, cooperation, and altruism in twin pairs. Compared with dizygotic twins, monozygotic twins worked harder for their cotwins on tasks, maintained greater physical proximity, expressed more affection, and suffered greater loss following bereavement.
6.3. A genetic basis for friendship?
Friendships also appear to be formed on the basis of similarity. This assumption holds for similarity as perceived by the friends, and for a variety of objectively measured characteristics, including activities, attitudes, needs, and personality (Berscheid 1985; Thiessen & Gregg 1980). Moreover, in the experimental literature on who likes whom, and .why, one of the most influential variables is perceived similarity. Apparent similarity of personality, attitudes, or any of a wide range of beliefs has been found to generate liking in subjects of varying ages and from many different cultures (Berscheid 1985; Byrne 1971). According to the genetic similarity view, there is a genetic basis to friendship and friendship is one of the mechanisms that leads to altruism.
Social psychological studies show that altruism tends to increase with the benefactor's actual or perceived similarity to the beneficiary (Krebs & Miller 1985; Rushton 1980). For example, Stotland (1969) had subjects observe a person who appeared to be receiving electric shocks. When Stotland manipulated the subjects' beliefs about their similarity to that person, perceived similarity was correlated with reported empathy as well as with physiological skin conductance measures of emotional responsiveness. Krebs (1975) has found that apparent similarity not only increases physiological correlates of emotion such as skin conductance, vasoconstriction, and heart rate, but also the willingness to reward the victim. In young children, the frequency of social interactions between friends corresponds closely to the frequency of acts of altruism between them (Strayer et al. 1979).
Data show that the tendency to choose similar individuals as friends is genetically influenced. In a study of delinquency among 530 adolescent twins by Rowe and Osgood (1984), path analysis revealed not only that antisocial behavior was about 50% heritable, but that the correlation of 0.56 between the delinquency of an individual and the delinquency of his friends was mediated genetically, that is, that students genetically disposed to delinquency were also genetically inclined to seek each other out for friendship. In a study of 396 adolescent and young adult siblings from both adoptive and nonadoptive homes, Daniels and Plomin (1985) found that genetic influences were implicated in choice of friends: Biological siblings were more similar to each other in the types of friends they had than were adoptive siblings.
To test further whether friends are more similar to each other genetically than they are to an average person and whether, like spouses, their resemblance is most marked in the more heritable components of shared traits, Rushton (1989b) examined blood types and differential heritability estimates. Their methods paralleled those used in the studies on heterosexual partners described in sections 6.1.1 and 6.1.2.
Seventy-six long-term, nonrelated, nonhomosexual male Caucasian friendship pairs ranging in age from 18 to 57 years were recruited by advertisements from the general community; the friendships had to have existed for at least one year. A control group was formed by randomly pairing individuals from the sample. At the testing session, a 1 9.- 11) 14-milliliter blood sample was drawn from each person and many variables were measured, including those explicitly chosen because estimates had been calculated of the degree of genetic influence on the various components. For examples, 36 heritabilities were available with respect to 50 social attitude items (see Table 4) from data on 3,810 Australian twin pairs (Martin et al. 1986). For 90 items from the Eysenck Personality Questionnaire, two independent sets of heritability estimates were available for a total of 81 of the items, one set from 3,810 Australian twin pairs (Jardine 1985), and the other set from 627 British twin pairs (Neale et al. 1986). These intercorrelated with r = 0.44 (p < 0.001) and were aggregated to form a more reliable composite. For 13 anthropometric measures, estimates of genetic influence were available based on midparent offspring regressions from data on 125 families in Belgium (Susanne 1977). Test-retest data were available for the two questionnaire measures and pilot work had shown that the anthropometric measures could be made with very high levels of precision (Rushton 1989b).
6.3.1. Blood tests. The percentage of similarity between the friendship pairs was calculated for the same 10 loci from 7 polymorphic blood systems — ABO, Rhesus (Rh), MNSs, P, Duffy (Fy), Kidd (Jk), and HLA — used in the study of sexually interacting couples. The percentage of similarity for the same measure was also calculated for an equal number of randomly paired individuals from the same sample. The similarities, presented in Table 5, are significantly different from each other (t(150) 3.13, p < 0.05). It is unlikely that this outcome is due entirely to stratification effects because within-pair differences in age, education, and occupation did not correlate with the blood similarity scores (mean r = —0 .05). Thus, friends are more similar to each other, genetically, than they are to randomly paired persons from the same sample.
