Are Anglers the Highway for Invasive Fish? A Negative-Control Test on 4,370 Angler Movements in Taiwan

A 2022 study of 4.9 million Fishbrain records argued that anglers moving between lakes create superhighways for aquatic invaders. Taiwan's invasive fish problem is denser than North America's, and we hold data of the same shape, so we asked the same question. But we added one step that study did not take: we applied the same connectivity metric to native fish. If it predicts natives just as well, it is not measuring dispersal.

Finding: The angler movement network is real and dense, but it predicts the appearance of native fish exactly as well as invasives. It maps where anglers go, not where fish are carried. On 21 months of data we find no evidence that angler movement drives invasive spread.

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

• H1 Connectivity: the more anglers arrive from waters already holding an invasive species, the more likely the destination holds it (controlling for angler-days). → holds, but see the next line.
• H2 Negative control: if connectivity reflects dispersal, it should predict non-invasive fish markedly less well. → rejected: equally strong for both.
• H3 Time order: among nodes clean in period one, period-one exposure predicts first appearance in period two, more so for invasives. → rejected: exposure adds nothing; no invasive-specific effect.
• H0: connectivity is a proxy for angler traffic; once effort is controlled it does not distinguish invasives from other species. → consistent with the data

Two consecutive freshwater catches by the same angler at different water bodies count as one move. 17,490 freshwater records across 3,251 nodes (cells within a named parent water body merged; 35% of records). Of 6,486 raw moves, 1,160 are under 2 km, mostly adjacent cells of the same river, and are not treated as inter-water-body.

We keep the 4,370 moves of at least 3 km within 60 days, from 1,272 anglers linking 1,988 nodes, median distance 15.3 km. The network is real and dense; that is not in dispute. What it represents is.

For each species, a node's exposure is the number of distinct anglers arriving there who had previously fished a node where that species was already recorded. We then fit, species by species, a logistic regression: presence ~ log angler-days + log exposure, for all 56 species with at least 25 nodes (11 monitored invasives, 45 others), and compare the distributions of exposure coefficients.

Species with at least 40 nodes; species names shown in Chinese as recorded. Bars are 95% confidence intervals; dashed lines are group means. Bold teal marks monitored invasives.

Mean exposure coefficient for invasives 0.50, for other species 0.58, Mann-Whitney p = 0.67. The two clouds overlap completely. Four of the five most connectivity-predictable species are not invasive (marble goby, walking catfish, convict cichlid, Japanese seabass). Striped snakehead, Taiwan's most widespread invasive, sits near zero because it is nearly everywhere already.

This paragraph is the whole study. A dispersal metric that works equally for natives and invasives is not measuring dispersal. It is measuring angler traffic: where crowds go, every species gets recorded, and the same crowds carry their records to the next place.

A cross-section can be explained by targeting: snakehead anglers already move among snakehead waters. So we split the data at 2026-01-01, kept nodes with no record of a species in period one, and asked whether period-one arrivals from occupied nodes predicted a first record in period two, controlling for period-two angler-days.

Raw "exposed vs not" rates favour exposed nodes for almost every species, natives included (Amur catfish 19% vs 5%, convict cichlid 20% vs 4%), because exposed nodes are crowded nodes. After controlling for angler-days, two of fourteen species reach p < 0.05: giant snakehead and the native Taiwan barbel minnow. Two in fourteen is chance.

Pooled into a species-by-node panel with species fixed effects: period-two angler-days β 0.81 (p < 0.0001), period-one exposure 0.14 (p = 0.34), invasive × exposure -0.12 (p = 0.53). What appears in period two is decided by who fishes there in period two. The connected-first-then-present signal is no stronger for invasives than for natives.

Rows, top to bottom: baseline (≥3 km, ≤60 days); strict distance (≥10 km); short gap (≤14 days); loose distance (≥1 km); exposure counted as arrivals rather than distinct anglers.

Tightening distance to 10 km, shortening the gap to 14 days, loosening to 1 km, or counting arrivals instead of anglers changes nothing. A reverse-time placebo (period-two exposure "predicting" period-one presence) gives +0.33 for invasives and -0.22 for others, small and inconsistent in sign, consistent with the metric capturing association rather than transmission.

