Fish tag–recapture studies estimate how many fish may occupy a defined population by asking a deceptively simple question: after a known group has been marked and released, what fraction of a later sample carries those marks? The calculation is easy to demonstrate, but translating observed recaptures into abundance requires a clear population boundary, credible mixing, comparable capture probabilities and an account of tags that disappear or go unreported. An abundance figure with no account of these conditions can look exact while answering the wrong question.
This guide explains fish mark–recapture from first principles, including the Lincoln–Petersen ratio, the Chapman small-sample estimator, field recording, tag retention, reporting rates, estimation uncertainty and the distinction between abundance and biomass. Every number in the worked examples is hypothetical teaching data. None is a reported fishery result. We will follow each denominator through its meaning rather than treating tag returns as interchangeable with recaptures observed in a controlled second sampling event.
The question for the reader is not only ‘How many fish were tagged?’ It is ‘Which animals were eligible to appear in both samples, what made a tag visible, and what biological population did that sampling process represent?’ The answers decide whether a neat equation is informative, biased, or inappropriate. We also show how to report a limited result honestly, test alternative assumptions and pass estimates into a wider assessment without claiming that every fish in the sea was counted.
Start with the estimand, not with the formula
An estimand is the quantity an investigation intends to estimate. For a closed two-occasion fish study it might be the number of catchable fish of one species within a particular enclosed lake during a specified short interval. It is not automatically the number of fish in the entire watershed, all age classes, or every fish that will ever pass the sampling gear. Write the spatial boundary, dates, biological size range and capture method before processing a single tag code. Otherwise researchers can accidentally change the population halfway through their calculations.
An observed recapture count is a property of a sampling protocol; a population abundance estimate is an inference conditional on that protocol. If a lake contains juveniles too small for the net, the method may estimate only the vulnerable part of the population. If tagged fish use a channel that the second survey never enters, their presence or absence in the catch says more about spatial coverage than total abundance. Good interpretation states the frame visibly: who was eligible, when and where, and by which fishing gear.
A stable notation that keeps the denominator honest
Let M be the number of fish actually marked and effectively released into the target population after the first occasion. Let C be the number of eligible fish examined in the second capture occasion, including both marked and unmarked animals. Let R be the number of those C animals identified as carrying a valid mark from the first occasion. Let N be the unknown abundance at the study’s target time and within its defined population. All four variables are counts of individual fish, not kilograms, hooks, hours or net hauls.
Some references use T for tagged fish, n for a later sample and m for recoveries. Those symbols can be perfectly valid, but moving among documents without a notation table invites mistakes. In particular, a tag returned by a fishing customer months later is not automatically an R observation from a controlled random sample of C fish. It may require a separate recovery and reporting model. The labels should preserve how the observations were generated, not merely make the numbers fit a familiar equation.
Map the evidence chain from animal to analysis
The first capture yields individuals that can be identified, assessed for condition and marked. The release record establishes the time and place at which each animal re-entered the study. A later sampling event selects a set of eligible animals under a stated gear and effort protocol. Each encountered fish must be checked for marks, including tags that are damaged, obscured or incorrectly read. Only then can the analyst classify a valid first-occasion recapture. The equation uses the resulting records, but quality begins much earlier on deck or shore.
Imagine a team initially handles 220 fish. Twelve are found outside its target length range, five are held for veterinary observation and three tags cannot be securely attached. The number M must not be reported as 220 merely because 220 fish were touched. If 200 eligible individuals are ultimately tagged and released alive under the protocol, M is 200. This bookkeeping is not cosmetic: including ineligible or unreleased fish inflates M and therefore can inflate inferred abundance.
Design a marked group that represents the later sampling frame
A tag–recapture study is most interpretable when fish in the initial marked set are exposed to the same later encounter process as unmarked fish. Sampling only one shallow shoreline while the second occasion covers the whole basin can violate that comparison. Marking only large fish and recapturing all sizes creates a particularly clear mismatch. A defensible design can stratify by habitat, size or subarea, but the sample definition and estimator must then acknowledge the strata rather than assuming one perfectly mixed pool.