6.3.2. Genetic similarity detection between friends. Examples of varying heritabilities include: 51% for attitude to the death penalty versus 25% for attitude to the truth of the Bible (see Table 4), 41% for having a preference for reading versus 20% for having a preference for many different hobbies (Neale et al. 1986), and 80% for midfinger length versus 50% for upper arm circumference (Susanne 1977). In the studies summarized in Table 3, estimates of genetic influence in relationships between spouses were based on parent-offspring regressions; however, in the study by Rushton (1989b), described in Table 4, heritabilities calculated from the comparison of monozygotic and dizygotic twins raised together were the measures primarily used. When evaluating these results, it should be kept in mind that the friendship heritabilities were generalized from one sample (e.g., Australian twins) to another (Canadian friends). Behavioral scientists usually consider heritability estimates to be properties of particular populations and not to be highly generalizable (Falconer 1981; cf. Rushton 1989a). This results in a conservative test of the genetic similarity hypothesis because the predicted effect has to be sufficiently generalizable to overcome this problem.
Across the measures, close friends were found to be significantly more similar to each other than to randomly paired individuals from the same sample. Pearson product-moment correlations showed that compared with random pairs, friendship dyads are more similar in age (0.64 vs. -0.10, p < 0.05), education (0.42 vs. 0.11, p < 0.05), occupational status (0.39 vs. -0.02, p < 0.05), conservatism (0.36 vs. -0.02, p < 0.05), mutual feelings of altruism and intimacy (0.32 vs. -0.04 and 0.18 vs. -0.08, ps < 0.05), 13 anthropometric variables (mean = 0.12 vs. -0.03, ns), 26 personality scale scores (mean = 0.09 vs. 0.00, ns), and 20 personality self-rating scores (mean = 0.08 vs. 0.00, ns). Although these similarities are very small, significantly more are positive than could be expected by chance (13/13 of the anthropometric variables, 18/26 of the personality scale scores, and 15/20 of the personality self-rating scores, all p < 0.05, binomial sign test). It should be noted that these relative magnitudes parallel the between-spouse similarities (Buss 1985; Epstein & Guttman 1984; Rushton et al. 1985; Thiessen & Gregg 1980).
As with marriage partners, similarity between friends was most marked among the more genetically influenced of the characteristics. For the 36 conservatism items (see Table 4), the correlation of the estimates of genetic influence and between-friend similarity was 0.40 (p < 0.01); this relationship was not altered when corrected for test-retest reliability or when similarity in a composite of age, education, and occupational status was partialled out (r = .40, p < .01; r = .32, p < .05, respectively). For the 81 personality items, the correlation 0.20 (p < .05) was also not changed by a correction for test-retest reliabiity or socioeconomic similarity. For the 13 anthropometric variables, however, the correlation was not significant (r = 0.15). Given the stringent between-sample rather than within-sample test of the hypothesis, it seems reasonable to conclude that friends are choosing each other more on the basis of the genetically influenced components of similarity than on the basis of environmentally influenced components.
7. Discussion
The preliminary evidence presented so far supports the hypothesis that both friends and spouses choose each other partly on the basis of genetic similarity. The blood antigen data clearly show that friendship dyads as well as sexually interacting couples who produce children together are genetically more similar to each other than to random pairs from the same samples. The fact that similarity is greater for the more genetically influenced components of traits than for the environmentally influenced ones suggests that positive assortment is genetically mediated. Objections to these conclusions can certainly be raised, and alternative hypotheses are possible. Different theories, interpretations of data, and underlying mechanisms are discussed in the following sections.
7.1. Theory
It has been objected that genetic similarity theory is implausible and fallacious (for an exchange of views, see Mealey 1985; 1989; Rushton 1989c; Rushton & Russell 1985; see also Trivers 1985, p.423). The main point made by critics is that the overall proportion of genes shared by two individuals is irrelevant unless the genes are linked to a "gene for altruism" and such a link is unlikely to remain across generations because genes assort independently. Two ways to avoid this problem are (1) to follow Hamilton's rules and depend on the statistics of identity by common descent to ensure the presence of altruistic genes in others and (2) to discover some phenotypic character that is very closely linked to or associated with altruism (as in Dawkins's [1976] "green beard" idea, described in section 2). Critics point out that the existence of such a character is considered unlikely even by the theory's formulators (Hamilton 1964; Dawkins 1976; 1982) because it would be in the interest of unlinked genetic loci to disrupt the altruist locus either by mimicking the phenotypic marker for parasitic purposes or by modifying the marker so the recognition system is foiled. Indeed, if one gene can evolve to produce such a complex phenotypic effect, alleles at other loci might also do so, resulting in an intragenomic "tug of war" as each gene attempts to influence the behavior of its bearer in its own interest (Alexander & Borgia 1978; Dawkins 1982).
