Fisheries Spatial Closures, BACI and Spillover

A closed area is not a controlled experiment by itself

A fisheries spatial closure changes where fishing is allowed, but an observed increase in fish abundance inside a closed area is not automatically proof that the closure caused the increase. Fish can move, climate conditions can change, sampling gear can improve, and managers may have selected a naturally productive area for protection. An evaluation therefore needs an explicit comparison over time and space. Before–after control–impact, often shortened to BACI, compares the change in a treated area with the change in an appropriate comparison area. Its core idea is simple: what changed where the intervention occurred beyond what also changed somewhere comparable without that intervention?

Marine protected areas, seasonal closures, gear-restriction zones and temporary fishing exclusions are not identical policies. One may prohibit all extraction, another only selected gear, and a third may operate for a short spawning season. Evaluating their effects requires knowing exact boundaries, activities prohibited, start date, enforcement, compliance, and the species or outcomes of interest. A report that says fish recovered after the reserve opened should identify whether the measured outcome was fish density, biomass, juvenile recruitment, catch rate outside the boundary, or a socioeconomic result for displaced vessels. Each outcome can move differently, and none represents every consequence of the closure.

The four BACI cells

A basic BACI design records an outcome before and after an intervention in both the impact site and a control site. Consider invented standardized fish counts per survey station. The impact area has mean 6 before the closure and 10 afterward, a change of +4. The control area has mean 5 before and 7 afterward, a change of +2. The difference of changes is (+4)−(+2) = +2 detections per station. This estimates the additional change associated with the intervention under the design’s causal assumptions. It is not the same as the raw before–after gain of four, because two of those four units might reflect a broader regional trend.

A second expression uses a ratio. In the same example, the impact area’s proportional change is 10/6, approximately 1.667, while the control area’s proportional change is 7/5 = 1.4. Dividing these ratios gives approximately 1.190, corresponding to a 19 percent relative difference in the changes under a multiplicative framing. This percentage is not interchangeable with the additive effect of two detections per station. Different effect scales answer different questions, and zeros complicate log-ratio methods. A scientific article should identify whether its reported BACI effect is an additive contrast, a log response ratio, a model coefficient, or a transformed estimate with confidence bounds.

A control site must be more than a place marked outside the closure. It should resemble the treated site in relevant habitat, fishing pressure, ecological trends and observation methods, while avoiding direct contamination from the intervention. If the comparison area receives displaced fishing vessels, it may no longer represent what would have happened without the closure. If the control is much deeper and colder, fish might respond differently to a marine heatwave. These are not minor aesthetic choices on a map. They affect the counterfactual that gives the BACI contrast a causal interpretation. A reported contrast can be calculated correctly and still be biased by an unsuitable comparison.

Time is part of the design

A single year before and a single year after protection can be misleading if fish recruitment varies naturally or weather prevents complete sampling. Multiple pre-intervention years help test whether treated and control sites followed broadly comparable trends before the policy. Multiple post-intervention years can reveal delayed biological responses and distinguish sustained change from a one-season anomaly. Slow-growing species may require longer observation windows than short-lived species. A report should state the time horizon and whether it includes an implementation phase when compliance was incomplete. The date a law was announced need not be the date fishing pressure actually changed.

Spatial replication matters too. Comparing one protected reef with one unprotected reef leaves site-specific peculiarities difficult to separate from treatment. Several independent protected areas and comparable control sites can make conclusions more general, provided sampling and statistical analysis recognize that stations within the same reef are not independent policy replicates. The treatment unit may be the entire closure, even if hundreds of underwater transects are recorded. Treating each transect as a separate independent closure inflates apparent sample size. A responsible analysis distinguishes station replicates, survey years, sites and policy units before computing standard errors.

Fishing effort moves

When fishing becomes unavailable in one area, some vessels may stop fishing, reduce effort or move to other grounds. This displacement can increase pressure just outside the boundary, alter travel costs, change total catch per trip, or redistribute economic benefits among communities. Consequently, outcomes measured outside a closure cannot be assumed to represent unchanged conditions. A nearby comparison zone may be ecologically similar but directly affected by displaced effort. Managers may need vessel tracking, logbooks, landing records and interviews to understand how fleets respond. These measurements describe actual fishing behaviour, not merely the boundaries drawn on a planning map.