Capture gear matters because nets, traps, hooks and electrofishing each encounter particular animals. Fast swimmers, sheltering fish and fish that have learned to avoid gear may not have equal capture probabilities. An investigator should record the gear configuration, soak time or tow duration, weather and site selection, and whether individuals could be captured more than once on the same occasion. Those details let a later reviewer distinguish a population change from a sampling change.
The classical Lincoln–Petersen proportion
Under a simplified closed-population model with random mixing and equal capture probabilities, the marked fraction in the population is approximately M/N, while the fraction observed in the second sample is R/C. Equating these proportions gives M/N ≈ R/C and the familiar estimate N ≈ M × C / R when R is positive. This is a proportionality argument, not proof that the system has all the required properties. Its power comes from a defined probability model and a clear sampling frame.
The ratio also reveals why small numbers of recaptures create problems. R is in the denominator. When R decreases, the estimate rises sharply; when R equals zero, the simple estimator is undefined. A result based on one or two observed marks is extremely sensitive to one reader missing a tag, one fish being misclassified or one individual changing its availability. We should inspect R and its detection process before getting impressed by a large calculated N.
Fully checked example: two planned sampling occasions
Suppose an entirely hypothetical lake survey marks and releases M = 200 eligible fish. After a suitable mixing interval, a separately selected second survey examines C = 150 eligible fish, of which R = 30 carry valid first-occasion tags. The second-sample marked proportion is 30/150 = 0.20. The first-sample marked fraction would therefore be about one fifth of the population under the model. The elementary Lincoln–Petersen result is 200 × 150 / 30 = 1,000 fish.
Check the units and the interpretation. Multiplying 200 fish by 150 fish and dividing by 30 fish produces a count of fish, so the output is an estimated number of individuals. It is not 1,000 kilograms, a catch quota or a forecast for next year. The calculation describes the eligible population represented by the two sampling occasions, and only if the statistical assumptions are sufficiently defensible. The mark proportion can be independently checked: 200/1,000 = 0.20, matching 30/150.
The value 1,000 is not a literal census. If tags make fish shy or if tagged individuals cluster where the second survey samples, the ratio will misrepresent the true marked fraction. Treat the result as a model-dependent estimate and pair it with sample details. The most useful report includes M, C and R beside N so readers can retrace the inference without trusting a rounded headline on its own.
Why the Chapman correction is often preferred
The Chapman estimator for the simple two-sample closed-population case is N̂ = [(M + 1)(C + 1)/(R + 1)] − 1. Its small-count adjustment is useful because the uncorrected ratio M C / R has finite-sample bias and becomes undefined when R is zero. The correction does not cure unequal catchability, tag loss, movement or an incorrect sampling frame. It addresses a particular statistical issue under the underlying two-sample model, not every biological imperfection.
For M = 200, C = 150 and R = 30, calculate (201 × 151)/31 − 1. The numerator is 30,351; dividing by 31 gives approximately 979.0645; subtracting one yields 978.0645 fish. A sensible display is about 978 eligible fish, or roughly 980 given the scale of uncertainty. The result differs modestly from the classical 1,000 because there are 30 recaptures; for very small R, the difference can become far more consequential.
The +1 terms should be retained through the calculation instead of rounding each factor prematurely. The estimator applies to the same kind of two-occasion protocol as the proportion argument, with random mixing and comparable capture probabilities. It does not justify changing a commercial tag-return data set into a textbook two-sample survey. Before choosing Chapman, ask whether all C fish were inspected in a consistently designed second occasion and whether the recapture process genuinely matches that model.
Variance is part of the result, not an appendix
For this simple Chapman model, an often-used variance estimate is [(M + 1)(C + 1)(M − R)(C − R)] / [(R + 1)²(R + 2)]. For the hypothetical M = 200, C = 150 and R = 30 example, the numerical estimate is approximately 20,134 fish squared, giving a standard error of about 141.9 fish. The squared unit matters: variance is in fish², while the square root restores the same count unit as abundance.
