Capture-Recapture Check¶
Method — instantiates Exhaustive Population Mapping
Estimates how many units were never seen from the overlap between two independent enumeration passes, without treating either as the final list.
A Capture-Recapture Check is the method that lets an enumeration estimate its own invisible shortfall. Its defining trick is counterintuitive: it reads the size of what you missed from the amount that two independent lists agree on. If two passes over the same population overlap heavily, they are both seeing most of the population, and few units are hiding; if they overlap little, each is catching a different slice and a large unseen remainder is implied. The output is not a list of the missing units — it never names them — but a number: an estimate of the total population, including the units neither pass saw, with an uncertainty band. It answers the question a complete-enumeration project most fears and most struggles to face: "how many did we still miss?"
Example¶
Ecologists need the size of a great-crested-newt population in a protected pond complex, where protection buffers depend on the true count rather than a density estimate. On the first night they capture, mark, and release 80 newts. A week later they return for an independent second pass and capture 100, of which 20 carry the earlier mark. The overlap is the signal: because only one in five of the second catch was previously marked, the marked 80 must represent roughly a fifth of the whole — implying a population near 400.[1]
The check's honesty is in what it refuses to claim. It does not assert it has found every newt; it estimates that roughly 400 exist and that the two nights, between them, saw about 180 distinct individuals — so on the order of 220 were never captured at all. That gap is exactly what a naive "we counted 180" would have buried. The estimate carries assumptions on its sleeve (the population didn't change between nights; a marked newt was no easier or harder to recatch than an unmarked one), and when those assumptions bend, the method reports a wider band rather than a false precision.
How it works¶
- Run two genuinely independent passes. Independence is the load-bearing assumption: whatever made a unit visible to pass A must not also make it visible to pass B.
- Cross-match to find the overlap. Identify which units appear on both lists — the recaptures — using a stable identity key.
- Invert the overlap into a total. A small overlap fraction implies a large hidden remainder; a large one implies near-complete coverage. The estimate follows from the ratio.
- Report a band, not a point. Sampling variability and assumption violations widen the interval; the method surfaces sensitivity to dependence rather than hiding it.
Tuning parameters¶
- Number of passes — two lists give the classic estimate; three or more allow the independence assumption itself to be tested and relaxed, at more cost.
- Independence stance — how strongly you assume the two passes are uncorrelated. Assuming independence when captures are correlated systematically under-estimates the true total.
- Matching strictness — how confidently a unit must appear on both lists to count as a recapture. Loose matching inflates overlap (and undercounts the population); strict matching does the reverse.
- Estimator choice — small-sample-corrected versus naive ratio; the correction matters most when recaptures are few.
When it helps, and when it misleads¶
The check is most valuable exactly where exhaustive enumeration is hardest — hidden, mobile, or hard-to-reach populations — because it puts a defensible number on the residual instead of leaving completeness as an act of faith. It is the discipline that keeps a project from declaring victory when two lists happened to agree merely because they shared the same blind spot.
Its central hazard is correlated capture: when the same trait that hides a unit from one pass also hides it from the other, the passes are not independent, the overlap is inflated, and the population is confidently under-estimated — the failure hides in the direction of comfort.[n1] The method is also blind to units invisible to every source (it can only reason from lists that exist), and a tidy estimate invites false precision over assumptions that never held. The guarding discipline is to engineer genuine independence between passes, prefer three-plus sources so dependence can be measured, and carry the interval forward rather than quoting the midpoint.
How it implements the components¶
A Capture-Recapture Check fills the estimate-the-unseen slice — the diagnostic no single sweep can produce:
missingness_and_overlap_diagnostic— its core output: an estimate of units present but caught by neither pass, derived from the overlap rate, with an uncertainty band.independent_source_crosswalk— it requires two or more independent passes and matches them to measure how many units appear on both.
It counts the missing but does not go find them — the field_verification_protocol and coverage_sweep_plan that add units on the ground are Door-to-Door or Field Sweep's — nor does it join records to reveal new units and cut fieldwork, which is the enumeration_frame_inventory work of its fellow method-type sibling Administrative Record Linkage; and it produces a number, not a picture, so the refresh_and_staleness_policy view of where gaps fall belongs to Coverage Gap Heatmap.
Related¶
- Instantiates: Exhaustive Population Mapping — the check supplies the residual-uncertainty estimate a completeness claim rests on.
- Consumes: two independent enumeration passes — e.g., a Door-to-Door or Field Sweep crossed against an Administrative Record Linkage list.
- Sibling mechanisms: Census Protocol · Enumeration Area Map · Master Unit Index · Door-to-Door or Field Sweep · Administrative Record Linkage · Duplicate Resolution Queue · Coverage Gap Heatmap · Late-Unit Inclusion Window · Enumeration Quality Backcheck
Editorial Notes¶
Form Classification¶
Form family: Analysis, Modeling & Optimization
Rationale: Estimates how many units were never seen from the overlap between two independent enumeration passes, without treating either as the final list, making its operative form a computation, comparison, model, or analytic representation used to infer, estimate, or choose.
Independent corroboration: The frozen evidence defines Capture-Recapture Check as 'Estimates how many units were never seen from the overlap between two independent enumeration passes, without treating either as the final list', so its operative form is Analysis, Modeling & Optimization.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Biology & Ecology
Origin pattern: Single lineage
Present-day reach: Multi-domain
Rationale: Population ecology developed mark–recapture and the Lincoln–Petersen estimator to infer unseen population from overlap between independent capture occasions.
Related originating lineages:
- Statistics & Experimental Design — Sampling theory supplies independence, equal-catchability assumptions, uncertainty, and small-sample correction.
Review resolution: Biology and ecology is the agreed primary lineage because capture-recapture originated as a population-estimation method based on overlap across samples. Statistics formalizes estimators and assumptions, but the check remains a single ecological lineage with multi-domain application.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
The check and its method sibling Administrative Record Linkage both cross-reference independent sources, but for opposite ends: linkage joins lists to add named units to the frame, while capture-recapture reads the overlap statistic to estimate the units no list contains. One grows the roster; the other measures what the roster is still missing.
[n1] The Lincoln–Petersen estimator (with the small-sample Chapman correction) is the foundational mark-and-recapture formula: total ≈ (first catch × second catch) / recaptures. Its key vulnerability is the independence/equal-catchability assumption — when capture on the two occasions is positively correlated, the estimator is biased downward, understating the true population. ↩
References¶
[1] Chapman, D. G. Some Properties of the Hypergeometric Distribution with Applications to Zoological Sample Censuses. University of California Publications in Statistics 1(7), 131–160 (1951). Chapman's capture–recapture estimator makes overlap the population-size signal: a recapture fraction near one fifth means the initially marked 80 represent about one fifth of the population, yielding about 400. The cited work does not supply the mechanism's exact observed one-in-five premise. registry ↩