For an invented displacement example, suppose 100 annual standardized fishing days previously occurred inside a now-closed zone. After closure, 60 of those days are transferred to adjacent grounds, 20 move to distant grounds, and 20 are not undertaken. Total recorded effort in the broader fleet would fall by 20 days if all other activity were unchanged, but adjacent pressure would increase by 60. A simplistic comparison of catch outside the closure could mistake this redistributed effort for biological spillover. To interpret catches, inspect effort, gear, vessels, season and fishing locations. Fish abundance and fishing activity are related but distinct processes.

Spillover needs a mechanism

Spillover can describe movement of adult or juvenile fish from protected habitat into adjacent fished areas. Larval export is another mechanism, involving reproduction, dispersal and recruitment over broader spatial and temporal scales. Increased catch near a boundary is not sufficient evidence on its own: vessels may concentrate there, targeting fish more efficiently, and catch per unit effort may change for reasons other than abundance. Tagging, telemetry, independent surveys, spatially explicit catch data and appropriate comparison designs can provide complementary evidence. A scientifically careful translation identifies which mechanism was tested rather than treating every increase in nearby catch as proof that fish swam out of the reserve.

Effects can take time. A protected stock may build biomass inside the boundary before older fish contribute significant recruitment or movement outside. Larval development, age at maturity, natural mortality and distance to suitable habitat all affect the timing of fisheries benefits. A NOAA-hosted 2021 methodological study investigated the lag between biological change within protected areas and projected changes in fishery yield outside. That does not establish one universal waiting period for all marine reserves. A report claiming that spillover must begin within two years, for example, would need species-specific evidence and assumptions. The causal story has to follow the life history.

Units, comparisons and claims

Biomass density measured in kilograms per hectare cannot be directly compared with catch per vessel-day without an explicit model connecting the measurements. A fixed-camera encounter index, transect counts, commercial landings and modelled spillover flux are different observables. Likewise, a 30 percent increase inside a closure does not mean the same 30 percent increase in total catch, nor a 30 percent improvement in nearby livelihoods. Every headline should attach the numerical result to its outcome, population, area and time. Good communication makes clear whether the claim is an ecological response, a fishery response, or a causal estimate under specified assumptions.

NOAA Fisheries publishes research on spatial choice and the economic effects of marine protected areas, including the possibility that different vessels face different costs when access to grounds changes. NOAA’s repository also provides primary studies of spillover using differences in vessel catch rates near and far from a large protected area before and after a boundary expansion. These demonstrate why temporal and spatial comparison are important, but they do not prove that every closure delivers the same outcome. This article’s purpose is to explain the causal logic, displaced effort and evidence requirements, rather than announce a universal benefit or cost.

Define the intervention before estimating its effect

A spatial closure can be defined at several levels, and those definitions must not be silently exchanged. The legal intervention is a boundary with specified activities prohibited during specified dates. The experienced intervention is the amount of fishing pressure actually removed, which depends on compliance and enforcement. The biological intervention is the consequent change in fishing mortality for particular species, sizes and seasons. An analysis that measures the legal boundary but claims a precise reduction in mortality has crossed two unmeasured steps. Write the treatment definition first, then list which intermediate quantities are directly observed, inferred from tracking data, or unknown.

For a seasonal example, suppose bottom-trawling is restricted for four months but hook-and-line fishing remains legal. A map coloured as a fully protected reserve would misdescribe both the policy and the ecological contrast. Even if vessel monitoring indicates fewer trawler-hours, the effect on one species depends on overlap with its habitat and whether alternative gears catch it. A defensible description says that a seasonal gear restriction reduced recorded exposure to one kind of gear within specified boundaries. It does not automatically say all extraction ceased, that all fish experienced equal protection, or that a population recovered. Language is part of the causal design.

