Effective Population Size

The mathematics of Genetic drift was originally built around an idealized population: the Wright-Fisher model. In this model, generations are discrete and non-overlapping, population size $N$ is constant, mating is random including the possibility of self-fertilization, every individual has an equal chance of contributing gametes, and the number of offspring per individual follows a Poisson distribution (mean and variance both equal to the average number of offspring per individual). Under these assumptions, the rate of drift — how fast heterozygosity is lost, how fast allele frequencies wander, how fast alleles fix or are lost — depends in a precise, calculable way on $N$. Real populations satisfy almost none of these assumptions. Sexes are rarely in a 1:1 ratio contributing equally; some individuals produce far more offspring than others (violating the Poisson variance assumption); population size itself fluctuates from generation to generation, sometimes crashing through bottlenecks; and generations overlap, with parents and offspring breeding in the same time window. Effective population size, $N_e$, is defined as the size of the idealized Wright-Fisher population that would show the same magnitude of drift as the actual population being studied. Formally, you measure some drift-related quantity in the real population — the rate of loss of heterozygosity, the variance in allele frequency change across one generation, or the rate of inbreeding accumulation — and then ask: what value of $N$ in the ideal model would produce that same number? That value is $N_e$. Because $N_e$ is defined relative to a specific process, there are technically several kinds of effective size — inbreeding effective size, variance effective size, and eigenvalue effective size — which coincide exactly only for a population that is constant in size and otherwise idealized. In fluctuating or age-structured populations they can diverge somewhat, but for a foundational treatment the unifying intuition holds throughout: $N_e$ is always the number that determines how strongly chance, rather than selection, shapes the genetic composition of the next generation. It is the currency in which drift is paid.

Step 1: Start from the variance of drift in the ideal model. In the Wright-Fisher model, if an allele has frequency $p$ in a population of $N$ diploid individuals ($2N$ gene copies), the frequency in the next generation is a random draw with expected variance:

This is the baseline "speed of drift" for a population of census size $N$. Real populations are compared against this yardstick.

Step 2: Identify the demographic distortions that change that variance, and solve for the $N$ that would reproduce it. Unequal sex ratio. If $N_m$ males and $N_f$ females contribute gametes, each parent's genetic contribution is not spread evenly — the rarer sex bottlenecks genetic transmission because every offspring must draw one allele from that limited pool. The standard result (Wright, 1931) is:

This is a harmonic-mean-like relationship: it is dragged down disproportionately by whichever sex is rarer. With $N_m = 1$ bull elephant seal effectively siring an entire colony and $N_f = 50$ cows, $N_e = 4(1)(50)/51 \approx 3.9$ — a colony of 51 breeding adults behaves, genetically, like a population of about 4. Variance in reproductive success (family size). The Poisson assumption requires that the variance in offspring number across individuals equals the mean. In real populations — oysters producing millions of eggs with almost all offspring dying, elephant seals where a few bulls sire nearly everyone — the variance in family size ($V_k$) vastly exceeds the mean ($\bar{k}$, normally ≈2 for a stable population). The correction is:

When $V_k$ is large (skewed reproduction), $N_e$ shrinks sharply. When $V_k < 2$ (more equal reproduction than the Poisson expectation — e.g., some managed or monogamous populations), $N_e$ can actually exceed $N$, though this is uncommon in nature. Fluctuating population size across generations. Populations are rarely constant; they boom and crash. Because drift is strongest when populations are smallest, a single-generation bottleneck disproportionately affects the long-run rate of variation loss. The effective size over $t$ generations with sizes $N_1, N_2, \dots, N_t$ is the harmonic mean:

The harmonic mean is pulled down toward the smallest value in the series — a single generation with $N=10$ embedded in nine generations of $N=10{,}000$ collapses the harmonic mean to roughly 91, not anywhere near the arithmetic average of ~9,001. This is precisely why a brief historical bottleneck leaves a genetic signature for many generations afterward, long after census numbers rebound. Overlapping generations and population structure. When generations overlap, or when a population is subdivided into semi-isolated demes with limited migration, the effective size calculation must additionally account for age-specific survivorship/fecundity and for the degree of genetic exchange between subpopulations. Subdivision can, counterintuitively, inflate the total $N_e$ relative to a single panmictic unit (because local drift pushes demes to different allele frequencies, increasing total variance captured across the metapopulation) — while simultaneously making each local deme behave as though it had a much smaller $N_e$ than its census size.