An illustrative symmetric normal calculation would be 978.1 ± 1.96 × 141.9, roughly 700 to 1,256 fish. This is a rough teaching approximation, not a recommended universal confidence interval, and it ignores systematic bias from broken field assumptions. Exact, profile-based, Bayesian or other interval approaches may be more appropriate depending on study design and recapture count. A precise-looking confidence interval cannot rescue a poor sampling process; it quantifies only the uncertainty included in its model.
Zero and very few recaptures change what can be inferred
Consider a different hypothetical dataset: M = 80, C = 60, R = 2. The classical estimate would be 80 × 60 / 2 = 2,400 fish. Chapman gives (81 × 61)/3 − 1 = 1,646 fish. That substantial difference is a warning that the estimate depends strongly on a tiny number of observed overlaps. Neither value deserves to be displayed as a settled census. The sample could be too small, poorly mixed or unrepresentative, and interval uncertainty will be wide and asymmetric.
With R = 0, the Chapman arithmetic still returns (M + 1)(C + 1) − 1, but this does not mean the true population is identified by that finite number. A single no-recapture event can be consistent with a wide range of sufficiently large populations and with failures of mixing or detection. Analysts should use an appropriate interval or likelihood formulation and report that the data are weakly informative rather than treating a mechanical plug-in output as a reliable abundance measurement.
Choosing sample sizes and survey intervals
The design objective is not merely to tag as many fish as possible. It is to obtain an informative recapture distribution without harming the population or changing behavior. A pilot study may help assess likely overlap, fish handling survival, tag retention and achievable inspection effort. If recaptures are rare, adding more randomly selected second-occasion observations can be more useful than repeating an already unrepresentative marking site. The study must still be feasible, permitted and ethical.
The interval between occasions poses a trade-off. Fish need sufficient time to mix, yet long intervals increase opportunities for birth, death, harvest, migration, tag shedding and changes in vulnerability. The appropriate period depends on fish biology and system geography. A week in one enclosed setting can be very different from a week in a migratory coastal stock. Document both the elapsed time and evidence supporting the assumptions, rather than simply asserting that ‘the tags had time to mix.’
Population closure: what must remain stable
A closed population has no relevant additions or losses between the two occasions for the abundance parameter being estimated. In practice, researchers distinguish geographic closure from demographic closure. Geographic closure concerns immigration and emigration; demographic closure concerns births and deaths, including fishing mortality. Fish can also move between habitats while remaining physically inside the lake, making them differently available to gear. That internal redistribution still matters for detection even when the total number stays unchanged.
Closure is an assumption about the actual target population and period, not a claim that fish never move. If a temporary barrier separates two coves after initial marking, later samples may encounter few tags despite no net loss from the whole lake. A high inferred abundance could then be an artefact of missed movement. Site maps, receiver detections, environmental records and repeated survey coverage can help evaluate whether one closed-population estimate is an appropriate description of the experiment.
Mixing and equal catchability are different questions
Random mixing means the initial marked animals are not systematically clumped relative to the later sample. Equal catchability means marked and unmarked eligible fish have comparable probability of entering that later sample. A school of tagged fish may mix broadly across a basin yet still avoid a net after their first capture. Conversely, tags may have no behavioral effect but be concentrated in one region due to a restricted release location. Both situations can distort the observed marked fraction for different reasons.
An informative field report describes the release distribution, target habitat, gear and any evidence of tag-induced behavior. If possible, assess whether recapture fractions differ by size, site or fishing method. Such diagnostics do not prove assumptions in a mathematical sense, but they can reveal obvious contradictions. A student should learn to identify the direction of possible bias: if tagged fish are more likely to be caught than others, R/C becomes too high and a naive abundance estimate tends to be too low.
Tag loss is a measurement problem with biological consequences
A tagged animal may survive but shed its mark, suffer damage to an external tag, or retain a tag that an observer fails to recognize. In a two-sample estimator, an effectively unmarked former tagged fish can be incorrectly placed in the unmarked group. That reduces observed R relative to the number of fish that truly participated in the first marking occasion. Because R sits in the denominator of both classical and Chapman estimators, uncorrected loss can make abundance appear higher than it is.
