Estimating Population Size

Ecologists distinguish population size (N, the total number of individuals) from population density (N per unit area or volume), a distinction already established in Population Density and Dispersion. Estimating either quantity from incomplete data requires a sampling design plus a mathematical rule — an estimator — that converts sample data into a population-level number. Two broad families of estimator dominate practical ecology. Plot-based (spatial) sampling — quadrats and transects — searches a known, bounded fraction of the habitat, counts individuals within it, and multiplies up by the ratio of total area to sampled area. This works well for sessile or slow-moving organisms: plants, barnacles, sessile invertebrates, and any species that can be exhaustively counted within a small plot. Mark-recapture sampling instead treats the population itself as the thing being sampled twice: individuals are captured, marked, released, and the population is resampled later. The overlap between the two samples (marked individuals recaptured) reveals what fraction of the total population the first sample represented. This works for mobile animals — fish, birds, mammals, insects — where you cannot fence off a plot and count everyone inside. Both families rest on the same statistical logic: a small sample, if drawn representatively, contains information about proportions that generalize to the whole. The mathematics differs, but the core assumption is identical — the sample must be representative, meaning every individual (or every unit of area) has an equal, independent chance of being included. Violations of this assumption — non-random dispersion, trap-shyness, differential mortality of marked animals — are the single largest source of error in real-world abundance estimates, and much of the methodological literature is devoted to detecting and correcting for them.

Quadrat sampling proceeds in four steps. First, the habitat area (A_total) is defined and a sampling frame established — a grid over a field, a set of dive transects over a reef. Second, a number of quadrats (plots of fixed area, a_quadrat) are placed at randomly chosen locations (using random coordinates, not observer judgement, to avoid bias toward obviously dense patches). Third, every individual of the target species within each quadrat is counted exhaustively — this is feasible because the quadrat is small enough to search completely. Fourth, the mean density from the sampled quadrats is scaled up:

where d̄ is the mean number of individuals per quadrat. The precision of this estimate depends on the number of quadrats (more quadrats shrink the standard error) and on how patchy the organism's dispersion is — clumped distributions require many more quadrats to achieve the same precision as a randomly dispersed population, because variance between quadrats is higher.

Transects extend this logic linearly: a line or belt is walked through the habitat, and either every individual within a fixed belt width is counted (belt transect, mathematically identical to a long thin quadrat), or the perpendicular distance from the line to each detected individual is recorded (line-transect / distance sampling). The latter corrects for the well-known fact that animals farther from the observer are less likely to be detected, using a detection function g(x) that declines with distance x; the effective strip width is estimated from the shape of this decline, and abundance is back-calculated from it. This is the statistical basis of aerial and ground surveys of large mammals and birds.

Mark-recapture works differently. In the classic Lincoln-Petersen method: 1. A first sample of size M is captured (e.g., trapped, netted). 2. Each individual is marked (tag, fin clip, leg band, dye) in a way that does not harm it or alter its behavior. 3. Marked individuals are released and allowed to remix randomly with the rest of the population. 4. A second sample of size C is captured at a later time. 5. Among the second sample, R individuals bear marks (recaptures). The logic: if marked individuals mixed randomly, the proportion of marked animals in the second sample (R/C) should equal the proportion of the whole population that was marked in the first sample (M/N). Setting these proportions equal:

This is the Lincoln-Petersen estimator. Because it is biased upward when R is small, Chapman's correction (1951) is preferred in practice:

with an approximate variance:

from which a 95% confidence interval can be built as N̂ ± 1.96√Var.

Regardless of method, three assumptions must hold for the estimator to be unbiased: 1. Population closure — no births, deaths, immigration, or emigration between the two samples (or these are explicitly modeled in open-population methods like the Jolly-Seber model, an extension covered at a more advanced tier). 2. Equal catchability — every individual, marked or not, has the same probability of capture in the second sample. Violations include trap-happiness (animals learn to seek bait) and trap-shyness (animals learn to avoid traps), both of which bias N̂. 3. No mark loss or mark effect — marks must persist and must not alter survival, behavior, or detectability.

The mark-recapture logic was first formalized quantitatively by the Danish fisheries biologist C. G. Johannes Petersen in the 1890s, studying plaice (Pleuronectes platessa) in Danish waters. Petersen marked fish with small metal tags, released them, and used recovery rates from later catches to estimate both population size and migration patterns — establishing that a fishery's standing stock could be inferred without draining the sea. This is the origin of the term Petersen estimator. Independently, the American naturalist Frederick C. Lincoln applied the identical logic in 1930 to estimate continental duck populations from band-return data. Waterfowl were banded (ringed) by the thousands across North America; hunters who shot banded birds reported the band numbers. Lincoln reasoned that if X% of all banded ducks were reported shot, and hunters shot some known total harvest of ducks, then the total duck population must be scaled accordingly — precisely the M/N = R/C logic, applied at a continental scale. His 1930 USDA circular, Calculating Waterfowl Abundance on the Basis of Banding Returns, is the direct namesake of the "Lincoln index," and the method is now universally called Lincoln-Petersen in recognition of both independent originators. The critical validating evidence for the method's reliability came from studies where the true population size was independently known, allowing direct comparison against the mark-recapture estimate. In closed experimental populations of Drosophila and in fenced small-mammal enclosures, researchers released a known number of animals, applied mark-recapture protocols blind to the true count, and compared N̂ to the actual N. These validation trials — repeated through the mid-20th century by fisheries biologists and mammalogists — consistently showed:

• N̂ tracks true N closely when R (recaptures) is large (typically R > 20 gives reasonably tight confidence intervals).
• N̂ is biased high when R is small, exactly as predicted by the mathematics of the Lincoln-Petersen ratio (a small denominator inflates the quotient) — the empirical justification for Chapman's 1951 correction, which was tested against simulated and real trapping data and shown to reduce bias substantially when M and C are modest.
• Trap-shy species (e.g., some rodents that learn to avoid traps) produce systematic overestimates of N, because R is depressed relative to what random mixing would predict — a finding that emerged from comparing live-trapping mark-recapture estimates against total enclosure counts in fenced field populations, and that motivated the development of removal methods and behavior-specific capture probability models as corrections.

