What a selectivity curve measures
Fishing-gear selectivity describes how the probability of capturing or retaining a fish can depend on its length, shape, behaviour, the gear and the way that gear is used. The central caution is that a size-selective catch is not a direct picture of the length distribution living in the sea. Small fish may escape through meshes, large fish may avoid hooks of some sizes, and intermediate fish may be especially likely to become entangled in a gillnet. A reader must distinguish the size distribution of fish available, the chance of encountering the gear and the probability of retention after encounter.
In trawl codends, a common retention curve rises from low probability for shorter fish toward high probability for longer fish. Its shape is often described by a logistic or other sigmoid function. L50, in this context, is the fish length at which the model predicts 50% retention for fish exposed to the selection process. It is not the length at which 50% of fish are sexually mature, even though maturation studies also use the label L50. The two curves concern different events and cannot be substituted for one another.
An effective translation names the conditioning event. ‘Fifty percent retained at 30 cm’ can mean that, in the specified experimental configuration, approximately half of fish of that length that reach the relevant selection stage are retained rather than escape. It does not imply that half of every 30 cm fish in a stock is caught by a single fishing trip. Encounter probability, capture efficiency, availability and fishing effort remain outside the retention proportion unless a broader model explicitly includes them.
Gear selectivity matters for fisheries science and conservation because biased sampling changes the observed catch composition, which may affect assessments of growth, age structure and biomass. It matters operationally when managers compare mesh modifications or the fate of undersized fish. However, a curve established for one species, mesh, twine, codend design, current, towing condition or catch load cannot automatically be transferred to all fishing conditions. The boundary of the experiment is part of the scientific claim.
The logistic retention equation explained
A convenient ascending logistic curve is S(L)=1/{1+exp[−a(L−L50)]}, where S is a probability between zero and one, L is fish length, a is the positive slope coefficient with units reciprocal length and L50 is the length giving 0.5 retention. The expression a(L−L50) must be dimensionless. If lengths are recorded in centimetres, the slope coefficient needs units cm⁻¹. The function is symmetric on the log-odds scale, not necessarily on a biologically meaningful length scale.
At L=L50, the exponent is zero; exp(0)=1, so S=1/(1+1)=0.5. That identity is a valuable sanity check. For fish shorter than L50 and a positive slope, S is below one half; above L50 it is greater than one half. A negative a would reverse the direction and should not be quietly accepted when the intended model describes larger fish being more likely retained by a trawl codend. Other gears require different curve shapes.
A frequently used alternative expresses the curve through L50 and L95: S(L)=1/{1+exp[−ln(19)(L−L50)/(L95−L50)]}. This follows because logit(0.95)=ln(0.95/0.05)=ln(19), approximately 2.9444. With L95 above L50, a=ln(19)/(L95−L50). The difference L95−L50 controls steepness: a small difference gives a sharp change in retention over a narrow length span, whereas a large difference gives a gradual transition.
Suppose L50=30 cm and L95=36 cm. The slope coefficient is ln(19)/6≈0.4907 cm⁻¹. At L=33 cm, the logistic exponent magnitude is 0.4907×3=1.4722, giving S≈1/(1+exp(−1.4722))≈0.813. About 81% retention is predicted for fish of that length under this illustrative curve. This is a model-predicted probability, not a direct observation that exactly 81 of every 100 fish on every haul will be retained.
At L=24 cm, the term a(L−L50) is −2.9444, so S=1/(1+exp(2.9444))=0.05. Thus the illustrative model yields L5=24 cm, L50=30 cm and L95=36 cm because of its logistic symmetry on the length scale. The fitted curve should still be checked against observations rather than assuming symmetry describes every biological gear interaction.
Selection range, mesh size and the denominator
L25 and L75 are lengths at which predicted retention is 25% and 75%. Their difference, SR=L75−L25, is called the selection range in many trawl studies. For a logistic curve, logit(0.75)=ln(3) and logit(0.25)=−ln(3), so SR=2ln(3)/a. Using a≈0.4907 cm⁻¹ gives SR≈4.48 cm. A smaller selection range means a steeper transition around L50; it does not by itself prove improved stock sustainability.