Tag retention varies with tag design, fish morphology, handling and time. A double-tagging sub-study, conducted with attention to animal welfare and tag dependence, can supply information about loss. NOAA-hosted research on blue sharks illustrates why different tag designs have different shedding performance; its empirical findings are species- and tag-specific and must not be transplanted as universal correction percentages. A useful report identifies which design was used, whether loss was tested and how uncertainty in retention enters the analysis.
Distinguish detection failures from physical tag shedding
A tag can remain physically attached yet fail to enter the data because an observer overlooks it, a code becomes unreadable or a database rejects the identifier. These failures have different repair strategies from actual shedding. Observer training, dual reading, clear photograph protocols and identifier validation can reduce missed or mistaken detections. For an electronic tag, receiver coverage, power, code collisions and device failure also matter, but those mechanisms should not be collapsed into a single unexplained ‘tag lost’ category.
Suppose ten recaptured fish have damaged tags, and the team cannot verify whether eight or nine are valid first-occasion marks. Do not quietly assign all ten to R or disregard them without consequence. Preserve an unresolved category, document the adjudication rule and evaluate sensitivity. One plausible classification could produce R = 30 and another R = 31, leading to different estimates. In a small study, that one-count change may dominate the arithmetic more than an extra decimal of precision.
Reporting rates are not recapture probabilities
In a controlled second survey, trained researchers may inspect every eligible fish they catch. In a commercial fishery tag-return programme, however, an animal may be captured, its tag seen and still not reported to researchers. A tag-reporting rate is the fraction of recoverable encounters that reach the intended reporting system under a defined process. It is not the same as the fraction of all fish that were caught, and it is not always equal to the fraction of tags physically retained.
NOAA Fisheries has published work on sablefish reporting rates showing why this component must be estimated explicitly, with differences across regions and years. Rewards, observers, fish handling and outreach can affect whether an encounter becomes a reported return. Those historical empirical rates are not constants to borrow for a new species or contemporary fishery. Analysts may use high-reward tags, controlled comparisons or observer data to estimate reporting, but the method and associated uncertainty need to be recorded.
For illustration only, suppose 40 tag recoveries are reported and an independent experiment suggests 60% reporting. A naive expansion would give 40/0.60 ≈ 66.7 expected encountered recoveries under simple conditions. This does not convert the dataset into a valid Chapman R of 66.7: the underlying sampling event, fishing effort, movement and tag survival still require a compatible recovery model. Report a reporting-adjusted count only within the model that defines it, not as a substitute for a random second capture sample.
Capture, survival and reporting form separate paths
Think of the evidence as a sequence of conditional events. A fish must remain in the relevant population and be vulnerable to sampling. It must be encountered, its tag must persist, the mark must be recognized, and the information must be recorded or submitted. These are distinct probabilities, often changing over time. Multiplying them in a simplified illustrative model can clarify where missing observations arise, but only if independence or dependence assumptions are made explicit.
A common interpretation error is to see a low number of returned tags and conclude immediately that the population is enormous. Perhaps very few tagged fish were harvested; perhaps many recovered tags were not reported; perhaps tags were shed; perhaps the tagged fish moved to a less sampled region. The correct diagnosis requires discriminating evidence. A tag-return curve alone may not identify abundance, mortality, migration and reporting simultaneously without outside information or strong modeling assumptions.
Handling effects and post-release survival
Tagging itself can affect a fish’s survival, behavior and subsequent vulnerability to capture. Holding time, water temperature, gear injury, air exposure and tag placement may change those effects. A valid study should describe the relevant permits, welfare protocols and release criteria, and avoid tagging animals for whom the intervention is inappropriate. Ethical practice is not separate from statistical validity: an animal that dies immediately after release cannot mix as an ordinary member of the intended population.
