Why a fish’s landing location is not its biological stock
A fish landed at a harbour is not necessarily a member of the population nearest that harbour. Many harvested species migrate, mingle during feeding, and return to different spawning areas. A mixed-stock fishery can therefore catch individuals belonging to several biological groups in the same fishing operation. Genetic stock identification, usually shortened to GSI, compares genetic markers from harvested fish with a reference collection of fish whose origins are reasonably established. It asks which stock or reporting group most plausibly contributed to a catch, rather than simply identifying the species. To read a GSI report correctly, separate the sampling location, biological origin, genetic reporting unit, and management unit before interpreting any percentages.
The central distinction is between a fish’s individual assignment and a catch’s estimated mixture. An assignment algorithm may assign one fish to group A with a probability or posterior weight, while a mixture analysis estimates the proportions of groups A, B and C across a sample. The group with the highest weight for one fish is not the same quantity as the estimated group share of a fleet’s total landings. Neither outcome is automatically an exact birthplace, and the labels depend on how the reference baseline was constructed. Translating an ambiguous group name as a precise river of origin would add a claim that the genetic analysis may never have supported.
The first question is what a ‘stock’ means in the study. A biological population may reproduce with some degree of isolation, a genetic reporting group may aggregate several similar populations, and a management stock may be defined geographically for assessment or regulation. They can overlap without being identical. Genetic similarity and demographic independence are different concepts. Two spawning groups can exchange enough genes to look similar at selected markers but still differ in productivity or exposure to fishing pressure. A reader should seek the study’s definitions rather than assuming every place name on a graph represents an independently managed spawning population.
A useful interpretation begins by identifying the fishery, species, year, sampled gear or port strata, laboratory marker panel, baseline groups, analysis method, uncertainty intervals and catch-weighting procedure. Without these fields, a diagram of coloured origins is more illustration than decision-ready evidence. It can show that several contributing groups were detected without demonstrating that all groups were sampled or that percentages represent all landings. A clear public explanation says which question was measured, which question was inferred, and which additional data would be required to connect the biological result to harvest management.
How a genetic baseline gives meaning to an unknown specimen
A reference baseline is built from specimens collected from populations of known or well-supported origin, often at spawning sites or under other sampling arrangements that establish group membership. Researchers genotype those fish at selected markers and estimate how marker patterns vary within and among groups. Single-nucleotide polymorphisms, or SNPs, record genetic variation at selected genome positions. The important thing is not that one allele uniquely names a river in every fish. Statistical information arises because combinations of marker frequencies can differ across sampled groups. A larger marker panel may improve discrimination, but only when data quality, marker choice, population representation and statistical assumptions support it.
Imagine two hypothetical reference groups at one genetic marker. The frequency of allele X is 0.80 in group A and 0.30 in group B. Observing X gives some evidence toward A, but it does not prove group A origin because the allele exists in both groups. If prior probabilities are equal, a simplified single-allele comparison yields relative weights 0.80 and 0.30; normalizing them gives 0.80/1.10, about 72.7 percent, for A under this deliberately simplified model. Real multilocus genotype likelihoods account for two alleles at each diploid locus and other details. This teaching calculation illustrates why a marker is evidence, not a geographical barcode.
Missing source populations create a more serious problem than a slightly small sample. If the baseline includes only A and B but fish from unsampled group C enter the catch, the analysis may assign those fish to the most similar represented group. The output can look numerically complete, with proportions adding to one hundred percent, even though the biological alternatives were incomplete. Analysts should describe baseline coverage, possible unsampled sources and how an unknown-origin or outlier scenario was tested. A definitive-looking pie chart does not repair missing reference groups. The denominator remains the fish analysed under the available baseline model.
The timing of reference collection also matters. Allele frequencies may vary through recruitment, sampling error or demographic change, and a baseline assembled from different years may not exactly represent every season of the target fishery. Genetic reporting groups can be consolidated to preserve identification accuracy, so a group called ‘Northern Region’ may contain several rivers whose individual assignments are unreliable. A sound reader looks for the baseline version, list of populations, sample sizes and nested group hierarchy. A version change that splits one region into three should not be narrated as a sudden emergence of three fish populations.