A selection factor may be defined as L50 divided by the measured mesh size, with both dimensions in the same length unit. If L50=30 cm and mesh size=10 cm under a particular measurement method, the factor is 3.0 and has no dimension. But mesh size measurement protocols vary: stretched mesh, opening, mesh gauge and material deformation can produce different values. Stating only ‘mesh 100’ without units or measurement convention makes the selection factor unverifiable.
The denominator of observed retention also needs scrutiny. In a covered-codend experiment, fish in the codend and those collected in the cover can provide a basis for estimating retention among fish entering that experimental stage, assuming the cover itself does not strongly change escape behaviour or lose fish. A simple observed fraction for a length bin is retained count divided by retained plus escaped count. It is conditional on what the experiment detected and what its sampling design represents.
Suppose 60 fish at 30 cm appear in the codend and 40 in the cover. The observed retention fraction is 60/(60+40)=0.60. That single bin does not prove the fitted L50 is 30 cm; sampling variation, design effects and the rest of the curve matter. If only 20% of the cover catch is subsampled and not raised appropriately, the naive fraction will overstate retention. Sampling fractions and different capture efficiencies belong in the estimation.
When comparing studies, an identical nominal mesh size need not yield the same retention curve. Twine thickness, number of meshes around the circumference, catch volume, towing speed, fish morphology, water conditions and cover effects can influence escape. A reported shift in L50 is descriptive evidence of a gear-condition difference under the study’s design, not automatically proof that exactly that shift will occur in every commercial tow.
A complete worked example with interval checks
Consider a purely hypothetical paired-codend experiment with estimated L50=28 cm and L95=34 cm. First compute a=ln(19)/(34−28)=2.9444/6=0.4907 cm⁻¹. At 28 cm, S=0.5 by definition. At 34 cm, a(L−L50)=2.9444, and S=1/(1+exp(−2.9444))=0.95. That second calculation verifies the parameterisation and catches the common mistake of using ln(20) instead of ln(19).
What is predicted at 31 cm? The length difference is 3 cm, giving a product 1.4722. exp(−1.4722) is roughly 0.229. Then S≈1/1.229≈0.813. The intermediate retention is about 81.3%, as expected for a length halfway between L50 and L95 on a logistic scale where the response is not linear in probability. A careless arithmetic average of 50% and 95% would give 72.5%, which is not the logistic model’s answer.
At 25 cm, the product is −1.4722 and retention is about 0.187, or 18.7%. The curve therefore predicts substantial escapes for that smaller length. If an observed length class showed 40% retention at 25 cm, this discrepancy might reflect sampling noise, model misspecification or a different operational condition. The appropriate response is to inspect sample sizes and confidence intervals and evaluate the fit; one point cannot be used to claim the entire fitted curve is fraudulent.
Find L75 algebraically. Since logit(0.75)=ln(3)≈1.0986, L75=28+1.0986/0.4907≈30.24 cm. Similarly L25=28−1.0986/0.4907≈25.76 cm. The selection range is therefore approximately 4.48 cm. Check that the two lengths straddle L50 symmetrically in this logistic model, and that the probabilities at those values are indeed 25% and 75%.
Now consider an experiment reporting L50=28±2 cm without identifying the interval type. The ±2 might be a standard error, confidence interval half-width or another summary. It cannot be translated as the range containing 95% of individual fish lengths. A confidence interval for L50 reflects parameter uncertainty; individual fish may differ in condition and behaviour even if the curve parameters were known exactly.
Gillnets and hooks are not necessarily ascending sigmoids
A trawl codend generally allows smaller fish to pass and tends to retain larger fish, producing a common ascending sigmoid. Gillnets may capture fish most efficiently near an optimum size relative to mesh, with both smaller and larger fish less likely to be gilled. Their selectivity curves can be bell-shaped or sometimes multi-peaked because fish can be gilled, wedged, snagged or entangled by different mechanisms. Applying an ascending logistic everywhere can incorrectly imply that the largest fish are always most vulnerable.