If ten initially tagged fish die before a second sample, adjusting the effective number marked might be necessary in a model that has trustworthy survival information. Yet mechanically subtracting ten from M without accounting for when or how mortality occurred may produce another error. A correction should be tied to measurements, survival studies or an explicitly modeled process. The first duty is to reduce preventable harm and report its possible effect, not to assume every released fish behaved normally.
Avoid double counting and identifier collisions
A tagged fish may be encountered more than once during a sampling occasion. If the estimator requires unique individuals in C, counting the same recaptured fish twice inflates both the sample total and possibly R. Assign one record per eligible individual per defined occasion, with repeated observations linked to a shared identifier. This is especially important in trap arrays where fish can leave and re-enter multiple times. A computer table of encounters is not necessarily a table of unique fish.
Tag codes themselves can collide through transcription errors, reused identifiers or incomplete field notes. Standardized formatting, check digits where supported, photographs and a separate tag deployment registry reduce these risks. Never treat a merely similar code as proof of identity. An invalid or uncertain match belongs in an auditable review category. Analysis scripts should flag impossible sequences, such as a tag recaptured before its release or a supposedly single-use tag allocated to two animals.
Size, age and sex may demand strata
Fish abundance is rarely sampled with perfect uniformity across size classes. Nets exclude some small individuals; older fish can use different habitats; seasonal sex differences can affect movement or catchability. One aggregate M, C, R calculation can conceal these contrasts. If the scientific target includes several biologically distinct groups, define strata in advance where feasible, calculate within valid sampling frames and combine estimates only with an appropriate design and variance treatment.
Consider two hypothetical length classes. Small fish may be easy to mark but hard to recapture because the later gear has larger mesh. Large fish may produce most recovered tags even if they are less numerous. Pooling both groups and assuming equal capture probabilities is unjustified. A useful diagnostic table displays marked, examined and recaptured counts by size class and reports where cells are too sparse for reliable separate estimates. Do not invent unobserved size classes to make totals appear complete.
Spatial movement can make one abundance result ambiguous
Tag-return positions can reveal movement as well as abundance, but different questions call for different designs. If fish leave the study region and later return, a closed estimate for the original boundary is difficult to interpret. If the goal is exchange between fishing zones, a multistate or movement-aware model may be more useful than a single two-occasion Petersen ratio. The analyst should not call fish ‘dead’ simply because their tags no longer appear in a particular survey area.
Spatial clustering can also hide information. An initial tagging vessel may work near a reef while later sampling preferentially covers sandy habitat. A low R might indicate habitat segregation rather than low stock density. Present release and encounter locations in the same coordinate system, describe survey effort and examine coverage gaps. If the observations cannot distinguish movement from population size, the report should preserve that ambiguity rather than assigning it an arbitrary correction factor.
Time, season and changing availability
A cohort marked before spawning may behave differently during an autumn survey, while a seasonal fishery may encounter only a transient part of a stock. The abundance implied by a tagging model should therefore carry an explicit time definition. Using tag returns collected over many months without modeling exposure, movement, fishing effort and loss can mix different probabilities into one denominator. More elapsed calendar time does not automatically mean more independent information about the original population.
Survey schedules can be designed around stable biological windows, but ‘stable’ must be justified by species and environment. Temperature changes can shift catchability, a flood can alter lake connections and management closures can change fishing effort. When multiple occasions are available, analysts can consider models that allow time-varying capture or survival probabilities. Those models require sufficient data and diagnostics; complexity is not a substitute for field evidence.
What uncertainty should be reported?
Sampling uncertainty comes from seeing only a fraction of eligible fish. Process uncertainty comes from biological changes and movement. Measurement uncertainty arises when tags are missed, misread or shed. Model uncertainty concerns the validity of equal-catchability, closure and other assumptions. A single confidence interval from the Chapman variance formula addresses only a portion of this full uncertainty landscape. Users should not read it as an interval guaranteed to contain the truth under any field conditions.