Among nodes with a public name and at least 10 anglers, the twelve with the most outgoing anglers. They are hubs of angler traffic; if an invasive does appear somewhere new, these are where it would be detected first and carried onward fastest, so they remain the priority list for monitoring and outreach. Do not read them as "these waters are exporting invasives".

Measured

• 4,370 inter-water-body moves (≥3 km, ≤60 days), 1,272 anglers, 1,988 nodes.
• Exposure coefficients: invasives 0.50 (11 species), others 0.58 (45 species), p = 0.67.
• Temporal panel: angler-days β 0.81; exposure β 0.14 (p 0.34); invasive × exposure -0.12 (p 0.53).
• Invasives already present at 1,798 of 3,251 nodes (55%).

Inferred, to be read separately

• "No evidence" is not "anglers never move fish". European work attributes about 23.6% of invasive freshwater fish introductions over the past century to deliberate release for angling, which leaves no trace in catch records. What this study refutes is the inference from catch-network connectivity to dispersal.
• Twenty-one months is short. Spread on a scale of years is invisible here, and striped snakehead and tilapia have already compressed the clean-node space.
• "Appears" means "recorded", not "arrived". That is why angler-days dominate, and why the negative control cannot be skipped.
• Nodes are grid cells or parent water bodies; parents cover only 35% of records, so long moves along one river may still count as inter-water-body. Results hold at a 10 km threshold, but the error exists.
• The giant snakehead cell (p = 0.021, 13 new nodes) is worth following, but it co-occurs with a significant native, so we do not count it as evidence.

Weir et al. 2022 and Morreale et al. 2023 both infer invasion risk from angler movement or linkage networks, and neither runs a species-level negative control. This study does not say their networks are wrong; the networks are real. It says the step from connectivity to "dispersal pathway" needs a control group, and with that control in place the step does not hold in Taiwan's data. That applies to anyone building invasive-risk maps from app catch records.

• Weir, J.L., Vacura, K., Bagga, J., Berland, A., Hyder, K., Skov, C., Attby, J. & Venturelli, P.A. (2022). Big data from a popular app reveals that fishing creates superhighways for aquatic invaders. PNAS Nexus 1(3): pgac075. DOI4.9 million Fishbrain records across 130,000 lakes; infers invasion pathways from angler movement. We reuse the network construction and add a negative control.
• Morreale, S.J., Lauber, T.B. & Stedman, R.C. (2023). Anglers as potential vectors of aquatic invasive species: Linking inland water bodies in the Great Lakes region of the US. PLOS ONE 18(7): e0276028. DOISurvey-based reconstruction of angler links among Great Lakes inland waters to rank risk; also infers risk from connectivity, without a species-level negative control.
• Recreational angling as a pathway for invasive non-native species spread: awareness of biosecurity and the risk of long distance movement into Great Britain. Biological Invasions (2019). DOIAbout 23.6% of invasive freshwater fish introduced to Europe over the past century were released primarily for angling; the mechanism is deliberate release, not gear hitchhiking.

1. Nodes. Freshwater water bodies, merged to a named parent where one exists. Coordinates and names from the water-body registry; public pages show only named nodes, never grid cells.
2. Edges. Consecutive catches by the same angler at different nodes, great-circle distance ≥3 km, gap ≤60 days.
3. Exposure. For species s and node n: distinct anglers arriving at n whose origin node had a record of s dated before departure. Time order is strict.
4. Cross-section. Per-species logistic regression presence ~ log angler-days + log(1 + exposure), nodes with ≥2 angler-days; Mann-Whitney comparison of coefficient distributions.
5. Temporal split. Split at 2026-01-01; sample is nodes with no s in period one and any catch in period two; first record ~ log period-two angler-days + log(1 + period-one exposure). Pooled panel adds species fixed effects and an invasive interaction.
6. Robustness. Distance 1/3/10 km, gap 14/60 days, exposure as arrival counts; reverse-time placebo.

Read-only analysis; no production data modified. Node-level aggregates for replication are available on request, without coordinates or individual records.

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. Testing whether anglers actually move fish needs genetic data (source assignment of new populations) or direct measurement of release behaviour, not more catch records. If you work on invasive population genetics or conservation behaviour, we can provide de-identified node-level aggregates and the hub list to help design sampling. Write to support@dudefishing.net.