Start with the fishery question, not the catch graph
Catch per unit effort, usually abbreviated CPUE, expresses recorded catch relative to a defined measure of fishing effort. The tempting interpretation is that higher CPUE means more fish in the water. Sometimes it does indicate increased local availability, but CPUE also changes when vessels, gear, fishing behaviour, areas, environmental conditions and reporting systems change. Reading CPUE well requires reconstructing what was caught, the effort denominator, the sampled fleet, and the process by which the catch was observed. Without these details, a smooth annual index can conceal a shifting measurement system.
A CPUE series is useful because direct census of a mobile marine population is rarely feasible. Fishing activity creates observations repeatedly across time and space. If catchability remains stable under well-defined circumstances, CPUE may be related to abundance through a conceptual relation U=qB, where U is catch per unit effort, B is vulnerable biomass or abundance, and q is catchability under that particular operating definition. Both q and which population component is vulnerable matter. The formula is an assumption-bearing bridge, not a declaration that one kilogram caught proves one kilogram in the sea.
Suppose a longline vessel reports 600 kg of target fish from 1,200 hooks. Nominal CPUE is 0.5 kg per hook, or 500 kg per 1,000 hooks. That is a correct arithmetic summary of the record, provided the 600 kg belongs to the same target species, the 1,200 hooks represent the agreed effort and no incompatible trips have been mixed. It does not measure the ocean-wide stock biomass or even abundance per square kilometre. Its spatial and operational scope must travel with the number.
When a report describes CPUE as a ‘relative abundance index,’ relative is doing important work. A standardised index may be scaled to one in a chosen reference period, or it may be proportional to biomass only up to an unknown catchability coefficient. An annual value of 1.20 could mean 20% above the reference index, not 1.20 tonnes of fish, and not necessarily a 20% increase in absolute biomass. The accompanying uncertainty and model definitions determine what can defensibly be inferred.
Define the denominator before dividing
Fishing effort has many forms: hook numbers, hook-hours, trawl duration, swept area, trap-nights, net length multiplied by soak time, fishing days, vessel-days, search hours or another well-defined exposure. Different measures capture different opportunities to encounter fish. A trawl operated for one hour is not necessarily equivalent to another trawl working over twice the seabed area or at a different speed. A catch-per-hour index may need additional covariates or a different denominator to account for changing gear power.
Distinguish effort deployed from effort successfully observed. For example, 1,000 hooks set but 200 lost or not retrieved cannot necessarily be treated as 1,000 effective observed hooks. Whether to use set hooks, hauled hooks or effective hooks depends on the operational protocol and question. The choice must be consistent and justified. When an operation changes how it records failures, a CPUE trend can emerge from the denominator definition alone.
Numerators require equal care. Retained commercial catch, total catch including discards, landed catch, estimated live releases and scientific survey capture are different quantities. A rule restricting landings may make retained-catch CPUE decline while encounters stay similar. Species mixing, size grading, conversion from processed weight to live weight and observer coverage introduce additional changes. A report must identify whether weights are whole, gutted, dressed or converted, and the factor used if conversion is necessary.
Compare two hypothetical records. Trip A catches 240 kg in 12 trawl hours, so nominal CPUE is 20 kg/h. Trip B catches 240 kg in 8 trawl hours, so CPUE is 30 kg/h. The difference is a ratio of catches to stated hours. Before concluding availability increased by 50%, ask whether trawl width, speed, habitat, season, location, vessel and target strategy were comparable. If the trawl hour has become more effective, a rising CPUE need not reflect more fish.
Why standardisation is an inference rather than a cosmetic correction
Standardisation attempts to estimate an index comparable across years after accounting for changes in measured factors affecting fishing success. A model might include year, area, season, vessel, gear configuration, target strategy and environmental covariates. The year effect is often of central interest, but it does not become pure abundance merely because these terms appear in a formula. Unmeasured behaviour, changing catchability and spatial coverage can still bias the result.