One practical sensitivity exercise repeats the main calculation under defensible alternative classifications or assumptions. For M = 200 and C = 150, changing validated R from 30 to 29 changes Chapman abundance to (201 × 151)/30 − 1 = 1,010.7; changing R to 31 gives (201 × 151)/32 − 1 ≈ 947.5. Both differ from 978.1 without changing fish biology. This shows why mark identification deserves attention and why a single extra decimal in the displayed estimate is misleading.
A documented spreadsheet or script can be independently checked
A reproducible workflow has one row per fish–occasion encounter or one clearly documented aggregation rule. It validates unique tag IDs, species, release status, eligibility, sampling dates and recapture recognition. A second table links analysis-ready M, C and R to the specific records used. Store both the numerator and denominator and record exclusions. The result should be reproducible by another analyst from the same source data, without relying on an undocumented spreadsheet cell.
Before computing N, run logic checks: R cannot exceed M or C; all counts must be nonnegative integers; a release date cannot follow a recapture date; and fish excluded as ineligible must not reappear silently in the sample total. Verify that C is a count of examined individuals rather than an effort measure such as trap nights. After calculating the estimate, perform the inverse marked-fraction check and preserve the exact formula and software version used.
Turn abundance into biomass only with a valid weight basis
An abundance estimate is a count. Biomass is a mass. If the target consists of fish with a representative mean individual mass w̄, a simplified biomass estimate is B ≈ N × w̄. For example, an estimated 978 fish with a legitimately representative mean mass of 0.75 kilograms would correspond to approximately 733.5 kilograms. That multiplication assumes the weight sample represents the same eligible fish population. It is not permission to combine an all-ages fish count with the mean weight of selected adult landings.
Real populations vary by length, age, sex and condition. Stratified abundance estimates and size-specific weight models can be preferable to one pooled average. The uncertainty in mean weight also contributes to biomass uncertainty and may be correlated with sampling. When a study reports only tagged fish lengths, avoid extrapolating a whole-stock biomass unless the relationship to unsampled groups has been established. The nearby eduKateSG fish length–weight guide explains why an apparently straightforward mass conversion requires careful definition.
Compare with trawl, acoustic and age-based evidence
A swept-area trawl survey uses catch and a gear footprint to build a density index or an area-expanded estimate. Acoustic surveys use backscatter, target strength and calibration to infer fish density in sampled water. Age–length keys translate observed size distributions into estimated age composition. Tag–recapture instead uses known marking history and later overlap. The methods do not measure identical things, and their biases can differ. Agreement between estimates is encouraging only when spatial, temporal and biological coverage are comparable.
If tagging suggests one abundance while a trawl survey suggests a substantially different number, do not average them immediately. Ask whether both target the same species, sizes, region and dates. Evaluate whether tags alter behavior, whether trawls miss particular habitats and whether acoustic targets are correctly identified. A productive triangulation identifies why estimates disagree and what new data could discriminate among explanations; it does not manufacture a consensus value without an observation model.
Communicate a result without converting an estimate into a fact
A defensible one-sentence report could say: ‘In a hypothetical closed two-occasion survey, 200 fish were marked, 150 were examined later and 30 were recaptured, producing a Chapman abundance estimate of about 978 eligible fish under the stated mixing and equal-catchability assumptions.’ This communicates the data, method and scope. It is much stronger than ‘The lake contains exactly 978 fish’, which hides sampling uncertainty and the possibility that unobserved classes of fish were excluded.
Under uncertainty, wording becomes part of scientific accuracy. ‘Estimated’, ‘within the sampled size range’, ‘assuming no tag loss’ and ‘in the defined study period’ are meaningful qualifiers, not verbal clutter. Avoid implying that a tag-return estimate establishes a management quota, total annual production or animal welfare outcome. If managers need harvest advice, they will require additional population dynamics, recruitment, mortality, risk tolerance and policy considerations.
Independent transfer exercise 1: recompute and audit
Try this invented problem without looking at the answer first. A survey successfully releases M = 120 marked fish. The second sampling event examines C = 90 different eligible individuals and detects R = 18 true first-occasion recaptures. Calculate both the classical Lincoln–Petersen and Chapman estimates, check the marked fractions and explain why neither estimate is an exact census. Be explicit about which variable counts unique fish and whether recaptures are included in C.