Build a causal explanation instead of treating correlation as a verdict

Imagine a simple chain: closure policy changes access; access changes the location and intensity of fishing; altered fishing mortality changes survival; survival affects density, size structure and eventually reproduction; movement and recruitment affect catches beyond the boundary. Every arrow can be weakened by another cause. Enforcement may be uneven, vessels may switch gear, climate may change recruitment, and larvae may drift into and out of study areas. A useful report separates measured links from plausible links and from assumptions. This makes a strong explanation without pretending that one before–after graph has established the whole sequence.

Selection bias is particularly important when managers protect habitats that are already comparatively productive. Suppose managers choose an unusually healthy reef because it has rare coral and large fish. After protection, the reef may still outperform a damaged comparison site even if its improvement would have occurred anyway. The question is not whether the protected reef is healthier after a policy; it is how its trajectory compares with a credible estimate of the trajectory it would have followed without the policy. Matching on pre-policy habitat, trends, survey accessibility and fishing history can improve comparison, but no matching algorithm can manufacture the missing counterfactual with certainty.

A confounder is a cause that influences both exposure to protection and the outcome, such as management selecting areas with strong baseline abundance. A mediator is a consequence of the intervention on the path to the outcome, such as reduced fishing mortality. Adjusting for a mediator while estimating the overall closure effect can remove part of the effect one intended to measure. A collider is a variable affected by two other variables, and conditioning on it can introduce a false association. These are not just academic labels: identifying the causal role of enforcement, vessel effort and habitat quality determines which statistical adjustments clarify the result and which make it less trustworthy.

Check whether comparison trajectories were genuinely comparable

Difference-in-differences interprets the gap between before–after changes as an intervention-associated contrast. To read it causally, analysts usually need a parallel-trends assumption: without intervention, treated and control areas would have had comparable outcome trends in the units being analysed. Equal baseline abundance is not required, but comparable untreated changes are. Multiple pre-closure surveys are valuable because they may reveal different trends before the intervention. Similar looking lines for two years are supportive, not conclusive; the unobservable post-closure counterfactual remains an assumption. Report the pre-period and what environmental or management changes could break that assumption.

Consider another wholly illustrative series. A treated area averages 4, 5 and 6 fish per standard transect in three years before closure; a control area averages 7, 8 and 9. Both rise by one fish per year before treatment despite different levels. In the subsequent year the treated area is 11 and the control is 10. Comparing only the last pre-year with the after-year gives treated change +5, control change +1 and BACI contrast +4 fish per transect. The estimated contrast is meaningful only if the control’s post-period change is a defensible guide to the treated area’s unprotected trend. This example demonstrates an arithmetic estimand, not a measured ecological truth.

Now alter the control history. Suppose its pre-closure sequence is 7, 9 and 12 while the treated series remains 4, 5 and 6. Controls were already accelerating faster before intervention. A simple post-period difference-in-differences may confuse continuation of that faster trend with a closure effect. Analysts could seek more controls, include credible environmental predictors, model site-specific trends, or report the limitation as unresolved. None of these repairs is automatically valid: a complicated model can fit pre-period fluctuations while failing after intervention. An evaluation should expose this diagnostic before presenting a confident single number.

When sites are protected at different times, a pooled regression with site and time fixed effects may mix comparisons that use already-treated areas as controls. Under heterogeneous treatment effects this can produce weights that obscure the intended comparison. Modern designs often compare each newly treated group to not-yet-treated or never-treated groups, estimate time-relative effects and then aggregate transparently. The practical reading test is simple: ask which observations served as controls for which treated observations, during what periods. The phrase fixed-effects model alone does not settle validity. A reader should not need to inspect code to discover that the comparator was itself already protected.

Match the sampling design to the biological question

Sampling units should be described explicitly. A transect is a within-site observation; several transects on one reef do not automatically represent several independent reefs. If a closure was assigned at the reef scale, treating forty transects from that one reef as forty independent policy interventions produces pseudoreplication. Repeated observations from the same stations can also be correlated over time. Models may use site-level effects, temporal correlation structures or cluster-aware uncertainty, but the number and independence of policy-level units remain consequential. Always distinguish the count of individual observations from the count of independent treated and comparison areas.