Think of fishing as an observation process. A captain may pursue known aggregations, avoid depleted areas, change fishing depth, upgrade electronics or switch target species. Fishing may become more efficient even while abundance falls, a phenomenon sometimes called technological creep or effort creep. Conversely, management limits, market incentives or weather may reduce encounters independently of stock abundance. Standardisation is therefore closer to measurement calibration than to dividing by an arbitrary correction factor.
One common route is a generalised linear model, often using log effort as an offset for catch. In a simple count or catch-rate model, the expected catch changes with exposure, and the offset makes comparisons meaningful under assumptions about proportionality. A model of log CPUE can also be used but needs care with zeros, error distributions, heteroscedasticity and back-transformation. The regression form is a tool, not a universally correct recipe.
The model should produce a standardised prediction under a defined reference distribution of relevant covariates. It should not simply use whichever vessel or area happened to be popular in each year, because that could preserve the very compositional change standardisation was meant to address. Predictions may be averaged over a consistent design grid or another transparent target population. The weighting choice defines the estimand, so report it rather than calling every result the ‘corrected CPUE.’
Worked arithmetic: pooled versus averaged trip rates
Imagine three trips with catches and hook counts as follows: 40 kg over 100 hooks, 60 kg over 600 hooks and 20 kg over 100 hooks. Their rates are 0.40, 0.10 and 0.20 kg/hook. The simple mean of trip-level rates is (0.40+0.10+0.20)/3 = 0.2333 kg/hook. Pooled catch divided by pooled effort is (40+60+20)/(100+600+100) = 120/800 = 0.15 kg/hook. Both calculations are mathematically valid but estimate different summaries.
The pooled estimator gives greater weight to high-effort trips. The unweighted trip mean gives equal weight to each recorded trip regardless of exposure. If the analysis question is the catch rate of a randomly selected trip, an unweighted mean may have a natural interpretation. If the question is catch per deployed hook across the observed operation, the pooled ratio is appropriate. Neither should be labelled universal stock abundance without a sampling design argument.
Consider a second comparison involving a sudden change in fleet composition. In year one, nine small vessels each catch 10 kg/h and one large vessel catches 40 kg/h. The unweighted mean is (9×10+40)/10 = 13 kg/h. In year two, two small vessels and eight large vessels produce the same vessel-specific rates. The new mean becomes (2×10+8×40)/10 = 34 kg/h. This apparent 162% increase arises solely from the fleet mixture in this constructed example; each type’s rate was unchanged.
Standardising both years to a common fleet composition—say five small and five large vessels—would give (5×10+5×40)/10 = 25 kg/h for each year under a simple equal-vessel weighting. That thought experiment demonstrates what comparability means. Real models also need to estimate differing vessel and year effects, deal with sparse combinations, and represent uncertainty. If one fleet disappears from all later years, extrapolating its performance is more uncertain than interpolation across observed comparable effort.
The zero-catch problem and delta models
Fisheries catch data often include many zero observations, particularly for rare species, broad survey coverage or operations that targeted something else. Replacing zeros with a convenient small constant before taking logarithms can make results depend strongly on that arbitrary choice. A log transform of strictly positive CPUE may be reasonable for some data, but it cannot describe encounter probability on its own. Zero prevalence should be examined as an ecological and operational signal, not treated only as an inconvenient calculation.
A delta or hurdle approach separates whether the species was encountered from how much was caught when present. The first component models Pr(positive catch), often with a binomial logistic model. The second estimates positive catch rate conditional on an encounter, using a suitable positive distribution or link. The expected unconditional rate is the product of encounter probability and positive-catch expectation, under the model’s conventions. The two components can respond differently to changing environment, gear and stock distribution.
Suppose a standardised year has a predicted encounter probability of 0.40 and predicted positive catch of 12 kg per standard effort unit. Then the estimated unconditional rate is 0.40×12 = 4.8 kg per unit. In another year, probability is 0.30 but positive-catch expectation is 16 kg; the product is again 4.8. If a report looks only at positive catches, it would claim an increase from 12 to 16, obscuring the fall in encounter probability. If it looks only at zeros, it misses the shift in conditional catch.