The convergence of independent fisheries work (Petersen), independent wildlife-management work (Lincoln), and later statistical refinement (Chapman, and subsequently Jolly and Seber for open populations) constitutes the evidentiary backbone: multiple lines of investigation, in different taxa and different decades, converged on the same estimator and the same conditions under which it succeeds or fails.

Example 1 — Quadrat estimate of a clumped wildflower population. A meadow of 5 hectares (50,000 m²) is surveyed for a rare orchid. Fifteen 2 m² quadrats are placed at random GPS coordinates. Counts per quadrat: ``` 0, 0, 1, 3, 0, 5, 0, 0, 2, 0, 8, 0, 1, 0, 4 ``` Sum = 24, mean per quadrat d̄ = 24/15 = 1.6 individuals per 2 m², i.e., density = 0.8 individuals/m².

But notice the data: many zeros and a few high counts (0, 0, 1, 3, 0, 5, 0, 0, 2, 0, 8, 0, 1, 0, 4) — a classic clumped dispersion signature (variance-to-mean ratio well above 1). The variance across quadrats here is large, meaning the standard error of N̂ is large too, and a 95% confidence interval might span from roughly 15,000 to 65,000 — a wide range that reveals the danger of small sample sizes when dispersion is patchy. The fix is not a different formula but more quadrats concentrated proportionally across patch types (stratified random sampling), which is exactly the design refinement that clumped-dispersion species demand, building directly on the dispersion patterns introduced when population density was first defined.

Example 2 — Lincoln-Petersen with Chapman correction, and why it matters. A biologist estimating a trout population in a stream traps and marks M = 30 fish, releases them, waits three days for remixing, then traps a second sample of C = 25 fish, of which R = 3 are found marked. Naive Lincoln-Petersen:

Chapman's correction (better for small R):

The two estimates differ by nearly 50 fish — a 25% discrepancy — purely because R is small (only 3 recaptures). The approximate variance:

giving a standard error of √5985 ≈ 77.4, and an approximate 95% CI of 201 ± 1.96(77.4) ≈ 49 to 353 — an enormous range relative to the point estimate. This demonstrates concretely why field ecologists insist on R ≥ 20–30 recaptures before trusting a Lincoln-Petersen estimate: with so few recaptures, the confidence interval is nearly as wide as the estimate itself, and the biologist should increase trapping effort (larger M or C) before reporting a number to a fisheries manager.

C. G. Johannes Petersen (1860–1928) was a Danish marine biologist working for the Danish Biological Station, whose primary charge was to understand plaice populations to inform sustainable fisheries management — a very practical, economically urgent problem for a small coastal nation dependent on its fisheries. Petersen developed metal fish tags and systematic tagging-recovery programs, turning an intractable census problem (how many plaice live in the North Sea?) into a solvable ratio problem. His work also pioneered benthic community sampling using the "Petersen grab," a device still used to sample seafloor sediment communities — showing that his quantitative instincts extended beyond fish to whole-community ecology.

Frederick Charles Lincoln (1892–1960) worked for the U.S. Bureau of Biological Survey (a predecessor of the U.S. Fish and Wildlife Service) and built the continental duck-banding program into a systematic tool for estimating waterfowl abundance and tracking migration flyways — the origin of the "flyway" concept still used in North American waterfowl management today. His 1930 publication independently derived the same estimator Petersen had used decades earlier for fish, demonstrating the generality of the underlying logic across entirely different taxa and continents.

Fisheries agencies use mark-recapture and acoustic/trawl-based density surveys to set sustainable harvest quotas — directly preventing stock collapse in commercially exploited species like cod, salmon, and tuna. Wildlife conservation programs use camera-trap "recapture" (individual identification via unique coat patterns, as in tigers or jaguars) combined with Lincoln-Petersen-style statistics to estimate populations of endangered large carnivores without physically handling animals. Public health entomology applies mark-release-recapture to estimate mosquito population sizes and dispersal ranges when evaluating vector-control interventions such as sterile-insect releases or insecticide-treated net efficacy. Invasive-species management relies on quadrat and transect density surveys to track the spread of invasive plants and estimate the scale of eradication effort required. Epidemiologists borrow the identical mathematical logic (called "capture-recapture" in human contexts) to estimate the true size of hard-to-count populations — for instance, estimating total case numbers of a disease from the overlap between two independent surveillance systems, a direct methodological descendant of Petersen's plaice tags. These abundance estimates then feed forward into the age-structured demographic models covered in Life Tables and Age Structure and Population Pyramids, where knowing N is the essential first step before asking how that N is distributed across age classes and how it will change over time.

Ask the tutor: How would the Lincoln-Petersen estimate change if the marked animals were more likely to be recaptured than unmarked ones (trap-happiness), and how could a field ecologist detect this bias from the data?