Imagine two protected reefs and two comparison reefs, with ten transects visited on each reef each year. The dataset contains forty transect readings per year, but only four independent reefs for understanding reef-level policy differences. Reporting a confidence interval calculated as if all forty readings were independent is likely overprecise. Repeated observations help estimate each reef’s mean; they do not create thirty-six extra independent reefs. If the ecological conclusion depends on generalising to other reefs, explain how these four were selected, what habitats they represent and how much uncertainty remains because the policy-level sample is small.

Observation methods can create apparent change. Diver counts, baited remote cameras, acoustic surveys and bottom trawls do not observe an identical fraction of the population. Visibility, fish avoidance, bait attraction, camera placement, vessel speed and survey timing can all shift detectability. A closure may even make fish less wary, raising apparent visual encounter rates without the same proportional increase in fish numbers. Record the measurement protocol, equipment, observers and environmental conditions. If the survey changes method at the closure date, calibration data or an explicit discontinuity analysis is needed before claiming that the abrupt change is biological.

Missing stations need a reason code rather than silent deletion. A storm may cancel trips to exposed reefs in the after period, while sheltered reefs remain observable. If exposed reefs contain different fish communities, averages of the remaining stations can change solely because the sampled composition changed. Where scientifically justified, stratified estimators, model-based imputation or sensitivity bounds can explore this problem. The safest primary table reports planned stations, completed stations, missingness reasons, and coverage by area and period. An analysis can have flawless arithmetic and still be misleading if the denominator changed between surveys.

Calculate an area-weighted example with the right units

Suppose a study area contains two habitat strata. Shallow habitat covers 120 hectares and has estimated fish biomass density 60 kilograms per hectare; deep habitat covers 80 hectares and has 35 kilograms per hectare. A simple area-weighted biomass total is (120 × 60) + (80 × 35) = 7,200 + 2,800 = 10,000 kilograms, or 10 tonnes. The whole-area mean density is 10,000 / 200 = 50 kilograms per hectare. Taking the unweighted average of 60 and 35 gives 47.5 kilograms per hectare, which underweights the larger shallow stratum. Neither figure is automatically a fish-stock estimate beyond the sampled area.

For a hypothetical after survey, suppose shallow density rises to 70 kilograms per hectare and deep density to 40. The estimated total becomes (120 × 70) + (80 × 40) = 8,400 + 3,200 = 11,600 kilograms, an increase of 1,600 kilograms or 16 percent from 10,000. It is tempting to say protection caused a 16 percent improvement. That causal claim still requires a valid comparison, stable sampling and a defensible biomass estimator. The arithmetic is only a before–after descriptive change. If the sampling method measures an encounter index rather than density, multiplying it by hectares would be inappropriate unless a validated conversion exists.

Percentages often conceal denominators. If one protected reef gains 20 kilograms of fish per hectare from a baseline of 40, that is a 50 percent rise. A nearby reef gaining the same 20 kilograms per hectare from a baseline of 100 has a 20 percent rise. Neither percentage alone tells which reef contains more biomass after the change. A difference-in-differences in absolute density has units kilograms per hectare; a log-scale estimate relates to proportional changes. Readers should ask why the chosen scale matches the biological and policy question, and avoid switching scales only because one produces the larger-looking result.

Treat movement and spillover as quantities to test

Adult spillover usually means movement of individuals across a protected boundary in ways that may affect adjacent fisheries. A rise in catch just outside a reserve is consistent with some spillover mechanisms, but catches also respond to where vessels fish, market prices, catchability, species composition and reporting. Evidence strengthens when boundary-distance patterns are evaluated against pre-closure baselines, suitable comparison boundaries, habitat gradients and independent measures of fish movement. Tag returns, telemetry, size distributions and consistent abundance surveys offer complementary information. None can establish universal benefits for all stocks; species mobility and reserve size strongly shape the likely process.