For a bell-shaped selectivity curve, the mode or optimum length is the length of maximum retention or relative selection under the fitted model. That is a different concept from L50 on an ascending codend curve. Some reports scale selectivity so the highest fitted value is one; others attempt absolute capture probabilities. A value of one on a relative curve may denote the most selected size, not certainty that every encountered fish is captured.
Longline selectivity can also have different shapes depending on hook dimensions, bait, target behaviour, fish mouth morphology and deployment. A report describing one hook size as ‘selective for large fish’ may mean it catches a greater fraction of large fish relative to small fish, not that small fish can never be hooked. Gear modifications can change encounter, hooking, injury and release survival separately. They should not be collapsed into one unqualified selectivity claim.
When translating a gillnet study, ask whether lengths are divided by mesh size under a geometric similarity assumption. If a fitted master curve uses L/m, the proportionality may help compare nets, but it is not a law independent of species morphology and gear construction. Differences in twine, hanging ratio or fish body shape can violate the assumed similarity. A strong explanation tells the reader which relationships are assumed and which were experimentally observed.
How experimental design supports the curve
A paired-gear or covered-codend trial should be planned to expose fish of relevant lengths to comparable conditions while distinguishing retained and escaped groups. Randomisation or rotation of gear positions can reduce confounding by location and time. If a cover changes net shape or flow, the experiment itself can influence escape. The analyst should describe where fish were counted, whether subsampling occurred and how damaged or unmeasurable specimens were handled.
The length range sampled must cover the transition. If all collected fish are much larger than L95 and virtually all are retained, the data do not identify L50 precisely. Conversely, if almost every fish escapes because sampled lengths are far below the transition, the upper portion is weakly informed. Good estimation requires exposure across the part of the curve where selection changes, plus enough observations to distinguish the fitted shape from chance.
Binomial models may be used when retained and escaped counts are observed under conditions supporting that likelihood. Overdispersion and clustered hauls can make simple binomial standard errors too narrow. A mixed or hierarchical model can represent haul-to-haul variation, while bootstrap procedures should respect haul as a sampling unit when appropriate. Counting thousands of fish in one tow does not create thousands of completely independent experimental gear deployments.
In some experiments, retention is inferred indirectly by comparing different mesh sizes without directly observing all escapes. The resulting estimates depend on assumptions about the fish population available, relative fishing power, equal exposure and functional curve shape. A caught-length histogram alone generally cannot disentangle gear selection from underlying fish abundance at each length. The limitation is not just statistical inconvenience; the two influences are multiplicatively intertwined.
Report mesh measurements with their method, gear dimensions, codend circumference, towing conditions, species, length convention, number of hauls, retained and escaped counts, model form, parameter uncertainty and any subsampling weights. Without these, another analyst can reproduce neither the fitting nor the operational meaning. A figure alone cannot communicate whether it is a theoretical curve, an experiment, or a model borrowed from another fishery.
Selectivity, mortality and stock inference
A fish that escapes the codend is not necessarily a surviving fish. Contact with mesh, physical injury, predation after escape, temperature stress and handling can influence delayed mortality. Gear selectivity is typically about capture or retention; survival needs separate evidence. A conservation claim about reduced fishing mortality requires both the selectivity effect and survival consequences, along with the fleet’s response.
Fishing mortality at age or length may depend on encounter, accessibility, selectivity, effort and subsequent survival. A curve giving 60% retention at 30 cm cannot be read as ‘40% of the stock will survive the fishery.’ That would ignore how many individuals encounter the gear, repeat encounters, other gears and natural mortality. When a stock model uses selectivity, it should define whether the curve is relative vulnerability, retention probability or another composite component.