The classical calculation is 120 × 90 / 18 = 600 fish. Chapman is (121 × 91)/(19) − 1 = 11,011/19 − 1 ≈ 578.53 fish. The second sample has a marked proportion of 18/90 = 0.20. The classical implied population fraction is 120/600 = 0.20, confirming the ratio. Chapman is slightly smaller because of its adjustment. Both estimates apply to eligible, catchable individuals under a defined two-occasion model, not every animal outside the sampling frame.
A careful audit would confirm that all 120 were actually released and that all 90 later fish were examined once. It would also ask how the 18 marks were authenticated and whether enough time passed for mixing without meaningful loss or entry from the population. Those checks are not supplementary to the answer; they are what allow the numerical answer to represent a biological quantity.
Independent transfer exercise 2: a missing mark
Suppose the same study rechecks its photographs and finds that one fish previously classified as unmarked carried a valid first-occasion tag. Now R = 19 while M = 120 and C = 90 remain unchanged. Recompute Chapman, identify the direction of change and explain the causal reason. The new estimate is (121 × 91)/20 − 1 = 549.55 fish, compared with about 578.53 before the correction.
The abundance estimate decreases because the second sample is now understood to contain a larger fraction of marked individuals. The corrected observation suggests that the original marked group represented a larger fraction of the target population, so a smaller N is needed to explain the overlap. Nothing about the actual fish stock changed during the photograph review. Only the measurement record changed, illustrating why detection quality and traceable revisions belong in the scientific report.
Independent transfer exercise 3: reporting is not recapture
A field team tags 500 fish and receives 25 tags from a commercial fishery over six months, but it does not know how many eligible fish were caught and checked for tags. A colleague proposes calculating N = 500 × 25 / 25 = 500 because 25 tags were reported. Is that valid? No. The 25 reported returns are not a known second-sample total C and an observed marked subset R; they are recoveries filtered by fishing, survival, retention, detection and reporting.
To approach abundance, the team would need a suitable model and information on effort, encounter probability, tag reporting, mortality and movement, among other relevant quantities. A study might estimate exploitation or movement under an appropriately specified tag-return model instead. The transferable lesson is that the same symbol cannot be assigned to two incompatible data-generating processes simply to make a formula computable.
Independent transfer exercise 4: abundance versus mass
A hypothetical study estimates N = 1,800 eligible fish. A separate valid random sample of the same eligible population has mean mass 0.40 kg. Under the simplified assumption that the mean mass is representative, estimate total biomass: 1,800 × 0.40 = 720 kg, or 0.72 metric tonnes. If the weight sample instead contains only large spawning adults while N includes juveniles, that 720 kg is not a justified whole-population biomass estimate.
Explain the repair: obtain mass information representative of the modeled fish population or estimate abundance by relevant size groups and use appropriate mass-per-group relationships. Also state how uncertainty in count and mean weight affects the total. It is tempting to label any N × weight calculation ‘biomass’, but a checked unit is only necessary, not sufficient, for biological validity.
A practical failure atlas for reading tag–recapture studies
Failure one is treating every fish handled as successfully released; repair it with a release-status audit. Failure two is using the number of returned tags as the full second sample size; repair it by reconstructing the actual encounter and inspection process. Failure three is overlooking fish that shed tags; repair it with retention evidence and a compatible model. Failure four is assuming that one cove represents an entire lake; repair it with spatial design and coverage diagnostics. Failure five is translating a count into tonnes without representative weights.
Failure six is asserting that the Chapman correction removes all bias; it only makes a model-specific small-sample adjustment. Failure seven is reporting a symmetric interval as a complete uncertainty statement despite low recapture counts or unequal catchability. Failure eight is quietly smoothing over invalid tag codes. Failure nine is equating no recapture with no fish. Failure ten is mistaking the visible dates of tag reports for a single closed sampling occasion. These failures differ, so each requires its own investigation rather than a universal correction factor.