From DNA sample to genotype quality checks
A field worker may preserve a fin clip or another permitted tissue sample, label it with a specimen identifier and record catch details. Laboratory steps include DNA extraction, marker genotyping and quality controls. The genetics result has no practical meaning if the specimen identifier cannot be reconciled to the correct landing record. Sample switches, duplication, inconsistent vessel dates and lost extraction batches can propagate into polished mixture graphs. The critical data relationship is specimen to sample to laboratory plate to genotype to analytic assignment to catch stratum. Readers should inspect this chain before treating the final percentage as a precise management measurement.
Genotyping success is not identical across every specimen. Damaged samples may have missing loci, and some assays may show systematic differences between laboratories or batches. A report can filter specimens for call rate, duplicate concordance and other thresholds. Suppose 250 fish were collected but only 225 passed the stated quality criteria. The genetic result describes 225 analysable samples, unless an explicit adjustment covers the twenty-five failures. The pass fraction is 225/250 = 90 percent. If failures cluster by species condition or sampling location, treating the accepted samples as a random ninety percent of the original collection may misrepresent the fishery.
A marker’s apparent informativeness also depends on sampling. Rare alleles may be absent from a small reference group simply by chance. Strongly related reference specimens can distort the estimated frequency distribution if the design treats all specimens as independent, and contamination can create improbable heterozygosity or conflicting multilocus patterns. Readers need not become laboratory geneticists to understand these warnings. They should look for whether the publication reports reference sizes, low-quality exclusion rules, missing genotype handling, technical replicates and checks against known-origin individuals. These are the equivalents of calibration and instrument blanks in other scientific measurements.
The laboratory output should be separated from the biological conclusion. A variant call reports the genotype inferred at a selected locus under a laboratory and software pipeline. Assignment probability is a downstream estimate conditional on that genotype information, the selected groups, the statistical model and often prior assumptions. A strong genotype call does not imply an equally strong stock assignment if groups have similar allele frequencies. Conversely a probabilistic stock assignment does not mean the fish’s genotype was uncertain. Precision requires naming which stage is uncertain and avoiding the vague expression ‘DNA accuracy’ as though it were one universal percentage.
How mixture proportions are estimated
Consider a collection of sampled fish drawn from a mixed fishery. Each fish has a multilocus genotype that could have arisen under more than one reference group. Mixture estimation asks which combination of group contributions best explains the collection, taking account of genotype probabilities and the analytic model. The components are non-negative proportions constrained to total one. An estimated thirty percent contribution from group A is a statement about the analysed catch mixture under the reference system. It is not a claim that thirty percent of the entire ocean population lives in area A, nor that each fish carried a thirty-percent ancestry label.
A simple teaching example has one hundred sampled fish, with 45 strongly associated with group A, 35 with B and 20 with C. Counting hard assignments produces 45, 35 and 20 percent. But proper mixture methods need not equal these hard counts. Suppose several specimens have substantial probability under both A and B; their information may support adjusted mixture proportions such as 42, 38 and 20 percent. The point is not that the adjusted result is always preferable, but that the two estimands differ. If a report uses individual maximum-probability assignments, say so; if it fits mixture proportions jointly, preserve that method in the translation.
Prior information can enter explicitly. If one method expects a group to be rare and another uses equal prior probabilities, identical genotypes may receive different individual posterior assignments. Prior choices are not necessarily flaws; they are modelling assumptions that need disclosure. A useful sensitivity check repeats analysis under defensible alternative priors and reports whether the management conclusion changes. For a fish with 0.55 weight on A and 0.45 on B, reporting A with no qualification throws away important uncertainty. For a mixture, a ninety-five percent interval indicates uncertainty in estimated group share under the chosen method, not that ninety-five percent of individual fish were assigned with certainty.
Mixture analysis may report posterior medians, means or maximum-likelihood estimates. These are different summaries, especially for small or rare groups whose uncertainty distributions are asymmetric or near zero. The phrase ‘most likely contribution’ is incomplete without naming the statistic and interval. If group C’s estimated share is 2 percent with a credible interval spanning near zero to 8 percent, announcing that C supplied exactly two percent of catch would be misleading. It is more useful to say the estimated contribution was small and imprecise, then explain the sample size and classification limits that produced the width.