Step 3: Recognize the compounding effect. These factors multiply, not add, in their impact. A population with both a skewed sex ratio and high variance in reproductive success and a history of bottlenecks can have an $N_e$ that is 1-5% of its census size. This compounding is the single most important practical fact about $N_e$: small individual effects stack into large overall reductions.

Because $N_e$ is a theoretical construct — the size of an idealized population, not something you count — it must be inferred from real genetic data, and several classic lines of work established that this inference is both possible and reliable.

Buri's (1956) experimental confirmation of drift dynamics in Drosophila. Peter Buri set up 107 replicate laboratory populations of Drosophila melanogaster, each maintained at a small, fixed census size of 16 breeding adults per generation (8 males, 8 females), all heterozygous for two neutral eye-color alleles ($bw^{75}$ and $bw$) at equal starting frequency (0.5). Over 19 generations, Buri tracked allele frequencies in every replicate line. What he found matched the Wright-Fisher predictions almost exactly: the variance in allele frequency across replicate lines increased generation by generation at very close to the rate predicted by $p(1-p)/2N$ for $N \approx 16$ (though he found the effective size was somewhat below the census 16, roughly 9–10, attributable to unequal family sizes and viability differences even in this tightly controlled lab setting) — and the proportion of lines in which one allele had gone to fixation grew steadily, eventually fixing in the overwhelming majority of lines by generation 19, split roughly according to the initial 0.5/0.5 frequency, exactly as drift theory predicts (the probability of fixation of a neutral allele equals its initial frequency). This experiment is the clearest direct laboratory demonstration that finite population size — not selection — drives predictable statistical loss of variation, and it let researchers estimate $N_e$ empirically from the very variance-among-replicates data it produced, showing $N_e$ can be measured, not merely defined.

Wright's (1931, 1938) theoretical derivations and the sex-ratio/variance formulas. Sewall Wright formalized the mathematics connecting census demographic parameters (sex ratio, offspring variance, population fluctuation) to $N_e$, providing the toolkit still in use. Crucially, Wright showed these effects were not hypothetical fine print — plugging in realistic demographic data from real species (elephant seals, oysters, many others) routinely produced $N_e/N$ ratios far below 1, a prediction later confirmed genetically.

Molecular confirmation: heterozygosity and microsatellite variance in real wild populations. Modern studies estimate $N_e$ from genetic data itself, using methods based on: (1) the rate of linkage disequilibrium decay, (2) temporal variance in allele frequencies sampled across generations, and (3) sibship/relatedness reconstruction from molecular markers (microsatellites, SNPs). These independent genetic estimators, when applied to species with well-known breeding biology, consistently produce $N_e$ estimates dramatically below census counts and match the demographic-formula predictions closely. For northern elephant seals specifically — a species that underwent a severe population bottleneck (down to an estimated 20-100 individuals by the 1890s from sealing) and has since rebounded to over 200,000 — genetic surveys find drastically reduced heterozygosity relative to related species that were never bottlenecked, a molecular fingerprint consistent with an extremely low historical $N_e$ that census rebound alone cannot erase; the harmonic-mean effect means that brief historical bottleneck continues to depress genetic diversity today, generations after the census recovered. Together, these lines of evidence — controlled experimental populations (Buri), formal demographic theory (Wright), and molecular data from real bottlenecked species (elephant seals and many other conservation case studies) — establish both that $N_e$ is a real, measurable quantity and that it is routinely and substantially smaller than $N$.

Example 1: A managed conservation herd with skewed reproduction. A captive-breeding program for an endangered antelope maintains 40 breeding females and 4 breeding males (one dominant male per breeding group of 10 females, so mating is far from random with respect to males). Additionally, because of studbook management to reduce inbreeding, offspring number per female is more even than natural (variance in offspring number, $V_k = 1.2$, below the Poisson value of 2), while males show high variance ($V_k = 5$, since dominant males sire far more offspring than subordinates). Sex-ratio effect alone:

So even before considering variance in reproductive success, a census of 44 breeding adults behaves genetically like a population of about 14.5. If we additionally fold in the separate sex-specific variance-in-offspring corrections (using the general two-sex extension of Wright's formula), the male-side contribution shrinks further because $V_{k,male}=5$ exceeds the Poisson expectation substantially, while the female-side correction is improved slightly by the low $V_{k,female}=1.2$. The compounded outcome typically lands in the range $N_e \approx 10$–13 — meaning this "44-individual" breeding population is, for genetic purposes, roughly a tenth its census size. Conservation managers use exactly this kind of calculation to justify rotating breeding males, equalizing family sizes via studbook allocation, and reporting $N_e$ (not $N=44$) as the number that predicts future inbreeding accumulation and heterozygosity loss. A commonly cited conservation guideline (the "50/500 rule," itself debated and refined) uses $N_e$, explicitly not census size, as the threshold quantity.

