Do Anglers Release Invasive Fish? Release Behaviour in 17,305 Taiwan Freshwater Catch Records

In Taiwan, removing invasive freshwater fish leans heavily on anglers, and agencies have started to enlist them: the Forestry and Nature Conservation Agency runs angling removal days in Yilan, the Fisheries Agency commissions pleco clean-ups, and Nantou County electrofishes juvenile giant snakehead at Sun Moon Lake. Yet a basic question has never had data behind it: when an ordinary angler catches an invasive fish, do they put it back? We answer it with 17,305 freshwater catch records.

Finding: Monitored invasive species are released 87% of the time, as often as or more often than other freshwater fish. Release is a default of fishing-method culture rather than a judgement about the species; only the media-salient giant snakehead and plecos are exceptions.

Analysis date 2026-09-05 · Data 2025-01-01 to 2026-09-04 · 中文版

• H1 Conservation-consistent: invasive species are released less often than other species. → rejected; the direction is reversed
• H2 Method culture: release rate is driven mainly by fishing method, not species. → supported
• H3 Within-angler: the same angler releases invasives and other species at different rates. → a difference exists but is small (+3.2 points) and favours release
• H4 Exploratory: invasives framed as pests by the media are released less than other invasives. → supported, but post hoc

Before asking what gets released, ask who releases. Overall freshwater release is 81.8%, but by method the spread is enormous:

Lure anglers release 86%, rock anglers 53%. Striped snakehead are 95% lure-caught and tilapia 65 to 73%, so the first-order explanation of high invasive release is simply that these fish are caught by lure anglers, and lure culture is release. That is a method effect, not a species effect. The question is what remains once method is controlled.

Bars are record-level release rates; hollow circles are one-vote-per-angler rates, to show whether a few heavy anglers dominate a species. Teal = monitored invasives, orange = media-salient invasives, grey = other freshwater species. Species with at least 100 records.

Most species cluster between 82% and 91% regardless of origin. Only two kinds drop out: food fish (barramundi 70%, hybrid grouper 71%, marble goby 65%) and media-salient invasives (giant snakehead 59%, sailfin pleco 50%). Striped snakehead and the three tilapias sit at 86 to 88%, indistinguishable from catfish, carp or topmouth culter.

Comparisons across anglers are confounded by who fishes for what. The clean test is within-angler pairing: take anglers with catches in both groups and compare their own release rates across the two. Identity, region and habit cancel out.

The 1,315 anglers who caught both monitored invasives and other fish released the invasives +3.2 points more often. The 188 who caught both media-salient and other invasives released the giant snakehead and plecos 19.8 points less. The sharpest pair: 50 anglers caught both giant and striped snakehead, releasing 63% of the former and 81% of the latter (p = 0.022). Same person, two snakeheads, different behaviour. The difference is not the fish's ecological role but its public image.

Logistic regression with angler and month fixed effects, controlling for method, fish length, ruler measurement and night. The sample is the 754 anglers with variation in release behaviour, 9,011 records. Odds ratios are relative to other freshwater species:

With everything controlled, monitored invasives still carry 1.32 times the odds of release; giant snakehead and plecos carry 0.07 times. A model without angler fixed effects but with angler-clustered standard errors points the same way (1.56 and 0.24). Two secondary findings: bigger fish are kept more (odds 0.71 per log unit of length), and anglers who measure with a ruler release more (1.88 times). Measuring is a behavioural marker of release culture.

The ordering is the same in every cut: monitored invasives ≥ other species > media-salient invasives. One caution on the sailfin pleco: three anglers account for 31% of its records, and one-vote-per-angler moves its rate from 50% to 65%. Giant snakehead has no such concentration (7%).