Individual assignment, uncertainty and unassigned fish
An individual assignment result often looks like a table of probabilities across candidate stocks. One fish may have estimated weights of 0.92 for A, 0.06 for B and 0.02 for C. Another may have 0.40, 0.35 and 0.25. Both can receive the hard label A, yet their evidentiary strength is clearly different. A translation that reports only the winning label erases uncertainty. A good table includes the assignment method, the candidate groups, the leading score, the score margin and the threshold for leaving a specimen unresolved. It should also indicate whether the scores are calibrated probabilities, likelihood ratios or another classifier output.
An unresolved category is often a sign of restraint rather than scientific failure. If no candidate population is sufficiently supported, the analysis can leave an individual unassigned or report only a broad parent group. Such a decision can protect against overconfident fine-scale labels. Suppose 180 of 200 analysed fish exceed a declared assignment threshold; the resolved fraction is 180/200 = 90 percent. It is invalid to divide every group count by 180 and present those figures as proportions of all two hundred fish without stating that twenty were excluded. One can show both the group distribution among resolved fish and the unresolved share of the full analysis.
The word ‘confidence’ deserves caution in technical writing. An individual assignment confidence score is not automatically a frequentist confidence interval, and a Bayesian posterior probability is conditional on the supplied candidate populations and priors. A model’s calibration must be checked before one interprets a stated 90 percent score as nine correct assignments out of every ten comparable predictions. Known-origin validation specimens can test whether assignments with similar scores actually succeed at a similar rate. The practical question is whether the numerical uncertainty tells the truth about future mistakes, not merely whether the model outputs impressive decimals.
When two populations cannot be resolved reliably, researchers may merge them into a wider reporting group. This sacrifices geographical detail to improve correctness. A report that distinguishes a coastal region from an inland region with strong validation may be more useful than one that prints many unsupported river labels. A reader can ask whether the group hierarchy is nested and whether the same sample was analysed at both regional and fine scales. Changes in the hierarchy affect comparison across studies: a map depicting ‘A1 plus A2’ in one year cannot be directly compared with a later map using only parent group A unless the categories are reconciled.
Validation with known-origin fish and simulated mixtures
Validation is the bridge between an assignment method and a claim about its accuracy. One common approach temporarily withholds known-origin specimens from the training baseline and predicts their group, taking care that the held-out individuals are independent enough to be a meaningful test. Another constructs artificial mixtures with known proportions and checks whether the analysis recovers those proportions. The strongest validation resembles the field application: similar group frequencies, sample quality, missingness, candidate populations and scale of reporting. Validation on conveniently separable groups should not be advertised as proof that difficult neighbouring stocks can also be distinguished.
Suppose a cross-validation confusion matrix contains eighty known A fish and twenty known B fish. Of A fish, seventy-two are assigned A and eight B; of B fish, fifteen are assigned B and five A. Overall accuracy is (72+15)/100 = 87 percent, while A sensitivity is 72/80 = 90 percent and B sensitivity is 15/20 = 75 percent. The overall figure hides a weaker result for B. Among the seventy-seven fish predicted A, precision is 72/77, roughly 93.5 percent. These values refer to this artificial validation set; changing the prevalence of stocks in actual landings can change precision even if sensitivities remain stable.
Mixture recovery tests ask a related but different question. Create a hypothetical mixture containing 60 percent group A, 30 percent B and 10 percent C. If repeated simulated samples yield estimates that average 58, 32 and 10 percent, the study can examine bias, variability and interval coverage. The absolute percentage-point error for A is two points; its relative error is 2/60, about 3.3 percent. These are not interchangeable. A mixture estimator can perform adequately on broad groups while individual assignments remain ambiguous, because aggregate information from many specimens may be easier to estimate than the origin of one individual.
Simulation should not be tuned until it always produces reassuring results. Analysts can deliberately vary rare-group frequency, missing populations, sample size, marker missingness, prior choice and season. A result that collapses under plausible alternative baselines deserves an uncertainty warning. When a paper quotes an exceptional accuracy figure from a different species or baseline, the appropriate translation is that the earlier system achieved the figure under its own conditions. It is not a universal accuracy guarantee for all mixed-stock fisheries. The reader needs specific, independently relevant validation for the current population structure and sampling frame.