Example 2: Reconstructing a bottleneck from the harmonic mean. A population of Channel Island foxes is censused at $N = 5{,}000$ for many generations, then crashes to $N=15$ for one generation due to disease, then rebounds immediately to $N=5{,}000$ and stays there. A geneticist wants to know the effective size relevant to genetic diversity accumulated over a 6-generation window straddling the crash: sizes are 5,000, 5,000, 15, 5,000, 5,000, 5,000.

The four large terms each contribute $0.0002$; summed, $4 \times 0.0002 = 0.0008$. The bottleneck term contributes $1/15 \approx 0.0667$. Total $\approx 0.0675$, divided by 6 gives $1/N_e \approx 0.01125$, so:

Despite five of six generations having a census size of 5,000, the harmonic mean effective size over this window is only about 89 — dominated almost entirely by the single bottleneck generation. This is exactly the pattern seen in real Channel Island fox populations and in northern elephant seals: a brief historical crash leaves a long genetic shadow (reduced heterozygosity, longer stretches of homozygosity-by-descent) that persists for many generations after census numbers fully recover, because the harmonic mean is dominated by the smallest term in the series, not the average.

Sewall Wright, working largely independently of but in creative tension with R. A. Fisher and J. B. S. Haldane, built the mathematical machinery of population genetics in the 1920s-30s, including the diffusion approximation for drift, the concept of adaptive landscapes, and — central here — the formal apparatus for effective population size. Wright's background was unusual for a theoretician: he trained first as an animal breeding geneticist working with guinea pigs at the USDA, which gave him an intimate, hands-on appreciation for how skewed sex ratios, unequal litter sizes, and small breeding herds actually behave — the empirical intuition that let him see why real populations needed a corrected notion of size distinct from the number in the pen. His 1931 paper "Evolution in Mendelian Populations" laid out the core drift mathematics and the shifting balance theory, and his subsequent work through the 1930s-40s refined the sex-ratio and variance-in-offspring-number formulas for $N_e$ still taught today.

Conservation genetics relies on $N_e$, not census counts, to set breeding management and translocation priorities for endangered species — from cheetahs (extremely low genetic diversity linked to historical bottleneck-driven small $N_e$, now a textbook case) to Channel Island foxes to captive breeding programs coordinated through international studbooks. Regulatory frameworks for minimum viable population size explicitly reference effective, not census, numbers.

Fisheries management uses $N_e$ estimates from genetic samples to detect whether heavily harvested marine species (many of which show classic "sweepstakes reproduction," where a tiny fraction of spawners contribute most surviving offspring) are experiencing dangerously accelerated drift and loss of adaptive variation despite still-large census populations — a pattern documented in oysters, cod, and other high-fecundity marine species.

Human population genetics uses long-run estimates of human $N_e$ (roughly on the order of 10,000, far below the billions alive today) derived from genome-wide diversity data to date ancient bottlenecks (e.g., out-of-Africa expansions), calibrate the Neutral Theory of Molecular Evolution's molecular clock, and interpret patterns of linkage disequilibrium used in the coalescent framework for reconstructing demographic history.

Livestock and crop breeding programs monitor $N_e$ to avoid inbreeding depression in intensively selected breeds, since strong selection on a few superior sires or elite cultivars is functionally identical to the reproductive-skew scenario that collapses $N_e$ in wild populations.

Medical and pharmacogenomic research uses estimates of ancestral human $N_e$ to model how efficiently selection could have purged or maintained disease-associated alleles, informing debates over why certain deleterious variants persist at appreciable frequency in human populations.

Ask the tutor: How do the different definitions of effective size (inbreeding, variance, eigenvalue) actually diverge in an age-structured population, and which one should I use for a given research question?