Measured

• Overall freshwater release 81.8%; monitored invasives 86.8%, other species 78.5%, giant snakehead and plecos 54.0%.
• Method is the largest single factor: fly 97%, lure 86%, rock 53%.
• Within angler: +3.2 points for monitored invasives; about 19 points less for the media-salient group.
• Release is rising: 73.9% in 2025 Q2 to 86.5% in 2026 Q2; the 110 anglers present in both periods moved from 78.3% to 89.5%, so this is not a change in who uses the app.
• The release option defaults to unchecked; the three release fields agree 99.997%; the app has never labelled any species invasive.

Inferred, to be read separately

• "Media-salient" is a post hoc grouping. Striped snakehead also has official removal programmes yet is treated as a sport fish; the grouping captures public image rather than management status. That is interpretation, not measurement.
• "Released" is self-reported. We do not know whether released fish survive, nor whether unreleased fish were eaten, discarded or humanely dispatched.
• Invasive status follows the site's monitoring list. Exotic species off the list (convict cichlid 91%, redhead cichlid 82%) all sit in the high-release band; moving them into the invasive group would not change the direction.
• The sample is app users, predominantly lure anglers, not all Taiwanese anglers; the study is underpowered for invasive-release behaviour among rock and bottom anglers.

Official and volunteer removal efforts assume anglers are allies. The data says: anglers do kill giant snakehead, but they release striped snakehead and tilapia, and the latter vastly outnumber the former in both abundance and range (striped snakehead alone appears at 1,087 water bodies from 1,092 anglers). The difference is not indifference; the information has only reached one species. That is directly actionable: a single sentence at the moment an angler logs an invasive catch costs almost nothing. Our own app is that channel, and this study is deliberately published before any such intervention so that it can serve as the baseline.

Verified public sources

• Nantou County Government press release on giant snakehead control at Sun Moon Lake (electrofishing juveniles; buy-back judged impractical) source
• Liberty Times: giant snakehead removal at Sun Moon Lake source
• Taiwan Ocean Conservation and Fisheries Sustainability Foundation: pleco removal operation in Taipei commissioned by the Fisheries Agency source
• China Times 2026-04-13: anglers remove invasive fish, mostly striped snakehead, at Luodong Forestry Culture Park source
• Taiwan Environmental Information Center: angler-led invasive fish removal source

Related literature

• Culture, context, and fish length drives voluntary catch-and-release behaviour of recreational anglers. Reviews in Fish Biology and Fisheries (2025). DOITwo German fisheries, 19,558 trip-level records: voluntary release is driven mainly by angling culture, context and fish length. The structure matches ours, with method culture first and fish length second.

1. Data. Catch records joined to water-body classification; freshwater, from 2025-01-01, non-fish excluded, release field present: 17,305 records. Invasive status from the site's 12-species monitoring list.
2. Field reliability. Release is an opt-in checkbox defaulting to unchecked; three synonymous fields agree 99.997%; monthly release rates show no break that would indicate a default change.
3. Within-angler pairing. Per-angler release rate in each group, differenced; 5,000 angler-level bootstrap replicates for the CI; Wilcoxon signed-rank test.
4. Regression. Logistic regression with angler and month fixed effects, controlling for method (lure vs other), log length, length availability, ruler measurement and night. Restricted to anglers with variation in release. A no-fixed-effects model with angler-clustered standard errors as a check.
5. Robustness. Excluding the top 1% of anglers, EXIF timestamps only, 2026 only, length ≥30 cm, lure only, and one vote per angler.

The analysis was read-only and modified no production data. Aggregate tables for replication are available on request.

The aggregates, figures and methods here are free to read and cite; please retain the data period, sample size and analysis date. Public pages are free to read and cite page by page. Raw records are not available for download, and bulk reuse licensing for aggregated data is still under review.

We are looking for collaborators. If you work on invasive species management, conservation communication or recreational-fisheries behaviour, we can share de-identified aggregates and would co-design a randomised trial of a one-line prompt at the moment of logging, turning this baseline into a measured intervention effect. Write to support@dudefishing.net.