Do not confuse multiplying the two fitted component point estimates with fully propagating uncertainty. The covariance between components and variation in covariates may matter. A simulation or bootstrap can account for model estimation and sampling structure. In a publication, show the encounter trend, the positive-catch trend and the combined index where helpful. A reader can then see whether a stable combined index hides meaningful changes in spatial occupancy or encounter frequency.
Choosing distributions and diagnosing fitted models
Count data, continuous biomass and highly skewed catch weight need not follow the same probability distribution. A negative-binomial model may suit overdispersed counts; a gamma or lognormal component can describe some positive continuous catches; a Tweedie model can sometimes accommodate a mixture of exact zeros and positive continuous outcomes. These choices should be tested against observed data properties and design. Choosing the most fashionable model does not establish validity.
Residual plots should be interpreted alongside covariate coverage. A model may show small residuals in heavily fished areas yet make poorly supported predictions for unvisited habitat. Spatial residual structure suggests important geographical variation remains. Temporal autocorrelation, vessel effects, seasonality and unbalanced design can make nominal standard errors overly optimistic. Diagnose the sampling process as seriously as the numerical fit.
Environmental covariates such as temperature, oxygen, currents and habitat can improve comparability, but they can also remove a real abundance-related signal if modelled without clear causal intent. A changing temperature field might alter catchability, fish movement, survival and productivity simultaneously. If standardisation treats all temperature effects as nuisance, it may produce an index answering a narrower question than users expect. Compare reasonable model specifications and disclose how interpretation changes.
Model selection should consider predictive diagnostics, scientific plausibility and sensitivity to alternatives. A very complex model can fit historical catches yet produce unstable extrapolations when a new fleet enters. A simplistic model may miss strong spatial gradients. The goal is a defensible measure of relative availability under a declared prediction frame, not a single perfect numerical score. When two plausible models diverge materially, that divergence is a substantive uncertainty to communicate.
Avoid a misleading precision hierarchy. A CPUE index with many decimal places is not necessarily more reliable than a rounded index with credible intervals. Statistical uncertainty from model fitting is only one component; measurement error, unreported effort, shifting fleet behaviour and unsupported extrapolation may dominate. A report should say which uncertainty sources were quantified and which remain outside the formal interval.
Spatial dynamics and hyperstability
A fleet can maintain high catch rates by concentrating effort in the few places where fish remain aggregated, even as total stock abundance declines. This phenomenon is called hyperstability when CPUE declines more slowly than abundance because catchability changes with density or fishing concentration. The reverse, hyperdepletion, describes CPUE falling more rapidly than abundance. Neither word is established merely by noting a nonlinear curve; the mechanism and independent abundance information matter.
Imagine a species schooling into dense patches. In year one, patches occupy a wide area and fishers catch 50 kg/h where they choose to operate. In year two, the occupied area contracts but successful fishers still catch around 50 kg/h within the remaining patches. Fleet CPUE looks stable while spatial occupancy has changed drastically. To interpret the index, examine effort maps, occupancy, search time, failed trips and coverage outside the core fishing grounds.
Hyperdepletion can arise when fishers abandon a stock, management restricts accessible areas, or a fish moves to habitats where the fleet is less effective. Declining CPUE may therefore overstate the decline in total biomass under some circumstances. Correcting for area does not fully solve the problem when fish movement changes availability within areas. Fishery-independent surveys, tagging and environmental information can provide independent context.
Spatial standardisation requires a target area definition. Averaging predicted rates across a fixed grid can answer ‘what is expected availability over this defined domain?’ Weighting cells by historically observed effort answers a different question. Expanding the grid into unsampled regions after a habitat shift is extrapolation. Maps should show observed fishing footprints and prediction coverage so readers can distinguish evidence from model-supplied surface.