These arguments do not altogether refute the theory, however. The mechanisms of gene recognition (if they exist) will be complex, perhaps involving many genes and supergenes on many chromosomes. For example, large groups of genes could become linked and pleiotropic, producing both feature detectors and altruistic behavior. Moreover, if it is advantageous for a single gene to work for copies of itself, it should be advantageous for all genes to do the same; thus aggregation effects would be expected. This makes it reasonable to talk of overall genetic similarity and not to distinguish between the proportion of shared genes and the probability of a shared altruism gene. Waldman (1987) has developed the preceding argument most fully, pointing out that feature detectors, like other phenotypic characters, can be expected to be the product of multiple alleles; therefore, they accurately reflect the overall genome rather than particular parts. He cites hybridization studies of crickets and frogs showing that hybrid females orient preferentially toward vocalizations produced by hybrid males; this suggests that the mechanisms responsible for their detection and production are genetically coupled.
The strong version of genetic similarity theory thus suggests that some phenotypes are inherently more attractive to organisms than others. The evolutionary origin of such a mechanism could be simple: If like appearance is positively correlated with like genes, any mutation toward preferring like phenotypes would tend to proliferate. If feature detectors exist, they will lead not to kin recognition abilities, but to the discrimination of individuals who share appropriate phenotypic traits.
7.2. Interpretation of data
Several questions can be raised about the data sets. It might be suggested that the blood group similarities are due entirely to the effects of social stratification (i.e., finding oneself in the same location because of similar education and social background) rather than preferential assortment. The data do show that such stratification occurs. For example, Rushton (1989b) found friendship dyads to be significantly more similar in age (r = .64), education (r = .42), and occupation (r = .39) – variables for which the random pairs were unrelated. Moreover, investigators such as Beardmore and Karimi-Booshehri (1983) have found that blood groups are stratified by socioeconomic status (SES). In Britain, blood type A is found to occur more frequently in SES 1, the highest group (57% of the time), than in SES 5, the lowest group (41% of the time).
To test the stratification hypothesis, Rushton (1989b) calculated within-pair differences in age, education, and occupation and did not find them to significantly be correlated with friends' blood similarity scores as they should have been if the statification hypothesis was correct. (Nor was it found that such socioeconomic similarity altered the correlation between friendship similarity and the estimates of genetic influence.) It should also be noted that although evidence does show that stratification effects apply at a single gene locus (e.g., Beardmore & Karimi-Booshehri 1983), in our study we aggregated across 10 loci using 7 polymorphic marker systems on 6 different chromosomes to assess similarity. As mentioned earlier, such blood group differences provide a greater than 95% confidence rating for inclusion in cases of paternity dispute (Bryant 1980) and distinguish reliably between fraternal twins raised in the same family (Pakstis et al. 1972). On the basis of these preliminary attempts to test the social stratification hypothesis, then, such blood group similarities have not been explained.
With respect to the heritability analyses, one possible artifact could have been the differential reliability of the test items. If some had particularly low reliabilities, they would have reduced the estimate of both genetic influence and between-friend similarity, thus giving rise to a spurious correlation between them. For this reason, item reliabilities were calculated in the majority of the studies and the correlations were computed both with and without correction for item reliability. The estimates of genetic influence were consistently found to predict similarity scores, even across quite disparate samples, and even after controlling for within-pair similarity in age, education, and occupation.
Some confusion may result from a mistaken belief that heritability is being equated with importance and that more assortment should therefore occur in physical features than in social variables because the former are more heritable. We have consistently considered it necessary, however, to examine the relation between similarity scores and degrees of genetic influence within homogeneous data sets rather than comparing across heterogeneous traits, for two reasons. First, this presumably holds more constant the (unknown) number of genes involved (hair texture may be highly heritable but may only involve one or two genes; a behavioral item may be less heritable but may involve more genes), and the theory predicts that overall similarity is what matters. Second, since sequential filtering may be involved in the formation of interpersonal relationships, it may be best to make the comparisons at the same level.
