Genetic drift is a mechanism of evolution in which the frequency of an allele, a variant of a gene, changes through chance rather than systematic differences in reproductive success. It arises because each generation carries only a finite sample of the preceding generation’s genetic variation. Drift occurs in all finite populations but generally produces larger fluctuations in smaller populations. Over time, it can eliminate variants or make them universal within a population, reducing genetic diversity. (genome.gov)
Mechanism and distinction from selection
In population genetics, drift describes the random component of genetic change. Even when individuals have equal expected reproductive success, chance determines which reproduce, how many surviving offspring they leave, and which alleles their offspring inherit. Consequently, the next generation’s allele frequencies need not precisely match those of its parents. A population need not experience a catastrophe or decline for drift to occur: ordinary reproduction is sufficient. (cnsgenomics.com)
Unlike natural selection, drift does not consistently favor alleles that improve survival or reproduction. A beneficial allele can disappear by chance, while a harmful allele can increase despite selection against it. Drift and selection operate together, rather than defining mutually exclusive kinds of populations. Their relative importance depends on population size and the strength of the selective difference between variants. (pmc.ncbi.nlm.nih.gov)
Drift changes the representation of existing variants; mutation supplies new variants, while gene flow transfers variants between populations. These processes can replenish variation lost through drift, so the irreversible loss predicted by a drift-only model is not necessarily permanent in a natural population. (pmc.ncbi.nlm.nih.gov)
Mathematical description
The Wright–Fisher model provides a standard mathematical representation. Consider a randomly mating population of (N) diploid individuals, with discrete generations, constant size, and no mutation, migration, or selection. At a locus with two alleles, there are (2N) gene copies. If one allele has frequency (p), its copy number (X) in the next generation follows a binomial distribution:
[ X\sim\operatorname{Binomial}(2N,p),\qquad p'=\frac{X}{2N}. ]
The model treats reproduction as sampling gene copies with replacement from the parental population. (math.wustl.edu)
Its expected value and variance imply
[ \mathbb{E}[p'\mid p]=p,\qquad \operatorname{Var}(p'\mid p)=\frac{p(1-p)}{2N}. ]
Thus drift has no preferred direction in expectation, but individual populations fluctuate. Smaller (N) produces greater variance per generation. The sequence of allele frequencies forms a Markov chain: under these assumptions, its next-generation distribution depends on the current frequency rather than the entire preceding history. (math.wustl.edu)
For illustration, an allele initially at frequency (0.5) has a one-generation frequency standard deviation of approximately (0.112) when (N=10), compared with (0.0112) when (N=1{,}000). These are model-based fluctuations, not directional predictions. (cnsgenomics.com)
Effective population size
Real populations depart from the Wright–Fisher assumptions. Their drift strength is therefore commonly described using effective population size, (N_e): the size of an idealized population showing the same specified genetic behavior. Variance effective size matches allele-frequency fluctuations; inbreeding effective size matches the increase in shared ancestry or decline in heterozygosity. These measures need not coincide in every population. (pmc.ncbi.nlm.nih.gov)
Effective size differs from census size, the number of individuals counted. Unequal contributions to reproduction, an imbalanced breeding sex ratio, and other demographic features can make (N_e) substantially smaller. When population size fluctuates, its long-term effective size is often approximated by a harmonic mean under appropriate assumptions. Small generations consequently have disproportionate influence: later numerical recovery does not automatically restore lost alleles. (pmc.ncbi.nlm.nih.gov)
Fixation and loss of variation
Fixation occurs when an allele reaches frequency one; loss occurs when its frequency reaches zero. In a finite, constant-size drift-only model, these are absorbing states, and eventually one of the competing alleles becomes fixed. The probability that a neutral allele ultimately fixes equals its initial frequency. A single new neutral copy in a diploid population therefore has fixation probability (1/(2N)); most such copies disappear. (math.wustl.edu)
Drift also reduces heterozygosity, a measure of variation based on the probability that two sampled gene copies differ. For constant effective size and no mutation, migration, or selection,
[ \mathbb{E}[H_t]
H_0\left(1-\frac{1}{2N_e}\right)^t. ]
The expected proportional loss per generation is (1/(2N_e)). This is an average over possible population histories, not a requirement that diversity decline smoothly in every individual population. (nature.com)
Bottlenecks and founder effects
A population bottleneck is a sharp reduction in population size. Survivors carry only part of the original variation, and strong drift during the small-population phase can rapidly remove additional alleles. The reduction in genetic diversity may persist after abundance recovers. (evolution.berkeley.edu)
The founder effect occurs when relatively few individuals establish a new population. Their alleles may be an unrepresentative sample of the source population: some variants are absent, while others begin at unusually high frequencies. Subsequent drift can amplify those differences. Bottlenecks and founding events are important demographic settings for drift, rather than prerequisites for its operation. (evolution.berkeley.edu)
Detection and research applications
Researchers can estimate drift strength by comparing allele frequencies at neutral markers across generations or examining associations between variants. Temporal estimates must distinguish genuine population change from the sampling error introduced when only some individuals are studied. Migration and selection can also alter frequencies, complicating interpretation. In conservation biology, effective-size estimates help quantify the expected loss of variation in small populations; in evolutionary research, they provide a framework for interpreting patterns in DNA variation. (pmc.ncbi.nlm.nih.gov)