Purebred dogs are prone to deformities and crippling defects. The biologists John Paul Scott and John L. Fuller performed breeding experiments—that is, artificial selection—on five breeds of dogs:In our experiments we began with what were considered good breeding stocks, with a fair number of champions in their ancestry. When we bred these animals to their close relatives for even one or two generations, we uncovered serious defects in every breed.… [C]ocker spaniels [are] selected for a broad forehead with prominent eyes and a pronounced “stop,” or angle between the nose and forehead. When we examined the brains of some of these animals during autopsy, we found that they showed a mild degree of hydrocephaly; that is, in selecting for skull shape, the breeders had accidentally selected for a brain defect in some individuals. Besides all this, in most of our strains only about 50 per cent of the females were capable of rearing normal, healthy litters, even under nearly ideal conditions of care.Among other dog breeds, such defects are quite common.12
Similar genetic deficits are found in the limited data available on human incest in modern times. The increased infant death rate resulting from first cousin marriages13 is only about 60%. But in a Michigan study14 in the middle 1960s, eighteen children from brother-sister and father-daughter matings were compared with a control group of children from non-incestuous matings. Most of the children of incest (eleven out of eighteen) died within their first six months, or showed serious defects—including severe mental retardation. No history of such defects was found in the parents or their families. The remaining children seemed normal in intelligence and in all other respects, and were recommended for adoption. None of the children in the control group died or was institutionalized. Compared to brother-sister and father-daughter matings in other animals, though, these mortality and morbidity rates seem high; perhaps incestuous unions that produce abnormal children were more likely to come to the attention of the scientists making the study.
The dangers of repeated inbreeding seem so clear that we can safely conclude that unsanctioned sexual unions, impregnations of Queens of Egypt by someone other than the Pharaoh, occurred among Cleopatra’s immediate ancestors. Even a few sibling matings in consecutive generations would probably have led to death, or at least to a Cleopatra very different from the vital individual history reveals to us. But one generation of outcrossing goes far to cancel the previous inbreeding.
Inbreeding is a particular danger in very small groups, because in them it can hardly be avoided. If a new nonlethal mutation occurs in one individual, it either gets lost—because, for example, its bearer has no offspring—or it’s not many generations before it’s in nearly everybody, even if it’s slightly maladaptive. So now most males in the population have, say, a little too much testosterone; the conflicts and the distractions of conflict are taking their toll, and the youngsters are not being cared for as they should. The population has wandered from optimum adaptation; if inbreeding is intense, it may be that eventually none of the members of the group leaves offspring.
If inbreeding weren’t so risky, you might think that small populations are the way to get to gene frequency constellations that are not now especially adaptive, but that will be so at some time in the future. If the population is small, then new mutations or new combinations of letters and sequences in the genetic code can propagate through the entire population in only a few generations. New random experiments in biology are being conducted that could not occur in large populations. As a result, almost always, the group goes hurtling away from optimum adaptation. But comparatively rare genes and gene combinations can be tried out so quickly in a small population that it can swiftly cover a lot of ground in the possible range of gene frequencies.
What’s happening here is described as “accidents of sampling,” which have much more profound consequences in small populations than in large ones: Imagine you’re flipping a coin. Your chance of getting one head in one trial or flip is clearly 50%, one chance in two. The coin has only a head and a tail, and it has to turn up one side or the other. With two flips, the full menu of equally possible outcomes is: two tails, a head and a tail, a tail and a head, or two heads. So your chance of getting two consecutive heads is one in four, or, equivalently, one-quarter, or ½ × ½. With three flips, the chance that they’re all heads is one chance in eight (½ × ½ × ½), or one in 23. You can flip ten heads in a row once in about a thousand trials (210 = 1024). (If we’d witnessed only that trial, we might think you’re phenomenally lucky.) But a hundred heads in a row will take about a billion billion trillion trials (2100 roughly equals 1030)—which is the same as forever.
In small populations major accidents of sampling are inevitable; in large populations they are nonexistent. Were a national opinion poll to query three people only, there would be little reason to believe the results—that is, to think these three opinions adequately sampled the opinions of most citizens. One of the individuals polled might, by accident, be a Libertarian or a Vegetarian, a Trotskyite or a Luddite, a Coptic or a Skeptic—all with interesting perspectives, but none an accurate reflection of the general population. Now imagine that the opinions of these three were somehow proportionately amplified to become the opinions of the population of the United States as a whole; a major transformation in national attitudes and politics would have been worked. The same can be true genetically when a few individuals from a large population establish a new and isolated community.
Accidents of sampling happen when the population sampled is very small. In many elections, when the pollsters sample five hundred or a thousand randomly chosen people, the results repeatedly prove to be representative of the nation as a whole.* With five hundred or a thousand truthful random samplings, the findings are accurate to within a few percent. (The variation expected is the square root of the sample size.) If you ask a large number of randomly selected people, you will reliably sample the average*; if you ask only a few, you may sample atypical or fringe opinions. Pollsters would gladly sample smaller populations; it would save them money. But they dare not—the errors would be too large, the sampled opinions too unrepresentative.
As in opinion polls, so it is in the genetics of populations: With a small enough group, substantial deviations † from the average can be sampled and become established. With mutually isolated small groups, many different sets of gene frequencies get tried out—most maladaptive, but a few, fortuitously, poised for the future. This is called genetic drift.
Or suppose that your name is Theodosius Dobzhansky and that you live in New York City. Even if you have ten sons, your name will continue to be “rare and outlandish” so long as you continue to reside in the big city. But move the family to a small town, have many descendants, and Dobzhansky will eventually become a common and unremarkable name. Similarly, any extraordinary hereditary predisposition in the Dobzhansky genes will affect only a tiny fraction of the population while you’re in New York, but might in a few generations become a major genetic feature of the citizenry of the town.15