Gene frequencies can change from one generation to another by a process of pure chance known as genetic drift. This occurs because the number of individuals in any population is finite, and thus the frequency of a gene may change in the following generation by accidents of sampling, just as it is possible to get more or fewer than 50 “heads” in 100 throws of a coin simply by chance.

The magnitude of the gene frequency changes due to genetic drift is inversely related to the size of the population—the larger the number of reproducing individuals, the smaller the effects of genetic drift. This inverse relationship between sample size and magnitude of sampling errors can be illustrated by referring again to tossing a coin. When a penny is tossed twice, two heads are not surprising. But it will be surprising, and suspicious, if 20 tosses all yield heads. The proportion of heads obtained in a series of throws approaches closer to 0.5 as the number of throws grows larger.

The relationship is the same in populations, although the important value here is not the actual number of individuals in the population but the “effective” population size. This is the number of individuals that produce offspring, because only reproducing individuals transmit their genes to the following generation. It is not unusual, in plants as well as animals, for some individuals to have large numbers of progeny while others have none. In marine seals, antelopes, baboons, and many other mammals, for example, a dominant male may keep a large harem of females at the expense of many other males who can find no mates. It often happens that the effective population size is substantially smaller than the number of individuals in any one generation.

The effects of genetic drift in changing gene frequencies from one generation to the next are quite small in most natural populations, which generally consist of thousands of reproducing individuals. The effects over many generations are more important. Indeed, in the absence of other processes of change (such as natural selection and mutation), populations would eventually become fixed, having one allele at each locus after the gradual elimination of all others. With genetic drift as the only force in operation, the probability of a given allele’s eventually reaching a frequency of 1 would be precisely the frequency of the allele—that is, an allele with a frequency of 0.8 would have an 80 percent chance of ultimately becoming the only allele present in the population. The process would, however, take a long time, because increases and decreases are likely to alternate with equal probability. More important, natural selection and other processes change gene frequencies in ways not governed by pure chance, so that no allele has an opportunity to become fixed as a consequence of genetic drift alone.

Genetic drift can have important evolutionary consequences when a new population becomes established by only a few individuals—a phenomenon known as the founder principle. Islands, lakes, and other isolated ecological sites are often colonized by one or very few seeds or animals of a species, which are transported there passively by wind, in the fur of larger animals, or in some other way. The allelic frequencies present in these few colonizers are likely to differ at many loci from those in the population they left, and those differences have a lasting impact on the evolution of the new population. The founder principle is one reason that species in neighbouring islands, such as those in the Hawaiian archipelago, are often more heterogeneous than species in comparable continental areas adjacent to one another.

Climatic or other conditions, if unfavourable, may on occasion drastically reduce the number of individuals in a population and even threaten it with extinction. Such occasional reductions are called population bottlenecks. The populations may later recover their typical size, but the allelic frequencies may have been considerably altered and thereby affect the future evolution of the species. Bottlenecks are more likely in relatively large animals and plants than in smaller ones, because populations of large organisms typically consist of fewer individuals. Primitive human populations of the past were subdivided into many small tribes that were time and again decimated by disease, war, and other disasters. Differences among current human populations in the allele frequencies of many genes—such as those determining the ABO and other blood groups—may have arisen at least in part as a consequence of bottlenecks in ancestral populations. Persistent population bottlenecks may reduce the overall genetic variation so greatly as to alter future evolution and endanger the survival of the species. A well-authenticated case is that of the cheetah, where no allelic variation whatsoever has been found among the many scores of gene loci studied.

Selection against one of the homozygotes

Suppose that one homozygous genotype, A₂A₂, has lower fitness than the other two genotypes, A₁A₁ and A₁A₂. (This is the situation in many human diseases, such as phenylketonuria [PKU] and sickle cell anemia, that are inherited in a recessive fashion and that require the presence of two deleterious mutant alleles for the trait to manifest.) The heterozygotes and the homozygotes for the normal allele (A₁) have equal fitness, higher than that of the homozygotes for the deleterious mutant allele (A₂). Call the fitness of these latter homozygotes 1 − s (the fitness of the other two genotypes is 1), and let p be the frequency of A₁ and q the frequency of A₂. It can be shown that the frequency of A₂ will decrease each generation by an amount given by Δq = −spq²/(1 − sq²). The deleterious allele will continuously decrease in frequency until it has been eliminated. The rate of elimination is fastest when s = 1 (i.e., when the relative fitness w = 0); this occurs with fatal diseases, such as untreated PKU, when the homozygotes die before the age of reproduction.

Because of new mutations, the elimination of a deleterious allele is never complete. A dynamic equilibrium frequency will exist when the number of new alleles produced by mutation is the same as the number eliminated by selection. If the mutation rate at which the deleterious allele arises is u, the equilibrium frequency for a deleterious allele that is recessive is given approximately by q = Square root of√u/s, which, if s = 1, reduces to q = Square root of√u.

The mutation rate for many human recessive diseases is about 1 in 100,000 (u = 10⁻⁵). If the disease is fatal, the equilibrium frequency becomes q ≅ Square root of√10⁻⁵ = 0.003, or about 1 recessive lethal mutant allele for every 300 normal alleles. That is roughly the frequency in human populations of alleles that in homozygous individuals, such as those with PKU, cause death before adulthood. The equilibrium frequency for a deleterious, but not lethal, recessive allele is much higher. Albinism, for example, is due to a recessive gene. The reproductive efficiency of albinos is, on average, about 0.9 that of normal individuals. Therefore, s = 0.1 and q = Square root of√u/s = Square root of√10⁻⁵/10⁻¹ = 0.01, or 1 in 100 genes rather than 1 in 300 as for a lethal allele.

For deleterious dominant alleles, the mutation-selection equilibrium frequency is given by p = u/s, which for fatal genes becomes p = u. If the gene is lethal even in single copy, all the genes are eliminated by selection in the same generation in which they arise, and the frequency of the gene in the population is the frequency with which it arises by mutation. One deleterious condition that is caused by a dominant allele present at low frequencies in human populations is achondroplasia, the most common cause of dwarfism. Because of abnormal growth of the long bones, achondroplastics have short, squat, often deformed limbs, along with bulging skulls. The mutation rate from the normal allele to the achondroplasia allele is about 5 × 10⁻⁵. Achondroplastics reproduce only 20 percent as efficiently as normal individuals; hence, s = 0.8. The equilibrium frequency of the mutant allele can therefore be calculated as p = u/s = 6.25 × 10⁻⁵.