Some readers may doubt whether variance in measures can legitimately be apportioned into estimates of genetic and environmental influence. Such uncertainties are common in both the evolutionary and social sciences. Increasingly powerful behavioral genetic techniques are available, however, for testing hypotheses about development (Eaves et al. 1978; Plomin 1986). Twin and adoption studies converge in showing moderate to substantial effects of genetic influence on the transmission of both socially undesirable traits such as crime, obesity, and schizophrenia, and more conventional characteristics such as vocational interests and value systems (Loehlin et al. 1988). In fact, Martin et al. (1986), in their study on the heritability of social attitudes, felt confident enough about the reliability and validity of their measuring instrument, theoretical model, estimation techniques, and sample sizes (3,810 pairs of adult Australian twins plus a secondary analysis of 825 pairs of British twins) to predict between-person correlations in conservatism scores in other relationships if their model was accurate: 0.00 between foster parent and adult foster child; 0.52 between parents and children; and 0.62 for separated monozygotic twins. Recently, a study of 44 monozygotic twins reared apart has confirmed the last prediction of Martin et al., showing an intraclass correlation of 0.53 on a scale measuring traditional moral and family values (Tellegen et al. 1988).
7.3. Mechanisms
Section 3 included a discussion of four different mechanisms by which similarity detection can come about: innate feature detectors, phenotypic matching, familiarity or association, and location. The strongest version of genetic similarity theory suggests that individuals detect genetic similarity in the absence of previous familiarity or other proximal mechanisms. The weaker versions predict that organisms will tend to direct altruistic behavior toward similar individuals when the similarity is detected by means other than genetic recognition. One such means would be phenotype matching based on familiarity with self or kin. Location (section 3.4) may be an additional means. This is likely to be based on similar genes because intelligence, socioeconomic status, values, and vocational interests have been shown to be genetically linked (Loehlin et al. 1988). If there are evolutionary advantages to be derived from the ability to benefit individuals who are genetically similar to oneself, many mechanisms may be involved.
Although the data reported here are only correlational, their pattern allows us to eliminate certain alternative hypotheses as implausible. Reverse causality, for example, whereby mutual interaction within the partnership causes the similarity, cannot be occurring because the genes that cause the blood types and individual traits clearly preceded the onset of the relationship. Moreover, several studies have shown that human assortative mating for similarity on the more genetically based items can be predicted by measures obtained early in or even prior to the marriage and that the degree of resemblance between spouses does not change over time (see Russell et al. 1985, for review).
Hypotheses that "third variables" are causes of both similar genes and social assortment are also unlikely to be valid because they should predict zero correlations between heritabilities and within-trait similarity. Consider the hypothesis that people vary genetically only in the dispositions that cause them to enter different professions, geographical areas, or subcultures within a stratified society, but that thereafter they associate randomly. Associates might thus be expected to be genetically similar compared with individuals chosen at random from the global population, but only as an "artifact" of genes for location. This would fail to explain why similarity is more pronounced in the more genetically influenced components of anthropometric, attitudinal, personality, and cognitive variables. It is unlikely that the patterning of the item heritabilities will be relevant to the individual's location. Moreover, contrary to the stratification hypothesis, Rushton (1989b) showed that the degree of dyadic difference in variables such as level of education was uncorrelated with the similarity scores on the blood tests or with heritability estimates.
The validity of purely environmental theories of the similarity attraction hypothesis also seems unlikely. The widespread social psychological view that people choose similar friends and spouses because of past histories of socialization and enculturation that enable them to reinforce each other more effectively (Byrne 1971) seems to imply a prediction opposite from the one made by genetic similarity theory: There should be a negative, not a positive, correlation between item heritabilities and degree of similarity because the more environmentally modifiable an item, the greater the degree to which it can be shared by people as a result of common experiences. This differential prediction is disconfirmed.
Finally, other, very different hypotheses about human relationship formation can be ruled out. The views that "opposites attract" and that people assort on the basis of complementarity rather than similarity are not supported by the data. Whereas such effects may still occur in individual cases or for particular traits, no evidence for them was found in the present review. This is not to deny the anomalies that exist. For example, Johnson (1984) has shown that about 50% of all civilian marriages in Hawaii are cross-ethnic today. Although data indicate that in such marriages couples appear to "make up" for ethnic dissimilarity by choosing greater similarity in other areas (Ahern et al. 1981), this finding does indeed appear to contradict these views. However, it is also clear that human mate choice depends on many more variables than similarity [see Buss: "Human Mate Preferences" BBS 12 (1) 1989].