How to scale, index and explain the model output
An annual standardised CPUE series is often normalised by dividing each year’s prediction by a reference statistic, such as the mean over a specified range of years. If the reference mean is 8 kg/h and the prediction for a later year is 10 kg/h, the relative index is 10/8 = 1.25, meaning 25% above that defined reference in index terms. A different baseline would change all numbers but not the underlying prediction sequence. The baseline must be named and stable.
A reported index of 0.70 should not be translated as ‘70% of the original stock remains’ unless a defensible abundance calibration and reference definition support that statement. More safely, it means the standardised prediction is 70% of a chosen relative reference. Stock assessment models can combine CPUE with other observations and assumptions to infer biomass, but that inference is not contained in the CPUE number alone.
When converting from log-scale model predictions, simply exponentiating a fitted log mean may produce a conditional median or geometric mean rather than an arithmetic mean. A lognormal model needs an appropriate retransformation correction if arithmetic expected catch is the target; the correction depends on variance assumptions. In delta models, expected positive catch and encounter probability must be combined under compatible scales. Back-transformation is not a formatting detail.
The most helpful plot pairs annual index estimates with uncertainty intervals, identifies the baseline, and adds a description of the standardisation model and fleet coverage. A second panel showing nominal CPUE can reveal how much adjustment occurred. Neither line should be described as a direct census. A table of observed trip counts and effort by year helps identify where estimates rest on abundant data and where they depend on sparse model predictions.
For a reader whose first language differs from the report language, a clear translation preserves denominators, time and uncertainty: ‘The model-adjusted catch rate in 2025 was estimated at 1.25 times the 2015–2019 reference average, with the stated confidence interval; this is a relative index of vulnerable fish availability, not an absolute biomass estimate.’ This is more faithful than ‘fish numbers increased by 25%.’
Data integrity and sampling choices
Every CPUE analysis should begin with a data dictionary. Confirm how trips, sets, stations and vessels are identified, how effort was measured, how missing effort differs from zero effort, and whether duplicate catch records exist. Do not divide by zero or silently discard trips with missing denominators. Record inclusion rules, and report how many observations were lost at each cleaning step. A convenient analysis dataset is not automatically a representative fishery dataset.
Fishing strategy should be examined rather than assumed constant. A vessel that switches from targeting the studied species to targeting another species may show many more zero catches. That change could be recorded as a target variable or used to define a consistent directed-fishery subset, but excluding all zeros that happen to be inconvenient would inflate the index. The filtering decision should follow a scientific definition that a second analyst can reproduce.
Changes in regulation may affect the relationship between catch and abundance. Closed seasons, spatial exclusions, size limits, discard rules and reporting incentives can all alter the measured numerator or denominator. A time series that crosses a new reporting regime requires explicit evaluation. If catch data capture landed fish but not discarded fish, a landing regulation may shift CPUE even with unchanged encounter rates.
Observer programmes, electronic monitoring and logbooks provide different completeness and verification. A model built from voluntary logs can inherit selective reporting if high-success trips are more likely to be documented. Comparing independent data streams and conducting plausibility checks can expose systematic differences. Statistical standardisation cannot manufacture unbiased source observations when the original records are fundamentally unrepresentative.
A good reproducibility package contains the date range, taxon code, fleet and gear definitions, effort unit, catch conversion rules, inclusion and exclusion criteria, model formula, distribution, prediction grid, uncertainty approach and versioned data source. That package enables another analyst to determine whether two studies asked the same question before comparing their trend lines.
Independent transfer problems with answers
Problem one. An observer records 75 kg in 1,500 hooks. Express CPUE in kg per hook and kg per 1,000 hooks. Answer: 75/1,500 = 0.05 kg per hook, or 50 kg per 1,000 hooks. If an article reports 50 kg per hook, the mistake is a thousandfold denominator conversion. State the unit explicitly on every chart.
Problem two. Two trips catch 10 kg over 10 hours and 90 kg over 90 hours. What is their pooled CPUE and average trip CPUE? Answer: both equal 1 kg/h in this special case. The equality is not a general identity; it occurs here because both trip rates are the same. Vary one catch and the pooled rate becomes an effort-weighted average rather than a simple trip mean.