Fishing selectivity L50 and maturity L50 are separate thresholds. Imagine a species with 50% maturity at 32 cm and codend L50 at 28 cm. One might worry that the gear retains many fish before maturity, but a robust judgement also needs the whole selectivity and maturity curves, fishing effort, survival, reproductive output and population dynamics. Comparing two point estimates alone cannot establish a safe mesh rule or a sustainable quota.
Increasing mesh size sometimes shifts retention toward larger fish, but the response depends on gear geometry and behavioural conditions. Fishers may also change location, towing speed or effort when catch composition changes. That compensatory behaviour can affect the intended biological result. Evaluation should examine field performance and management context, not treat one theoretical increase in L50 as a guarantee of improved outcomes.
Stock assessment models may infer selectivity jointly with abundance, mortality and recruitment, often under parametric constraints. These components can be confounded. A changing age or length composition may arise from changing recruitment, shifting stock distribution, altered gear or selective mortality. Independent survey or experiment evidence can help constrain the inference. The translation should be clear about whether a curve was measured experimentally or estimated indirectly within a population model.
Uncertainty, diagnostics and communication
A selectivity estimate should be displayed as a fitted curve with observed length-class retention summaries, sample sizes and uncertainty where available. If the curve is conditional on a particular gear configuration, label that condition in the figure. Intervals at lengths far outside the sampled transition should be interpreted cautiously. An apparently precise sigmoid can be strongly determined by the chosen model family when the data barely touch either tail.
L50 and L95 estimates are often correlated because they both depend on the slope and position of the same curve. Propagating uncertainty into a length-specific retention prediction requires their joint distribution or a suitable fitting framework; changing each independently by one standard error can produce unrealistic extreme combinations. A credible figure communicates predicted retention and uncertainty without pretending every fish at a particular length behaves identically.
A logistic model may fit the middle of the curve yet fail for short or very large fish. Residual patterns, overdispersion, repeated-haul effects and alternate model shapes should be checked. For gillnets, the absence of the descending limb from observations is especially important. A bell curve with only its ascending side observed can look logistic locally, and extrapolating it indefinitely may be profoundly wrong.
Translate the term ‘selection range’ with its exact definition. Some studies use L75−L25, others report L95−L50 or the full L95−L5 difference. These are not numerically interchangeable. The formula and quantile labels should accompany the reported width. A difference of 6 cm between L95 and L50 means something different from a 6 cm L75−L25 range.
A sensible publication identifies actual observations, model predictions and downstream policy judgments as separate layers. ‘This tested codend had an estimated 50% retention length of 28 cm’ is an observational/model statement. ‘Changing mesh will reduce juveniles caught’ is a prediction needing comparison under field conditions. ‘The fleet should adopt this mesh immediately’ is a management decision needing broader risk and feasibility evidence.
Independent transfer problems and checked answers
Problem one. A trawl study gives L50=25 cm and L95=31 cm. Find the logistic slope. Answer: a=ln(19)/(31−25)≈2.9444/6≈0.4907 cm⁻¹. At L=25 cm, predicted retention is exactly 0.5; at L=31 cm it is exactly 0.95 under this parameterisation. If a is reported as 0.4907 cm per year, the units are wrong.
Problem two. Using those parameters, calculate retention at 28 cm. Answer: the difference from L50 is 3 cm. The exponent’s magnitude is 0.4907×3≈1.4722. S=1/(1+exp(−1.4722))≈0.813. A linear interpolation of the 50% and 95% probabilities would produce 72.5%, but logistic probabilities are nonlinear in length.
Problem three. An experiment finds 18 retained and 42 escaped fish in a length class. Calculate the observed fraction. Answer: 18/(18+42)=18/60=0.30. The denominator is all observed fish in that class in the two relevant experimental compartments, not just those in the codend. Check whether different sampling fractions require raising the counts.
Problem four. A study has L75=35 cm and L25=29 cm. What is the selection range? Answer: 35−29=6 cm. This is not the same as L95−L50 unless the model and conversion are specified. The phrase ‘six-centimetre selectivity’ without quantile names is ambiguous.