8. Distal-proximal levels of explanation
Some readers may be reluctant to accept the hypothesis that there is a genetic component in the "similarity attraction" link, arguing that we like those who are similar to us for cognitive reasons (such as the validation of our view of the world). Others will point to the experimental evidence that familiarity and social learning also mediate relationships (Berscheid 1985; Byrne 1971). There may also be a reluctance to accept the proposition that cognition, choice, and learning are partly derivatives of genetics. A preference for proximal explanations, however, should not rule out more distal factors. Consider Figure 1 in which explanations are depicted as varying in a time dimension; the figure shows that there need be no conflict between levels. Evolutionary biologists do not find the heritability of traits problematic, trait theorists can accept the view that dispositions are modified by later learning, and learning theorists can accept that the products of early experiences interact with subsequent circumstances to produce emotional arousal and cognition.
8.1. Epigenetic rules
In social development Cultural practices and social learning play such an important role in human social behavior that the concept of an "epigenetic rule" (defined as a program, complete with self-correcting feedback system, whereby individual development is guided in one direction rather than another) may help us understand how social influences are genetically channeled (Lumsden & Wilson 1981). Epigenetic factors are most apparent in embryology, where the physical development from fertilized egg to neonate follows a consistent course in normal environments. Channeled development is also illustrated by findings from behavior genetics (Bouchard 1984; Plomin & Daniels 1987; Rushton 1988b). Identical twins show a high degree of concordance in age of onset of puberty, timing of first sexual experience, and menopause. Genetic timing mechanisms also affect cognitive development, as shown in a large sample of twins who were studied from 3 months to 15 years of age; the synchronies between lags and spurts in mental development were found to average about 0.90 for identical twins, but only about 0.50 for fraternal twins (R. S. Wilson 1983).
Psychological development is also guided by epigenetic rules, from sensory filtering through perception to feature evaluation and decision-making (Lumsden & Wilson 1981). For example, whereas the brain perceives variation in luminance along a continuum, it divides hue into categories, using language to do so (Hamad 1987). Many social scientists used to believe that the division into red, green, and so forth were arbitrary, but linguistic and cross-cultural studies have shown that they are in fact closely tied to the inborn physiology of color perception. Epigenetic rules governing more complex social behavior have also been identified. Targeted endpoints appear to underlie the evolutionary function of smiling, attachment, and separation responses in infants (Freedman 1974). [See also Lamb et al: "Security of Infantile Attachment" BBS 7 (1) 1984.] Similar interpretations can be advanced for the life-cycle stages documented to occur in ego development, morality, and psychosocial functioning (Alexander 1987).
Other behavior genetic evidence also provides support for the role of epigenetic rules in social development. For example, whereas small fluctuations in one or two molecules might affect ontogeny, studies show that siblings raised apart for many years in complex environments grow to be significantly similar to each other in a variety of traits; their degree of similarity is predicted by the number of genes they share (Tellegen et al. 1988). It has also been found that the environmental factors influencing development are not shared but are unique to each sibling; that is, the important environmental variance turns out to be within a family, not between families. Such factors as social class, family religion, parental values, and child-rearing styles are not found to have a common effect on siblings (see Plomin & Daniels 1987). This is true even of traits such as altruism and aggression, which parents are expected to socialize heavily (Rushton, Fulker, Neale, Nias & Eysenck 1986).
These data suggest the existence of genetically based stabilizing systems that channel development in such a way that, within the constraints allowed by the total spectrum of cultural alternatives, people create environments maximally compatible with their genotypes (Lumsden & Wilson 1981; Plomin & Daniels 1987; Scarr & McCartney 1983; Rushton, Littlefield & Lumsden 1986). Thus, even within the same rearing environments, genetically different siblings are biased to learn different items of information because they have different sets of epigenetic rules channeling their common environment in individual ways.