Sampling design connects genetic shares to actual harvest
A mixture proportion among sampled fish may not describe the whole fishery if sampling favours large boats, accessible landing sites, particular gears or certain weeks. Imagine an illustrative fishery landing 80 tonnes in stratum A and 20 tonnes in stratum B. Researchers sample one hundred fish from each stratum. A simple unweighted average treats the strata equally even though A contributes four times the landing mass. If the within-stratum estimated contribution from a stock is 20 percent in A and 70 percent in B, an equal-stratum average gives 45 percent, but a catch-weighted estimate is (80×0.20 + 20×0.70)/100 = 30 percent.
Even catch weighting must match the estimand. If the question concerns share of individual fish, weighting by number of fish may be appropriate; if it concerns harvested biomass, weighting by landed mass can be more relevant, sometimes with stock-specific size differences. The phrase ‘thirty percent of the catch’ is ambiguous without specifying whether catch means fish counted, landed weight, sampled fish, retained product or total removals including discards. A genetic mixture derived from retained commercial landings may exclude discarded juveniles and unobserved fishing. Its valid conclusion should stay inside the sampled and adequately weighted population.
Sampling precision is affected by cluster structure. Fifty fish from a single large catch can share a vessel, location and time, so they may convey less independent information about the seasonal fishery than fifty fish sampled across diverse trips. An effective sample size can be smaller than the number of genotypes. Stratified designs should report trips, landings and sampled specimens, not only a laboratory specimen count. If seasonal composition changes, a pooled annual estimate may conceal a biologically vulnerable stock that dominates one short period. Time strata and uncertainty should be visible when management decisions depend on within-season exploitation.
Non-response and laboratory failure require attention at the same stage as vessel coverage. Suppose one port supplies 60 percent of total landings but only 15 percent of genetic samples because of logistical constraints. Applying the raw specimen proportion to all landings can be biased. A survey weighting plan may correct the imbalance if the port catch totals and selection probabilities are reliable, but no weighting scheme can guarantee that completely unsampled gear or time periods behave like sampled ones. Honest reports show coverage maps or tables and use sensitivity analysis where parts of the target fishery have no direct genetic evidence.
Worked example: a three-stock catch and the correct denominator
Consider an invented survey that samples one hundred eligible fish from each of two landing districts. In district North, estimated shares are A 50 percent, B 30 percent and C 20 percent. In district South they are A 20 percent, B 50 percent and C 30 percent. The sample-pooled shares are A 35 percent, B 40 percent and C 25 percent when districts contribute equal sample counts. If the real fishery landed 300 tonnes in North and 100 tonnes in South, the landed-mass-weighted mixture is A (300×0.50+100×0.20)/400 = 42.5 percent; B 35 percent; C 22.5 percent.
The calculation can be checked by summing the implied tonnes. Stock A contributes 150+20 = 170 tonnes, stock B contributes 90+50 = 140 tonnes, and stock C contributes 60+30 = 90 tonnes. These sum to the reported 400 tonnes. Dividing each by 400 yields 42.5, 35 and 22.5 percent. This arithmetic demonstrates consistency of the illustrative bookkeeping, not a claim that all stock assignments are exact or fish weights within groups are identical. An actual analysis would propagate genetic mixture uncertainty, catch measurement error and sampling design variation rather than merely multiplying point estimates.
Suppose the total catch itself has a measurement uncertainty and district catches are revised. If South becomes 150 tonnes while North remains 300, the stock A point estimate becomes (150+30)/450 = 40 percent, B becomes (90+75)/450 ≈ 36.7 percent, and C becomes (60+45)/450 ≈ 23.3 percent. The stock composition apparently changes even though within-district genetic mixtures remain fixed. This is why auditors should retain the catch table used for expansion, the date of its revision and the analysis code. A percentage from one report cannot be compared safely with a percentage from another until their weighting bases match.
A transfer exercise asks what happens if only eighty of one hundred sampled South fish pass genotype quality control, but every North fish passes. Should one automatically weight North by one hundred and South by eighty? No. Specimen acceptance counts do not replace the underlying fishery landing weights. One should first assess whether missing genotypes are random within South and whether group composition could differ among failures. If the accepted South specimens remain representative, analysis can estimate South’s mixture with increased uncertainty and still use South’s documented landing weight. If missingness is selective, adjustment or uncertainty bounds may be necessary.