A boundary-distance graph needs physical interpretation. Distance of a vessel from the nearest boundary is not necessarily distance from suitable fish habitat. If one side of the boundary contains a deep channel and the other a shallow shelf, a catch gradient may be a habitat gradient. Fishing vessels might also crowd a legal boundary because navigation and enforcement make its position conspicuous. Analyse effort and habitat alongside catches. Prefer coordinates, distances, time windows and fishing gear definitions to a phrase like nearby catches improved. A spatial association is a beginning for mechanism testing, not its conclusion.

Larval export is not the same as adult spillover. Reproductive adults inside a reserve may produce larvae that drift and settle elsewhere, but identifying where larvae came from can require oceanographic transport modelling, genetics, otolith chemistry or carefully calibrated demographic inference. Increased recruitment outside a reserve does not prove source attribution if ocean conditions changed across the region. A proper claim might be that observed recruitment patterns are consistent with potential larval connectivity under stated circulation assumptions. It should not be translated into a measured number of exported adults, because larvae and adults represent different life stages, units and survival filters.

Movement also creates interference between treated and control areas. If protected fish regularly enter a nominal control region, the control region is not unaffected by the treatment. If fishing vessels displaced from the closure intensify harvesting in that control region, the same problem appears through human behaviour. The classic difference-in-differences comparator assumes the control trajectory is not changed by treatment; spillover can violate that assumption. This is especially awkward because detecting spillover may be one goal of the study. Analysts may use more distant controls, spatial exposure measures or structural models, but they should openly state which comparisons are potentially contaminated.

Measure displacement rather than treating it as a footnote

Displaced fishing effort can be described in vessel-days, trips, trawl-hours, hooks deployed, or another gear-specific exposure measure. These units are not interchangeable. A day of fishing with one net is not necessarily equivalent to a day with several nets. Suppose tracked vessels formerly used 120 trawl-hours per month inside a closed area and afterward recorded 75 extra trawl-hours in a neighbouring zone plus 20 extra hours in distant zones. The observed redistribution accounts for 95 hours, not 120. The remaining 25 hours may represent reduced effort, untracked activity, seasonality or altered vessel behaviour; it must not be invented as a flow to an unspecified destination.

A closure may change costs as well as effort. Moving farther from port can increase transit fuel and time, while moving to closer but less productive grounds can reduce catch per hour. Small vessels may have less flexibility than large vessels, and crews may lose income even when regional landings remain stable. If annual net revenue changes, distinguish gross sales from costs and distinguish vessel-level changes from community totals. A model that predicts stable total yield can coexist with losses for particular groups. Policy evaluation should make distributional consequences visible rather than treating one average welfare estimate as every fisher’s experience.

Suppose a vessel earns illustrative gross revenue of 40,000 currency units in a season before a closure, with operating costs of 25,000, leaving 15,000 net operating return. After changing grounds, gross revenue is 42,000 but costs rise to 31,000, leaving 11,000. The 5 percent increase in revenue conceals a 26.7 percent decline in the stated net return, calculated as (11,000 − 15,000)/15,000. This is a simple arithmetic example, not a forecast: it omits capital costs, taxes and uncertainty. It demonstrates why a claim about improved landings is not equivalent to improved economic outcomes.

Make uncertainty and sensitivity visible

A point estimate without uncertainty should be interpreted as a summary of data and assumptions, not a settled effect. Survey sampling variance, spatial correlation, variation among years, missing observations and model uncertainty can all matter. If an estimated BACI change is +2 fish per transect with an illustrative 95 percent interval from −1 to +5, the interval includes declines and increases. It would be inaccurate to present +2 as proof of a positive biological effect at a conventional threshold. Yet an interval spanning zero also does not prove no effect; it describes uncertainty under the method used to construct it.

Confidence intervals and prediction intervals answer different questions. A confidence interval may describe uncertainty about the average intervention contrast under a model; a prediction interval describes likely variation for an individual future site or period. Both depend on assumptions and coverage. A model can estimate a positive average effect yet predict wide site-to-site variation. If a study includes only highly productive reefs, transporting that average to low-productivity habitats introduces an additional uncertainty not necessarily captured in the original interval. Explain the target population before allowing a figure to circulate as an all-reserves conclusion.