In an analysis of the effects of television, for example, Rowe and Herstand (1986) found that although same-sex siblings resemble one another in their exposure to violent programs, it is the more aggressive sibling who (1) identifies more with aggressive characters, and (2) views the consequences of the aggression as positive. Within-family studies of delinquents have found that both intelligence and temperament distinguish delinquent siblings from those who are not delinquent (Hirschi & Hindelang 1977; Rowe 1986). It is not difficult to imagine how intellectually and temperamentally different siblings might seek out different social environments.
Of all the decisions people make that affect their environment, choosing friends and spouses may be one of the most important. Thus, epigenetic rules, by influencing preferences, may prove useful in ordering the hypothetical levels in Figure 1, for any distal "purpose" of the genes must necessarily be mediated by proximal mechanisms.
9. Ethnocentrism and ideology
The potential effects of epigenetic rules on behavior and society may go well beyond ontogeny. Through cognitive phenotypes and group action, altruistic inclinations may find their expression in charities and hospitals, creative and instructional dispositions can give rise to academies of learning, martial tendencies to institutes of war, and delinquent tendencies to social disorder. The idea that genes have such extended effects beyond the body in which they reside, biasing individuals toward the production of particular cultural systems, is a central focus of current thinking in sociobiology (Boyd & Richerson 1985; Dawkins 1982; Lopreato 1984; Lumsden & Wilson 1981; Ruse 1986). [See also multiple book review of Lumsden & Wilson in BBS 5 (1) 1982.]
The implications of the finding that people moderate their behavior as a function of genetic similarity may be far-reaching. They may suggest a biological basis for ethnocentrism, for example. Despite enormous variance within populations, it can be expected that two individuals within an ethnic group will, on average, be more similar to each other genetically than two individuals from different ethnic groups. According to genetic similarity theory, people can be expected to favor their own group over others. Ethnic conflict and rivalry is one of the great themes of historical and contemporary society (Horowitz 1985; Reynolds et al. 1987; Rushton 1986; van den Berghe 1981). Local ethnic favoritism is also displayed by group members who prefer to congregate in the same area and to associate with each other in clubs and organizations. Many studies have found that people are more likely to help members of their own race or country than they are to help members of other races or foreigners, and that antagonism between classes and nations may be greater when a racial element is involved (see Cunningham 1981, for review).
Traditionally, political scientists and historians have seldom considered intergroup conflict from an evolutionary standpoint. That fear and mistrust of strangers may have biological origins, however, is supported by evidence that animals show fear of and hostility toward strangers, even when no injury has ever been received. For example, direct analogies have been drawn between the way monkeys and apes resent and repel intruding strangers of the same species and the way children attack another child who is perceived as being an outsider (Gruter & Masters 1986; Hebb & Thompson 1968). Many influential social psychologists have pondered whether the transmission of xenophobia could be partly genetic. W. J. McGuire (1969) wrote that "it appears possible for specific attitudes of hostility to be transmitted genetically in such a way that hostility is directed towards strangers of one's own species to a greater extent than towards familiars of one's own species or towards members of other species. It would not be impossible for xenophobia to be a partially innate attitude in the human" (p. 265).
Theorists from Darwin and Spencer to Allport and Freud and now Alexander, Campbell, and E. 0. Wilson have considered in-group/out-group discrimination to have roots deep in psychobiology. (For a historical review, see van der Dennen 1987.) Recent developmental psychological studies have found that even very young children show clear and often quite rigid disdain for children whose ethnic and racial heritages differ from their own, even in the apparent absence of experiential and socialization effects (Aboud 1988).
Despite this background, many sociobiologists are equivocal about whether there are fitness implications for ethnic or national preference (Dawkins 1981; Lopreato 1984; Reynolds et al. 1987; Trivers 1985; van den Berghe 1981; E. 0. Wilson 1978). Dawkins (1981) has written: "The equating of 'kin-ship', in the sense of kin-selection, with 'ties of race' appears to result from an interesting variant of what I have called the fifth misunderstanding of kin-selection" (p. 528). Trivers (1985) notes: "Indeed, for large political groups like the United States of America, degrees of relatedness between virtually all members are nearly zero" (p. 135). Many sociobiologists, in an effort to condemn racism, may inadvertently have minimized the theoretical possibility of a biological underpinning to ethnic, national, and racial favoritism (see, for example, almost all contributors to Reynolds et al. 1987). As Hamilton (1987) has remarked, in the context of why kin discrimination, even among animals, is not more readily expected: "in civilized cultures, nepotism has become an embarrassment" (p. 426; see also Alexander 1987, p. 192).