Rare stock detection versus precise estimation
A rare stock may be biologically important even if its contribution to a large fishery is small. Detection asks whether at least one individual is likely represented; precise estimation asks how much the stock contributes. These have different sample requirements. If a perfectly classified stock truly contributes one percent of independently sampled fish, the probability of seeing no individuals in n simple random draws is 0.99 raised to n. With one hundred draws this is approximately 36.6 percent, so even a genuine one-percent contributor can easily be missed by a modest sample. Real genetic misclassification and clustering can make detection harder.
At three hundred independent samples and the same idealized one-percent prevalence, the chance of at least one draw is 1−0.99^300, approximately 95.1 percent. This calculation assumes random, independent fish and perfect identification. It does not guarantee a precise estimate of the one-percent share, because the expected observed count is only three fish and percentage estimates based on such few observations can vary considerably. Nor does it establish that every one-percent stock will be seen in reality. The exercise clarifies why a report must distinguish ‘not detected’ from ‘absent’ and ‘detected’ from ‘accurately quantified’.
If an analyst chooses a threshold requiring multiple supporting genotypes, sensitivity to a rare group can decline while false-positive control improves. Different thresholds make different trade-offs. A report should explain what counts as a detection, how ambiguous assignments are handled and whether the baseline can distinguish the rare group from its neighbours. A single apparent rare assignment may reflect biological presence, laboratory error, contamination or an unsampled related stock. A careful statement expresses the competing explanations rather than either dismissing the observation or treating it as conclusive.
The cost of an error may be asymmetric. Missing a vulnerable spawning stock can affect conservation decisions, whereas falsely attributing harvest to that stock can lead to unnecessary restrictions. Scientists may therefore report upper confidence or credible bounds, sensitivity scenarios and conservative management rules instead of one crisp percentage. Readers should not convert a precautionary upper limit into the best estimate, or a lower bound into a proof of zero. The chosen decision threshold belongs to the management question, not simply to the genetics software. A responsible translation preserves both statistical evidence and the consequence being considered.
Genetic groups, demographic stocks and management boundaries
Genetic divergence measures differences in inherited variation, whereas management often cares about birth, death, recruitment, productivity and fishery exposure. The biological and management maps need not coincide. Populations can exchange enough migrants each generation to show weak genetic differentiation but still have locally meaningful recruitment dynamics. Conversely a genetic grouping can persist while movement patterns change with the environment. Therefore ‘genetically distinguishable’ does not automatically mean ‘demographically independent’, and ‘genetically similar’ does not establish that managing all fish as one unit is risk-free. State which evidence supports each concept.
Geographic management boundaries are legal and administrative decisions. A baseline reporting group called ‘West Shelf’ might span several jurisdictions, while a catch limit might cover only one jurisdiction’s water. A mixture estimate grouped by West Shelf cannot automatically be mapped onto a legal quota category unless a documented crosswalk exists. The same concern arises when a fish is caught outside its spawning area. A map of capture positions answers where the encounter happened. A genetic mixture answers which baseline groups most likely contributed. Translators should label the two maps separately, even if they appear beside each other in one report.
Stock assessment integration needs care. An assessment that previously assumed a single population may be revised to represent migration or stock-specific catch components after genetic evidence becomes available. Yet a genetic survey is not itself an estimate of absolute population abundance. If 40 percent of the sampled catch originates from group A, it does not follow that group A accounts for 40 percent of the ocean’s biomass. Catchability and fishing distribution can favour one group, and the sampling may exclude portions of its habitat. Absolute abundance requires additional information about population size, fishery selectivity, survey catchability or other estimation processes.
Even agreement between genetic and conventional tags must be interpreted at the right level. Tags can provide movement or cohort information for fish that were tagged and subsequently observed, while GSI can include untagged specimens whose markers are informative. Their biases differ. Concordant results strengthen confidence only when the compared units, years, fishery strata and assumptions align. An independent validation table is more valuable than a general claim that two sophisticated methods agreed. Explain the shared estimand and the independent sources of error before using one result to corroborate the other.