Problem three. A report has 40 positive trips and 60 zero trips out of 100. Mean positive catch is 8 kg per trip. Estimate the unconditional mean under a simple two-part decomposition. Answer: encounter probability is 0.4; unconditional mean is 0.4×8 = 3.2 kg per trip. This descriptive calculation does not adjust for covariates and should not be called a stock-wide abundance estimate.
Problem four. A standardised annual index is 1.4 relative to a five-year baseline of 1.0. Write the strongest justified plain-language claim. Answer: the model-adjusted relative CPUE is 40% above the defined baseline. It is not established that total fish abundance or harvest quota is 40% higher. State the model’s assumptions and uncertainty interval.
Problem five. A vessel doubles sonar capability but retains the same catch rate per hook. What does that prove about abundance? Answer: very little by itself. Improved searching might raise catchability, whereas stock decline might offset that gain, or the enhancement might not affect the species. The observation requires behavioural and independent evidence before attributing a cause.
Problem six. One period reports ‘hours fished’ and the next reports ‘hours at sea.’ Why not concatenate them? Answer: travel, searching, handling and resting may enter one denominator but not the other. The exposure definitions differ and can create artificial trends. Reconstruct comparable effort or restrict the analysis to compatible records.
Problem seven. CPUE falls sharply when vessels stop targeting the study species. Does this prove biomass collapsed? Answer: no. Target switching may cause reduced encounters irrespective of stock size. Define fishery strategy, inspect effort allocation and include or standardise the targeting process when the data permit, while avoiding selection on catch outcomes alone.
Problem eight. A gamma model is fitted to catches including exact zeros without accounting for the zero mass. What is wrong? Answer: a standard gamma response is positive continuous and does not assign mass to zero. Use an appropriate two-part or other compatible response model, or justify a method that explicitly represents zeros.
Problem nine. A model reports an estimate of 3.0 kg/h and standard error 0.2, then says 95% of individual trips are between 2.6 and 3.4 kg/h. Diagnose. Answer: the standard error usually quantifies estimator uncertainty, not trip-level dispersion. An individual prediction interval may be much wider and needs a suitable distributional calculation.
Problem ten. A report applies the same environmental covariate to all historical years but lacks measurements in an offshore region for the first decade. Is the prediction supported? Answer: partly through model assumptions, not observations at those missing combinations. Extrapolation should be mapped and uncertainty examined. A precise model coefficient does not create historical data.
Problem eleven. A study removes all zero trips and reports ‘CPUE increased.’ What question does it actually address? Answer: it may address positive catch conditional on an encounter, if that conditioning and the sample are explicit. It no longer measures overall expected catch per unit effort across all defined operations. Restore an encounter component or revise the language.
Problem twelve. An index rises from 0.8 to 1.0 while confidence intervals overlap. Can it be declared a statistically established recovery? Answer: not solely from interval overlap, which is not the definitive test for a difference, nor from two point estimates. Examine the covariance, specified contrast, sampling frame, independent data and model sensitivity before making a recovery claim.
Frequently asked questions
Is CPUE the number of fish in the sea? No. It is a catch-rate measure under a specific effort and sampling definition, sometimes used as a relative abundance index with additional assumptions.
Why does standardised CPUE differ from nominal CPUE? Standardisation estimates a comparable prediction after accounting for selected changing factors. It may change a trend by adjusting differences in fleet, area, season, gear or other covariates, but unmeasured effects can remain.
Can zero catches be discarded? Not without changing the question. Zeros often carry information about encounter probability, stock distribution and targeting. Delta models are one way to analyse them.
Is kilogram per hour comparable with fish per 1,000 hooks? Not directly. They use different numerators and denominators. Any conversion needs appropriate species weight and fishing-operation information, not a generic multiplier.
Does a confidence interval capture all uncertainty? Usually not. Many reported intervals cover sampling and model estimation conditional on assumptions, while reporting errors, shifting catchability and model misspecification may be less well represented.