Many of those who have considered nationalist and patriotic sentiment from a sociobiological perspective have emphasized its apparent irrationality. Johnson (1986) formulated a theory of patriotism in which socialization and conditioning engage kin-recognition systems so that people behave altruistically toward in-group members, as though they were genetically more similar than they actually are. In Johnson's analysis, patriotism may often be an ideology propagated by the ruling class to induce the ruled to behave contrary to their own genetic interests, while increasing the fitness of the elite. He noted that patriotism is built by referring to the homeland as the "motherland" or "fatherland," and that bonds between people are strengthened by referring to them as "brothers and sisters."
According to genetic similarity theory, patriotism may often be more than just "manipulated" altruism working to the individual's genetic detriment. It may be an epigenetically guided strategy by which genes replicate copies of themselves more effectively. The developmental processes that Johnson (1986) and others have outlined undoubtedly occur (Rushton 1980), as do other forms of manipulated altruism (Dawkins 1982). However, if these were sufficient to explain the human propensity to feel strong moral obligation toward society, patriotism would remain an anomaly for evolutionary biology. From the standpoint of optimization, one might ask whether evolutionarily stable ethical systems would survive very long if they consistently led to reductions in the inclusive fitness of those believing in them (Alexander 1987; Ruse 1986; E. 0. Wilson 1978). [See also Vining: "Social versus Reproductive Success" BBS 9 (1) 1986.]
If epigenetic rules do incline people toward constructing (Findlay & Lumsden 1988) and learning those ideologies which generally increase their fitness, then patriotic nationalism, religious zealotry, class conflict, and other forms of ideological commitment (even "international socialism") can be seen as genetically influenced cultural choices that individuals make that, in turn, influence the replication of their genes. Religious, political, and other ideological battles may become as heated as they do partly because of implications for fitness; some genotypes may thrive more in one ideological culture than another. In this view, Karl Marx did not take the argument far enough: Ideology serves more than economic interest; it also serves genetic purpose.
Two sets of falsifiable propositions follow from this interpretation. First, individual differences in ideological preference are partly heritable. Second, ideological belief increases genetic fitness. There is evidence to support both propositions. With respect to the heritability of differences in ideological preference, it has generally been assumed that political attitudes are mostly determined by the environment; however, both twin and adoption studies reveal moderate heritabilities of specific social and political attitudes (Table 4), as well as of stylistic tendencies such as authoritarianism and the voicing of extreme views (Eaves & Eysenck 1974; Martin et al. 1986; Scarr & Weinberg 1981; Tellegen et al. 1988).
Obvious examples of ideologies that might increase genetic fitness are religious beliefs that regulate dietary habits, sexual practices, marital custom, infant care, and child rearing (Lumsden & Wilson 1981; Reynolds & Tanner 1983). Amerindian tribes that cooked maize with alkali had higher population densities and more complex social organizations than tribes that did not, partly because cooking with alkali releases the most nutritious parts of the cereal, enabling more people to grow to reproductive maturity (Katz et al. 1974). The Amerindians did not know the biochemical reasons for the benefits of alkali cooking, but their cultural beliefs had evolved for good reason, enabling them to replicate their genes more effectively than would otherwise have been the case.
By the way of objection, it could be argued that although some religious ideologies confer direct benefits on the extended family, ideologies like patriotism decrease fitness (hence most analyses of patriotism would ultimately rest entirely on social manipulation). Genetic similarity theory may provide a firmer basis for an evolutionary understanding of patriotism, for benefited genes do not have to be only those residing in kin. Members of ethnic groups, for example, often share the same ideologies, and many political differences are genetic in origin. One possible test of genetic similarity theory in this context is to calculate degrees of genetic similarity among ideologues in order to examine whether ideological "conservatives" are more homogenous than the same ideology's "liberals." Preserving the "purity" of an ideology might be an attempt to preserve the "purity" of the gene pool. Now that numerous measures of genetic distance are becoming available (Jeffreys et al. 1985; Jorde 1985; Stringer & Andrews 1988) this line of research is becoming feasible.