When version changes make the same catch look different
Genetic methods evolve as reference libraries grow, marker panels improve and analysts refine how groups are combined. Such improvements are welcome, but they complicate time-series interpretation. A historical catch batch can yield one set of mixture proportions under an earlier baseline and another under a newly expanded baseline. The underlying fish did not change. If the public is shown the revised result without a version note, it may mistake an analytical update for ecological change. Keep baseline identifiers, marker panel names, reference population lists, group hierarchy and software versions attached to the reported estimate.
Suppose an old baseline classifies a hypothetical catch as 70 percent Northern and 30 percent Southern. A new baseline splits Northern into River A 40 percent and River B 30 percent while Southern remains 30 percent. This does not demonstrate a new 40-percent River A catch that previously did not exist. It shows a change in reporting resolution. To construct a consistent trend, analysts can aggregate new River A and River B estimates into the old Northern category or reprocess old samples under the new baseline when available. Every bridge between naming systems should be explicit rather than inferred from similar-looking colours.
Another problem is changed sample exclusions. If a new genotyping pipeline retains eighty more fish from poorly preserved landings, the mixture may shift because the eligible sample changed. Separate that sampling effect from the re-estimation effect by analysing a shared specimen subset if possible. Comparisons should also preserve whether individual hard assignments or full mixture estimates were used. A historical table of top-label counts and a contemporary model-based mixture cannot be compared as though both were identical measures of stock share. Methodological continuity is more important than an uninterrupted line in a chart.
An honest revision notice can be quite concise: the estimates for prior years were recalculated with a specified baseline version and updated group definitions, and therefore differ from earlier reports even though the underlying samples are unchanged. Readers should retain access to both the original and revised tables, along with a reconciliation matrix of categories. Hiding a methodological break may create a persuasive but fictional trend. Genetic evidence helps fisheries management most when it can be audited across time, especially when catch limits or conservation priorities depend on a small estimated change.
Translation traps: technical words with ordinary meanings
The everyday word ‘assignment’ can sound like an official decision. In GSI it often means a model-based classification of a fish to a reference group. ‘Stock’ may mean a fishery management unit rather than an investment or a warehouse inventory. ‘Baseline’ is a genetic reference dataset, not a shoreline or a zero measure. ‘Mixture’ means the estimated contributions of candidate groups, not physical mixing of DNA between specimens in a tube. Define these meanings in the opening paragraphs so a general reader does not import false assumptions from ordinary vocabulary.
Similarly, ‘accuracy’ needs an explicit unit. A paper may report 95 percent correct broad-region assignments under cross-validation but lower accuracy for adjacent rivers. Saying the technique is 95 percent accurate without the tested scale is misleading. ‘Precision’ can refer to repeated estimate variability, confidence interval width or laboratory measurement repeatability depending on context. ‘Confidence’ is commonly used casually when a model actually reports a posterior probability. Preserve the technical statistic if available. Numbers are less informative when their measurement target has been silently erased.
‘Composition’ asks about proportions, not absolute fish count. ‘Contribution’ may refer to sampled numbers, estimated landed numbers, landed mass or total removals. ‘Unassigned’ does not mean the specimen has no origin; it means the available evidence or decision rule did not permit a defensible chosen label. ‘No detection’ is not the same as a confirmed absence. ‘Mixed-stock’ is not a claim that individuals are hybrids. These phrases are frequent sources of public misunderstanding because each resembles a familiar word with a different operational definition.
Good translation keeps the hierarchy visible: genetic locus, genotype, candidate reference population, reporting group, catch sample, weighted fishery stratum and management unit. It also avoids assigning legal blame to a place based on a biological mixture estimate. If an identified stock is associated with a named river, fishing that stock offshore may still be legal under applicable rules. The role of a technical explainer is to preserve what the genetics shows, to identify what it cannot show and to give the reader enough context to ask the next useful question.
Independent problems with checked answers
Problem one: a baseline test contains forty known-origin fish from group A and sixty from B. The classifier correctly labels thirty-six A and fifty-one B, assigning the others incorrectly. Find the overall accuracy and two group sensitivities. Answer: overall accuracy is (36+51)/100 = 87 percent. A sensitivity is 36/40 = 90 percent, and B sensitivity is 51/60 = 85 percent. These are validation-set results, not catch mixture proportions. If the real catch contains far more A than B, the precision of a predicted B label could differ from that in this test. State which population generated the validation counts.