Problem five. A report gives L50=28 cm and mesh size=70 mm. Compute the selection factor. Answer: convert 70 mm to 7 cm. The factor is 28/7=4, dimensionless. Dividing 28 by 70 without unit conversion would produce 0.4, a tenfold numerical mistake.
Problem six. A text says ‘the L50 is the fish size where half are mature’ while citing a covered-codend retention experiment. Repair it. Answer: the retention L50 is the fish length with a modelled 50% chance of codend retention under the experimental conditions. Maturity L50 requires reproductive maturity observations and describes a different biological event.
Problem seven. A gillnet catches few small fish, many intermediate fish and few very large fish. Must the selectivity curve be logistic? Answer: no. A dome-shaped or more complicated curve may be plausible because intermediate sizes can mesh or become entangled most efficiently. But the catch histogram alone also contains the population length distribution, so experimental evidence is required to identify selectivity.
Problem eight. A study reports only retained fish and no independent information about available fish or escape. Can it directly estimate an absolute capture probability by length? Answer: generally not. A length-frequency catch may reflect both abundance at length and selection. Specify the assumptions and seek covered-gear or comparative experiments before claiming a probability.
Problem nine. A confidence interval for L50 is 26–30 cm. Does this say that 95% of fish are retained between 26 and 30 cm? Answer: no. If it is a 95% confidence interval, it describes estimation uncertainty for the length of the 50% retention point under the model. It is not a band of lengths comprising the catch or a probability of retention for an individual fish.
Problem ten. A fish escapes but is injured and later dies. Is escape equivalent to survival? Answer: no. Physical retention and post-escape survival are separate endpoints. A conservation analysis should consider injury and delayed mortality where evidence exists, rather than equating an escape percentage with avoided mortality.
Problem eleven. A logistic curve predicts S(35 cm)=0.8. A report says every five 35 cm fish will yield exactly four retained. What is wrong? Answer: probability is an expectation across relevant trials, not a deterministic count in each small group. Variation and correlated exposure can make actual outcomes differ. Specify the experimental conditions and uncertainty.
Problem twelve. An analyst uses L95=30 cm and L50=35 cm in an ascending retention logistic. Diagnose. Answer: L95 should exceed L50 for a monotonic ascending curve with positive slope. The supplied ordering would give a negative coefficient and a descending curve. Check whether labels were reversed, the response is escape rather than retention, or a different model was intended.
Frequently asked questions
Is gear selectivity the same as the length distribution of the catch? No. The catch distribution combines fish availability, encounter, retention and sampling. A selectivity curve isolates a conditional component only under appropriate experimental or model assumptions.
What does L50 mean in a trawl codend? It commonly denotes the fish length with 50% predicted retention during the defined selection process. It must not be confused with L50 of reproductive maturity or a legal minimum landing size.
Is L95 a guarantee that 95% of fish longer than that will be kept? Under a monotonic fitted sigmoid it means 95% predicted retention at that particular length, increasing at larger lengths within the model. It is not a guarantee for every haul, species or condition, nor does it describe full stock capture.
Why do gillnet curves often look bell-shaped? Fish near an optimum size can be especially vulnerable to meshing, while much smaller or larger fish may pass through, avoid or interact differently. Entanglement and other mechanisms can produce asymmetry or multiple peaks.
What is a selection factor? In many trawl contexts it is L50 divided by mesh size under the stated mesh measurement convention. Both measurements require matching units and relevant gear details.
Can a larger mesh always save juveniles? No universal guarantee follows. Mesh changes can alter size retention, but exposure, species morphology, fish behaviour, injury, escape survival and fishing practice also matter.
How should uncertainty be shown? Ideally show observed retained and escaped counts by length, modelled retention, sample sizes, estimation uncertainty and relevant haul effects. Explain whether an interval describes curve estimates or individual outcomes.