Why not use the highest-catching vessel as the reference? That would define the index around an unusual or possibly changing operation. A reference or prediction grid should represent the intended comparison, with transparent weighting and support.
Can CPUE increase as abundance falls? Yes, under mechanisms such as concentrated fishing on remaining aggregations or increasing efficiency. This possibility needs investigation, not an assumption that it always occurs.
What should a good translation of an index include? Name the catch numerator, effort denominator, fleet, area, reference period, standardisation method, uncertainty and limits of abundance inference. Explain which comparisons are supported and which are not.
Field interpretation and error-repair cases
Two habitats with different rates
Reef operations average 15 kg/h and shelf operations 4 kg/h in both years. If effort shifts from mostly shelf to mostly reef, pooled CPUE rises without either habitat becoming more productive. Standardise to a declared spatial target or report habitats separately; do not describe the compositional effect as stock recovery.
Unreported unsuccessful trips
If unsuccessful trips are less likely to be entered into a voluntary logbook, reported CPUE can be biased upwards. A model cannot recover missing zeros from the observed log alone without additional assumptions or independent sampling. Compare vessel monitoring, observers, landings and audit records, and describe what remains unobservable.
A new electronics package
A fleet acquires improved acoustic fish finders. Higher catch per hour might reflect greater search efficiency rather than biomass. Whether to model equipment as a catchability covariate depends on the purpose of the index and support in the data. Document the introduction date rather than silently attributing its effect to year.
One exceptionally successful set
A single large catch may dominate the arithmetic mean. It might be a real aggregation encounter, a duplicate or a conversion error. Verify the record, examine heavy-tailed residuals and compare sensitivity to robust or appropriate distributional choices. Do not delete it merely because it changes the annual index.
Changing species names
Historical data may use common names or taxonomic codes that later split into several species. A trend that mixes taxa is not a species-specific index. Reconcile codes using verified taxonomic mapping and mark records that cannot be resolved; never silently relabel all old catches as the newer species.
An observer coverage change
Observer coverage expands from a few large vessels to many small ones. Average observed CPUE changes even if catch performance within each vessel class does not. The sampling frame should be explicit and a common target population considered before comparing annual predictions.
Missing units
A table labels effort simply ‘hours,’ while notes show that only some rows represent trawl-hours and others engine-hours. Without a consistent effort construct, dividing catch by the column is not meaningful. Reconstruct operational definitions from logs or exclude irreconcilable records with a transparent rule.
An area closure
Management closes the most productive grounds in the second year. Fleet CPUE falls, yet this need not mean overall stock biomass declined. Translate the change as an observation of the accessible fleet and discuss whether the standardised domain remains comparable when fishing can no longer occur in the closed area.
A skewed age or size target
Fishers switch to targeting larger individuals because market prices change. Weight CPUE may rise even while fish counts fall. Measure catch composition and align the index with the vulnerable component of the stock. Numbers and kilograms answer different questions and cannot be equated by an arbitrary average weight.
Too few positive catches
When a rare species appears in only a handful of trips, the positive-catch model may be unstable. Aggregating years, using an appropriate hierarchical structure or reporting that a reliable annual index is unavailable can be more defensible than publishing an irregular precise-looking line.
A changing baseline
An index series recalculated with a new reference period may jump numerically even when predicted catch rates are identical. Report the denominator and, for comparison, transform both series onto a common baseline. Do not call normalisation a change in stock conditions.
A misleading percentage
A fall from index 1.0 to 0.5 is a 50% decline relative to the first value. A later rise from 0.5 to 1.0 is a 100% increase relative to the lower starting point. The apparent asymmetry is arithmetic, not a biological paradox. State the reference and denominator for every percentage.
Trip versus haul data
A trip may include several correlated hauls under one vessel decision. Treating each haul as independent when the sampling design has vessel-trip clusters can make uncertainty too narrow. Model the clustering or use a resampling plan respecting the dependence structure; a larger row count is not always more independent information.