Problem two: an invented coastal fishery lands 240 tonnes in district X and 160 tonnes in Y. In X a stock’s estimated share is 25 percent; in Y it is 50 percent. What is the catch-weighted share? Answer: contributing mass is 240×0.25 + 160×0.50 = 60+80 = 140 tonnes. Dividing by total 400 tonnes yields 35 percent. The equal-district average would be 37.5 percent and is not the correct landed-mass-weighted result for these data. Both figures exclude uncertainty in mixture estimation and assume landing records and within-district estimates correspond to the same target species and time window.
Problem three: in a perfectly identifiable independent random sample, a source group contributes two percent of fish. What is the chance of sampling none in fifty fish? Answer: 0.98^50 ≈ 36.4 percent, so failing to observe the group in fifty draws is quite plausible even when it is present. This model is an idealized binomial calculation; real sampling can cluster by vessel or season and assignment may be imperfect. The practical interpretation is that negative evidence is weak for a rare group at small sample size. To make a strong absence claim, one must define a detection threshold, sampling design and acceptable error risk.
Problem four: a report finds group shares of 10, 35 and 55 percent, but describes its samples as collected only from retained fish in two main ports. Can it conclude that these are the group proportions of all fish removed by the fleet? No. Discards, smaller ports, other gear types, unreported landings and genotype failures may change the target population. The correct claim is restricted to the analysed and appropriately weighted retained catch. A defensible expansion to all removals would need supporting coverage data and uncertainty analysis. A figure’s percentages summing to one hundred does not prove that its sampling frame was complete.
Frequently asked questions and practical limits
Can DNA reveal the exact river where a fish was born? Sometimes strong markers and comprehensive baselines permit high-resolution inference, but not as a universal rule. A fish may be assignable only to a broad group because neighbouring rivers have similar genetic signatures. The model may omit plausible source populations, and an apparently strong best group still depends on its priors and validation. The plain-language answer should identify the tested reporting scale. Naming a precise birthplace without the required discriminatory evidence creates geographic certainty that the analysis does not possess.
Does a high mixture proportion mean that a particular stock is abundant in nature? Not necessarily. It means the estimated share of the defined catch sample or fishery removals is high under the method used. Different groups may have very different migration timing, fishing exposure and catchability. A relatively small population could be disproportionately harvested, while an abundant population might rarely enter the sampled gear. Biological abundance and fishery composition are different estimands. To examine stock health, combine appropriately weighted catch origins with surveys, recruitment, productivity and relevant management assessment information.
What should happen when an unrepresented stock is suspected? The study should not silently force its fish into the nearest reference category and report the outcome as certain. Analysts can seek additional reference specimens, test an unknown-group sensitivity scenario, use broader reporting groups, or qualify limitations. Another question is whether reference specimens were collected during spawning or through a method that truly supports origin. A geographically close sampling site is not automatically a valid origin label for a highly mobile species. Quality begins with a defensible biological sampling design, not with software output.
Is genetic assignment the same as parentage-based tagging or close-kin mark–recapture? No. Genetic stock identification primarily uses group-level frequency patterns to estimate origin or catch composition. Parentage methods match sufficiently informative relatives or parent–offspring relationships to infer particular sampling or abundance information under their own design. Close-kin approaches can help estimate adult abundance when sampling and demographic assumptions support them. Each uses DNA, but their target questions and error structures differ. For a related explanation, see the separate eduKateSG discussion of fish close-kin mark–recapture rather than treating the methods as synonyms.
Original methodological sources and next reading
For first-party research see the USGS study of genetic stock identification and parentage methods in Chinook salmon, the NOAA reference on sample sizes for stock composition, and the NOAA Fisheries methodological work on Bayesian stock mixtures. The numbers in this article’s exercises are deliberately hypothetical and are not measurements of those studies.
For adjacent but distinct purposes see eduKateSG’s verified guides on close-kin mark–recapture, fishery-independent surveys, and AIS–VMS tracks. None can be treated as the same measurement as genetic catch composition.