What makes a selectivity report reusable? Clear gear geometry, mesh protocol, species, length convention, environmental and tow conditions, experiment type, sampling fractions, model definition and uncertainty. Without these, a beautifully drawn curve can be misleading.
Practical case diagnostics and repairs
The wrong denominator in a codend comparison
A table reports 70 retained fish and 30 fish in a cover, then calculates 70/30=2.33 and calls it ‘retention 233%.’ Retention as a proportion of detected exposed fish is 70/(70+30)=0.70 under the design’s assumptions. A ratio of retained to escaped fish is an odds-like quantity, not a probability. The label determines the appropriate denominator.
A different haul loading
A codend packed with a large catch may have meshes obstructed by fish and debris, changing opportunities for smaller fish to escape. A curve fitted during light loads may be poorly transportable to heavy commercial hauls. Include catch loading or haul effects when evidence supports them, and do not describe a gear’s selectivity as independent of operation.
Mislabelled length conventions
An investigator compares a selection L50 in fork length with a maturity L50 in total length. Even if both were measured on the same population, the numbers need a validated conversion before comparison. The disagreement may be measurement convention rather than biological process. Label FL, TL or SL in every figure and results sentence.
A cover that changes water flow
A small-meshed cover attached over a trawl codend is meant to collect escapements. It may also alter water flow, shape or mesh opening, changing the process being measured. Cover effects are a design limitation, not necessarily a fatal flaw, and should be assessed by experimental controls or sensitivity analysis rather than ignored.
A mesh specification without a method
‘Mesh size 100’ has no usable unit, no indication of stretched or open mesh and no measurement protocol. A selection factor computed from it cannot be reproduced. Recover the original specification and its gauge method, then convert both mesh and fish lengths to compatible units before interpreting a reported ratio.
A rising sigmoid imposed on a gillnet
A model forces capture probability to increase forever with fish length, yet field evidence suggests large fish are less likely to mesh. The model may fit a limited ascending interval while failing beyond it. Compare dome-shaped models and the observed length range, and explain why extrapolated high-length vulnerability is not established.
An apparent impossible length
A catch includes a fish longer than the reported L95. That is completely compatible with an ascending logistic, where retention approaches one for longer fish. L95 marks a probability quantile, not a maximum possible size. The same reasoning explains why fish shorter than L5 can occasionally be retained.
An inappropriate maturity comparison
A result says selectivity L50 increased from 27 to 30 cm and calls this ‘maturity increasing.’ The gear’s retention curve shifted; no reproductive data were supplied. The correct interpretation is a change in predicted retention size under the specified test. Maturity requires separate gonad or reproductive-status evidence.
One haul treated as many independent trials
A single tow yields hundreds of fish, and a binomial analysis treats all outcomes as independent despite shared gear configuration and environmental conditions. The fitted curve can have understated uncertainty. Model haul-to-haul variation or resample hauls where justified, and distinguish fish sample count from experimental replication.
A steep curve mistaken for high catch
A small selection range indicates a rapid transition from low to high conditional retention over a narrow size span. It does not say the gear encounters many fish or has high total fishing power. A sharp curve and low overall catch can coexist if fish density or encounter probability is low.
An unobserved escape group
If escaped fish cannot be counted and no comparative sampling design exists, catch data alone do not identify retention probabilities. It is misleading to fit a logistic to caught lengths and call it an absolute selectivity experiment. At most, under explicit assumptions, one can explore relative vulnerability combined with a population model.
An unexplained percentage above one
Some studies scale relative selectivity differently or present indices rather than probabilities. A value of 1.2 cannot be interpreted as a 120% probability. Check whether the scale was normalised to a reference gear or a baseline selection coefficient. Retention probabilities must stay within zero and one.
Species mixed within length bins
The same net can retain two species differently at identical lengths because body depth, girth, behaviour and flexibility differ. If species are pooled before fitting, a changing species composition can shift apparent selectivity. Analyse species-specific evidence where possible and avoid claiming a universal mesh rule from a mixed catch.