Effort measured after catch
If reported effort is adjusted according to fishing success, denominator errors may be correlated with the outcome. This creates more than ordinary random measurement noise. Reconstruct independent exposure records where possible and explain any remaining endogeneity instead of assuming effort is an error-free covariate.
Spatial prediction into unvisited water
A map may colour an entire region with modelled CPUE even though observations cover only its coastal edge. Separate interpolation, extrapolation and unavailable areas. A polished continuous surface is not evidence that fish were sampled everywhere, and an index relying heavily on unsupported habitat needs explicit limits.
Discard reporting reform
A new rule makes vessels record discards. Total catch CPUE may rise even if encounter frequency has not changed, while landed catch remains similar. Identify the numerator’s definition in each year and harmonise records before translating the increase into a statement about availability.
A model with vessel effects
The best captains may remain in the industry while others exit. Including vessel effects can reduce bias from changing fleet composition, but vessels appearing in only one year cannot independently identify vessel and year effects. Report connectedness, shrinkage or comparison constraints rather than hiding the identification problem.
Environmental confounding
Warm-water years may move fish deeper and change hook encounters without changing biomass in the same proportion. If a temperature covariate is introduced, describe whether it is intended to adjust catchability, availability or biology. Do not automatically label any resulting flat index as the true stock trajectory.
Catchability and schooling
A schooling species may allow high CPUE in shrinking aggregations. Compare spatial occupancy and the effort required to locate schools. Search time and failed operations may tell a different story from successful-set CPUE. A single strong encounter is not evidence that the regional population is secure.
Cross-gear pooling
A trawl and a longline capture different species and size fractions. Combining their effort hours as though they were equivalent can destroy the index’s meaning. Gear-specific standardisation or explicit calibration is needed, and a combined result must name the constructed reference rather than imply a direct natural unit.
A change in catch processing
One fleet reports whole fish mass while another reports dressed weight. Even an accurate kilogram denominator gives incomparable numerators. Apply documented species-specific conversion factors where appropriate, carry the uncertainty, and preserve the original measurements for audit.
Regression on transformed catch
An analyst exponentiates a predicted log catch to produce an arithmetic mean. Because of the nonlinear transformation, this may underestimate the expected mean without a variance correction. Specify the response scale and predict the required expectation under the fitted distribution rather than treating exponentiation as mere presentation.
Confidence without validation
A model returns a narrow interval but fails on years withheld from fitting. The interval may describe parameter uncertainty conditional on a poor model. Predictive checks, blocked cross-validation and alternative specifications are important because mathematical precision is not the same as external validity.
Regulatory decision language
A minister asks whether the index of 1.3 permits 30% more catch. The correct response is that relative CPUE is not a quota formula. Advice requires population dynamics, catch histories, risk and biological reference points. Translate the observed signal accurately and keep the decision rule separate from the observation.
Sources and related verified learning
NOAA Fisheries and Hoyle et al., Catch per Unit Effort Modelling for Stock Assessment: A Summary of Good Practices (2024): https://www.fisheries.noaa.gov/resource/peer-reviewed-research/catch-unit-effort-modelling-stock-assessment-summary-good-practices
NOAA Fisheries, Stock Assessment Model Descriptions: https://www.fisheries.noaa.gov/insight/stock-assessment-model-descriptions
FAO fisheries length-based assessment methodological review: https://www.fao.org/4/t0535e/t0535e00.htm
Related eduKate tag–recapture abundance guide: https://edukatesg.com/2026/10/30/how-to-translate-easily-fish-tag-recapture-chapman-abundance-tag-loss-reporting-bias/
Related eduKate swept-area and catchability guide: https://edukatesg.com/2026/10/30/how-to-translate-easily-swept-area-trawl-survey-fish-density-catchability-biomass/
Related eduKate growth guide: https://edukatesg.com/2026/10/30/how-to-translate-easily-fish-von-bertalanffy-growth-linf-k-t0-age-length-uncertainty/
Google Search Central, people-first content principles: https://developers.google.com/search/docs/fundamentals/creating-helpful-content