10. Group selection
Humans have obviously been selected to live in groups, and the line of argument presented so far may have implications for determining whether group selection occurs among humans. Although the idea of group selection, defined as "selection that operates on two or more members of a lineage group as a unit" (E. 0. Wilson 1975, p. 585), and as "the differential reproduction of groups, often imagined to favor traits that are individually disadvantageous but evolve because they benefit the larger groups" (Trivers 1985, p. 456), was popular with Darwin, Spencer, and others, it is not currently thought to play a major role in evolution. Hamilton's (1964) theory of inclusive fitness, for example, is regarded as an extension of individual selection, not group selection (Dawkins 1976; 1982). Indeed, in recent times group selection has "rivaled Lamarkianism as the most thoroughly repudiated idea in evolutionary theory," as D. S. Wilson put it (1983, p. 159). Mathematical models (reviewed in D. S. Wilson 1983) show that group selection could override individual selection only under extreme conditions such Rushton: Genetic similarity as small intergroup migration rate, small group size, and large differences in fitness among groups.
In the recent past it was Wynne-Edwards (1962) who brought the altruism issue to theoretical center stage. He suggested that whole groups of animals collectively refrain from overbreeding when the density of the population becomes too great – even to the point of directly killing their offspring if necessary. Such self-restraint, he argued, protects the animals' resource base and gives them an advantage over groups that do not practice restraint and become extinct as a result of their profligacy. This extreme form of the group selection claim was immediately disputed by other biologists. A great deal of argument and data was subsequently against the idea (Williams 1966). There did not seem to exist a mechanism (other than favoring kin) by which altruistic individuals could leave more genes than selfish individuals who cheated.
One compromise position was offered by E. 0. Wilson (1975), who suggested that although genes are the units of replication, their selection could take place through competition at both the individual and the group levels; for some purposes these can be viewed as opposite ends of a continuum of nested, ever enlarging sets of socially interacting individuals. Kin selection is thus seen as intermediate between individual and group selection. Genetic similarity theory, according to which genes maximize their replication by benefiting any organism in which their copies are to be found, may provide a mechanism by which group selection can be enhanced.
Among humans, the possibility of conferring benefits on genetically similar individuals has been greatly increased by culture. Through language, law, religious imagery, and patriotic nationalism, all replete with kin terminology, ideological commitment enormously extends altruistic behavior. Groups made up of people who are genetically predisposed toward such moral behaviors as honesty, trust, temperence, willingness to share, loyalty, and self-sacrifice will have a distinct genetic advantage over groups that did not. In addition, if strong socialization pressure, including "mutual monitoring" and "moralistic aggression," is used to shape behavior and values within the group, a mechanism is provided for controlling, and even removing, the genes of cheaters (Campbell 1983; Rushton 1980). Several recent analyses suggest that evolution under culturally driven group selection, including migration, war, and genocide, may account for a substantial amount of change in human gene frequencies (Alexander 1987; Ammerman & Cavalli-Sforza 1984; Chagnon 1988; Melotti 1987; D. S. Wilson 1983).
Findlay, Lumsden, and Hansell (in press) have provided a mathematical group selection model in which genetic and cultural parameters are allowed to interact; this goes considerably beyond the previous group selection theories, which examined altruism only as a function of genetic information. The logic of Findlay et al. (in press) is as follows. Since humans have evolved to be "trend watchers," the most common phenotype is likely to be adopted, irrespective of its effect on fitness or on the genetic contribution to altruism. This means that relative to the pure genetic case (on which all traditional group selection models are based), the correlation between genotypic and phenotypic values is reduced. The intensity of individual selection against altruistic genotypes is consequently reduced. On the other hand, because of group structure and frequency-dependent learning, "selfish" groups become more selfish and "altruistic" groups become more altruistic, thereby increasing the variance among groups. Since variance is what drives group selection, the end result is an increase in the efficacy of group selection in biocultural systems compared with purely genetic ones.
Given also that human populations differ, genetically, in the mechanisms underlying their behavior (Rushton 1988c), group selection may be expected to have additional effects on human evolution. Genetic similarity theory, together with the Findlay-Lumsden view, suggests that there should be a correlation between an individual's genotype and the particular trend that is watched. As reviewed earlier, social learning is biased by individualized epigenetic rules. Social psychological studies of cultural transmission make clear that people pick up trends more readily from role models who are both similar and of high status (Bandura 1986). If different ethnic groups learn from different trend-setters, the group selection models of Findlay et al. (in preparation) may have further implications for fitness. In the evolutionary past, for example, those groups that adopted an optimum degree of ethnocentric ideology may have replicated their genes more successfully than those that did not.