A sensitivity analysis deliberately asks how the conclusion changes when a plausible assumption changes. Re-estimate the effect with alternative control sites, omit one influential year at a time, compare absolute and proportional scales, vary assumptions about missing stations, and examine whether results survive reasonable spatial buffer distances. These are diagnostic comparisons rather than a licence to select the largest number. Report the full set of prespecified checks, including negative or unstable estimates. A conclusion that changes sign under a small and credible choice of controls should be labelled fragile, not simplified into certainty for a headline.

Placebo tests are another useful discipline. Apply the same analysis at a pretend intervention date before the real closure, or to a species unlikely to respond through the hypothesised mechanism. A large placebo signal suggests that the design may be capturing background trends or measurement changes. A small placebo signal does not guarantee a valid causal effect because the placebo may be weak or poorly chosen. The task is to show what the design would falsely announce under conditions where it should find little. This is a more informative test than merely reporting a small p-value from the preferred specification.

Turn technical tables into honest language

The term impact is overloaded. In a management document it may mean the area subject to policy; in statistics it may mean an estimated treatment effect; in ordinary language it may imply a definite consequence. A translation should specify which meaning applies. Likewise reserve effect, abundance increase, redistribution of effort and fishery benefit are not synonyms. Prefer an operational sentence: Across the surveyed sites, the treated group’s mean biomass-density change exceeded the comparison group’s change by an estimated 8 kilograms per hectare, with uncertainty bounds and assumptions stated nearby. That carries more information than protection boosted fisheries by 8.

Translate measured outcomes without swapping their denominators. Catch per unit effort is not total catch, catchability is not abundance, and tonnes landed per season is not kilograms of standing biomass. An index can be valuable without being a census. If a table presents encounters per camera-hour, retain that exact unit; do not paraphrase as fish per square kilometre. If a spatial model predicts expected catch in a five-kilometre ring, state that it is modelled and that the ring is defined by distance to a boundary. The most readable technical writing is often the writing that makes its hidden measurement decisions visible.

A reader-friendly figure caption should specify the intervention, outcome, unit, times, sites and contrast. For example: Mean biomass density, kilograms per hectare, before and after a seasonal gear closure at three reefs, compared with three unclosed reefs using the same transect protocol; bars show uncertainty intervals clustered by reef. If the figure is a causal estimate, name the assumption; if merely descriptive, say so. The caption should not claim exported larvae, improved household income or higher total catch unless those were measured separately. Precision in language is an extension of precision in the study design.

Worked practice: distinguish a measured contrast from a claim

Exercise one uses invented abundance observations in fish per standard transect. Protected sites average 9 before and 15 after; comparison sites average 8 before and 11 after. Find the two changes, the BACI contrast and its unit. Answer: protected change is 15 − 9 = +6 fish per transect; control change is 11 − 8 = +3; the difference is +3 fish per transect. The result estimates an additional change associated with the intervention if comparison assumptions hold. It is not a 3 percent rise, not three tonnes of fish, and not a guaranteed causal effect in the absence of those assumptions.

Exercise two examines effort displacement. Before closure, 80 vessel-days were recorded inside an area. After closure, tracked vessels add 42 vessel-days to adjacent grounds and 18 vessel-days farther away. How much of the original activity has been accounted for, and what can be said about the rest? Answer: 42 + 18 = 60 vessel-days were observed elsewhere, equivalent to 75 percent of the former inside-area activity. The unaccounted 20 vessel-days cannot be labelled lost, illegal, or transferred without further evidence. Vessel-day comparisons also require consistent effort definitions and care about broader changes in total fleet activity.

Exercise three tests area weighting. Habitat A covers 30 hectares with biomass density 80 kilograms per hectare; habitat B covers 70 hectares with density 20 kilograms per hectare. Compute total biomass within these two strata and whole-area mean density. Answer: (30 × 80) + (70 × 20) = 2,400 + 1,400 = 3,800 kilograms. The total area is 100 hectares, so mean density is 38 kilograms per hectare. The unweighted average of stratum means is 50 kilograms per hectare and would be inappropriate for the combined area. This example also illustrates why a density estimate and a total are different reporting quantities.