Frequently asked questions from careful readers
Does a larger number of tags always improve the estimate? Not automatically. Greater marking effort can increase potential overlap, but poor release survival, concentrated locations or a mismatched second survey can still dominate bias. A smaller well-designed study with documented coverage may yield a more interpretable result than a large opportunistic programme. Sample size and sampling quality are separate dimensions of evidence.
Can a fish that was captured twice count twice? In a two-occasion estimate based on unique individuals, no: the analyst must distinguish encounter records from distinct fish and define a consistent rule for multiple captures within an occasion. Repeated encounters can support richer capture-history models, but then a model designed for repeated occasions is necessary rather than inserting duplicate rows into Chapman.
Can an external tag and an acoustic transmitter be treated as equivalent? They answer different observation questions. A visible external tag depends on physical encounter and inspection, while electronic telemetry depends on device operation, signal reception and receiver coverage. Both can be useful, but their detection probabilities and data structures are different. Tagging studies often combine methods only after explicitly modeling their respective observation processes.
Does a low recapture fraction prove the stock is healthy and large? No. It may be compatible with a large eligible population under strong model assumptions, but it can also arise from movement, poor mixing, tag loss, changed catchability or failure to report. Management conclusions need independent checks and uncertainty analysis rather than a single ratio.
Translate the technical vocabulary precisely
‘Marked’ refers to an animal identifiable as part of a prior release cohort; it need not mean a colored external tag. ‘Recaptured’ means encountered again under a specified observation process, not necessarily harvested or killed. ‘Released’ must be distinguished from merely handled or tagged. ‘Reporting rate’ concerns the fraction of recoveries submitted to investigators, while ‘detection probability’ concerns the opportunity to register an encounter under a defined system. These distinctions remain important in translations between languages and in popular summaries.
‘Abundance’ is the number of individuals within the defined population, while ‘biomass’ is their combined mass. ‘Catchability’ concerns the propensity of eligible fish to be caught by particular gear; ‘availability’ concerns whether fish are accessible to that gear in time and space. ‘Closed’ describes model assumptions over a stated interval rather than permanent isolation. Preserving these boundaries in translation helps readers avoid changing the scientific claim while simplifying the explanation.
A decision path for choosing what to say
First, identify the study’s target species, size range, boundary and time. Second, classify each datum as marked-and-released, second-occasion examined, verified recapture or opportunistic return. Third, check mixing, closure, survival, tag loss and encounter comparability. Fourth, choose a model whose likelihood matches those observations. Fifth, calculate the estimate with checked denominators and units. Sixth, quantify sampling uncertainty and discuss structural limitations separately. Finally, link any biomass or management interpretation to the extra evidence it requires.
If the source lacks the information necessary to resolve one of those steps, keep the limit in the explanation. A useful translation may conclude that a report supports a relative abundance comparison or evidence of fish movement but not a reliable absolute stock count. That is a successful reading, not a failure to provide an impressive number. Scientific literacy improves when we learn to distinguish a known observation from an inference that still needs testing.
Concluding interpretation: the mark is evidence, not the stock itself
Tag–recapture converts an observable overlap into a possible population estimate through a mathematical model. Its value lies in the combination of field design, a transparent estimator and honest uncertainty. An external tag, a reported return, a Chapman number and a fisheries stock assessment are four different layers of the evidence chain. Keep those layers separate, and the method becomes both easier to calculate and harder to misuse.
Primary sources and related eduKateSG explanations
For primary-method foundations read the FAO Manual of Fisheries Science: release and recapture, the NOAA Fisheries study of sablefish tag-reporting rates, the NOAA-hosted comparative shark tag-retention study and NOAA-hosted analysis of movement effects in mark–recapture. These are methodological examples, not sources of this article’s invented numbers. For the neighbouring calculations see swept-area trawl surveys, fish otolith ages and age–length keys, maturity ogives, length–weight conversions, and acoustic telemetry detection. For words and meaning see How English Works and How Vocabulary Works.