The hook size inference
A larger hook may change size selectivity but can also change bait presentation, hooking location and handling outcomes. Reporting a shift in fish lengths caught cannot alone prove the causal effect of hook size if fishing depth, bait or season changed simultaneously. Use controlled comparisons and keep retention separate from survival.
A neglected conversion
If an analyst inserts mesh size as 80 mm and fish length as 30 cm into L50/mesh, the result is 0.375 instead of the correctly converted 30/8=3.75. Ratio equations may appear unitless, but their inputs must share units before cancellation. This simple check prevents a tenfold error.
A heavily sampled central range
A curve may have thousands of observations close to L50 but very few on either tail. Its central estimate could be stable while L5 and L95 are sensitive to model form. Show the observed length support and uncertainty in the tails, especially when regulation focuses on high-retention thresholds.
The regulatory leap
A policymaker asks for a mesh change solely because L50 increases by 2 cm. A technical result may support the direction of size retention, but a management decision also considers stock maturity, mortality of escapes, total catch, compliance and fleet response. Preserve the evidence chain instead of turning a conditional gear measurement into a standalone policy mandate.
Recovery of lost experimental counts
A file includes only subsampled escape counts from the cover. If the fraction sampled varied by haul, treating the raw counts as full escape totals biases observed retention. Apply documented raising factors and propagate their uncertainty, or acknowledge irrecoverable gaps. The problem belongs to sampling design, not to a different logistic equation.
A selection range without quantiles
An article says ‘selection range of five centimetres’ but omits whether it means L75−L25, L95−L5 or another interval. These differ substantially for the same fitted logistic curve. The repair is to include the precise percentiles and mathematical definition before comparing selectivity across studies.
Selection versus catchability
Two gears have identical relative selection curves by length but one catches far more fish at every length. Their fishing power or encounter rates can differ even when the shape of selection is similar. A normalised curve strips out scale, so a comparison must retain the separate effort and catchability evidence if absolute capture is the question.
A shift in tow speed
An experiment compares codends across years but changes towing speed. Net tension, mesh opening and fish swimming behaviour may change with speed. Year and gear modifications are confounded unless the design separates them. Translate the measured curve for the actual operating conditions rather than an idealised net independent of use.
A minimum landing size
A legal minimum landing size is a regulation about what may be retained or sold; selectivity L50 is a biological/technical estimate of capture probability in the gear. They may have similar numbers but are not the same. Discards and escape survival must be considered when discussing how regulations interact with gear selectivity.
A reader misinterprets 0.05
The model predicts S=0.05 at a length class, and a summary says no fish of that size can be captured. Five percent is not zero. Under sufficiently many exposures, some fish may be retained. The proper sentence is ‘low predicted retention under the model’s tested conditions,’ with uncertainty and exposure clearly named.
A cross-study difference
One experiment reports L50 of 29 cm and another 32 cm. Before concluding a temporal change, compare mesh protocol, species, length definition, gear construction, sampling, haul loading and uncertainty. The curves may be valid for different operational systems. Harmonisation comes before causal explanation.
First-party methodological sources and related guides
FAO, Conceptual Clarifications — Size Selectivity, L50 and Selection Range: https://www.fao.org/4/Y4593E/y4593e03.htm
FAO, Manual of Fisheries Science, Mesh Selection: https://www.fao.org/4/F0752E/F0752E09.htm
FAO, Manual on Estimation of Selectivity for Gillnet and Longline Gears in Abundance Surveys: https://www.fao.org/4/x7788e/X7788E03.htm
FAO, Estimating Gillnet and Longline Selectivity Curves from Experiments: https://www.fao.org/4/X7788E/X7788E05.htm
Related eduKate maturity-ogive guide: https://edukatesg.com/2026/10/30/how-to-translate-easily-fish-maturity-ogives-l50-spawning-biomass/
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/
Google Search Central people-first content guidance: https://developers.google.com/search/docs/fundamentals/creating-helpful-content