Exercise four asks whether a control site is valid. A closure is introduced at a protected reef, but vessels shift fishing effort to the comparison reef. Both reefs use identical survey methods, and the control’s abundance declines. Can the difference-in-differences value be read directly as the closure’s isolated biological benefit? Answer: not without qualification. The comparison reef may have experienced an effect of the closure through displaced fishing, violating a no-interference assumption. Its decline can enlarge the estimated contrast even when the protected reef’s gain is smaller. Look for alternative controls, effort exposure data and a clearly defined estimand of the combined policy response.

Exercise five checks an economic headline. Gross revenue increases from 200,000 to 215,000 currency units while costs rise from 150,000 to 180,000. Calculate the change in net operating return. Answer: before, 200,000 − 150,000 = 50,000; after, 215,000 − 180,000 = 35,000; change −15,000, or −30 percent relative to the initial net return. The headline revenue increased is arithmetically true but financially incomplete. This example says nothing about real vessels. A real assessment needs compatible accounting periods, consistent cost categories, inflation treatment and uncertainty estimates before conclusions about fisher welfare.

Frequently asked questions about spatial closure evidence

Does BACI prove that marine protected areas work? No single acronym proves a policy effect. BACI is a useful design for comparing changes across treated and comparison sites, but causal interpretation depends on comparable untreated trends, measurement consistency, relevant sampling units and absence or modelling of interference. A carefully replicated design with pre-period checks is stronger than a one-site anecdote. A credible evaluation may find gains, losses, mixed effects or a result too imprecise to classify. Reading those distinctions is not a rejection of conservation; it is how we distinguish the intervention’s measured effects from expectations.

Why might fishing improve beyond a reserve boundary? Potential processes include adult movement, larval dispersal, habitat change and shifts in fishing distribution. Observed catch improvements could also arise from targeting behaviour or changes in detectability and stock availability. Demonstrating a mechanism means comparing observations to competing explanations, not merely noting proximity to a line on a map. The direction, size and timescale may differ by species and gear. A reserve designed to protect sedentary invertebrates will not necessarily affect highly migratory fish in the same way. Maps should therefore show the relevant ecology and fishing activity, not only political boundaries.

What should a non-specialist look for first? Identify the policy’s exact restriction and date, the outcome being measured, the units, the before and after periods, the control groups, the spatial sampling design and what happened to fishing effort. Then look for uncertainty intervals, tests of pre-intervention trends and sensitivity to alternative comparisons. If a report uses the word spillover, ask whether it means tagged adult movement, larval connectivity, catch patterns or model predictions. If a report uses benefit, ask who benefits and whether the benefit is ecological, economic or both. These questions make technical evidence substantially easier to translate without exaggeration.

How can two apparently contradictory studies both be valid? They may measure different species, reserve ages, enforcement conditions, survey outcomes, geographic scales or estimands. A study of large adult fish inside a well-enforced closure may find rising density while a study of regional fishing profit finds higher fuel costs from displaced effort. Both can be true. One might measure short-term catches and another long-term spawning potential. Before averaging conclusions, harmonise intervention definitions, denominators, methods and horizons. A disagreement over outcome definitions is not necessarily a disagreement about the same factual quantity.

Source and method notes

NOAA Fisheries: Marine Protected Area Economics Research.

Medoff, Lynham and Raynor, Spillover Benefits from the World’s Largest Fully Protected MPA, Science (2022).

Barceló, White, Botsford and Hastings, Projecting Timescale of Fishery Yield Following Marine Protection, ICES Journal of Marine Science (2021).

NOAA repository: Examining the Impacts of Spatial Closures in the Gulf Menhaden Fishery (2025).

All numerical examples above are invented for explanatory arithmetic and do not describe any actual protected area.

How to Translate Easily | How to Read Fisheries Spatial Closures, BACI Evidence, Displacement and Spillover