Method, Place and Species: 11,364 Taiwan Catches

Method and recorded species can be associated through gear, chosen water bodies and microhabitats within them. We compare information before and after conditioning on water body, then examine records from the same anglers. These observational data cannot identify a gear effect on their own.

Finding: About 21% of the method–species association remains after controlling for water body; the difference cannot be assigned solely to location choice. Pole-caught fish are shorter than lure-caught fish in this sample, but the data do not establish causality.

Analysis date 2026-09-26 · Data 2025-01-01 to 2026-09-25 · 中文版

• H1 Within-water-body association: method still carries information about recorded species after conditioning on water body. → signal detected; 21% of marginal information remains
• H2 Within-angler association: species composition differs across the same angler’s method records. → signal detected; time and target species may still confound it
• Original H3 layer mechanism: field review found this is the angler-recorded fishing layer. → the original mechanism cannot be tested; only descriptive statistics appear below
• H4 Length association: compare methods within species, water body and month. → pole versus lure is negative in three specifications
• H0: within-node method shuffling produces the same information. → rejected; this is not causal identification

We ask an information question: how much does knowing the method reduce uncertainty about the species caught? Species uncertainty is 5.47 bits. Knowing method alone removes 0.40 bits (7%). If the water body is already known, method removes only a further 0.09 bits (2%), still significantly above a null that shuffles methods within each water body (p < 0.01).

Of the method’s marginal information, 21% remains after conditioning on water body. The difference can reflect location choice and other confounding; it cannot be assigned to gear as a causal share. Water body alone carries 3.44 bits.

Within-angler comparison. Among 433 anglers who used two or more methods (4,358 records), bias-corrected conditional mutual information is 0.13 bits (p < 0.001). The same person may still select a method by time, microhabitat or target species.

For each method and species, expected counts sum the method’s catches in each water body times that body’s species share. The index compares observed with expected co-occurrence. An index of 2.0 means twice the local baseline in these records. Cells require ≥10 observed and ≥20 expected records. This is neither catch rate nor a causal gear effect.

Species names shown in Chinese as recorded.

The table shows descriptive high and low co-occurrence ratios in this eligible sample. Method, bait, microhabitat and target choice may move together, so the ratios alone cannot identify a gear preference.

This field records the fishing layer chosen by the angler: surface, mid-water or bottom. It is not a constant biological trait of the species. We compare its distribution by method within water bodies as description only, not as evidence for a selectivity mechanism.

Bottom-rig records choose the bottom layer 94% of the time, versus a within-water-body expectation of 84%. This reflects angler choice and recording; it does not imply that the caught species live at the bottom.

Within species, water body and month, how does logged catch length differ by method? Log length is regressed on method with species, water-body and month fixed effects absorbed and angler-clustered standard errors, using 8,890 eligible length records. Relative to lure:

* p < 0.05. The angler fixed-effects column uses only anglers who fished two or more methods.

Pole versus lure length estimate: -20%; ruler-only -15%; angler fixed effects -11%. All three p values are below 0.05. Shore casting and boat fishing did not meet the main-analysis threshold. Among exploratory species contrasts, 7/15 have unadjusted p < 0.05; median absolute effect is 22%. No multiple-testing adjustment was applied.

Measured

• Mutual information of method with species: 0.40 bits marginal, 0.09 bits conditional on water body (21% retained); water body alone 3.44 bits.
• Within-angler conditional mutual information 0.13 bits (p < 0.001).
• Length: pole -20.2% versus lure, p < 0.05 in three specifications. Exploratory species contrasts: 7/15 unadjusted p < 0.05.
• Bottom-rig bottom-layer share 94% against expected 84%.

Inferred, to be read separately

• Water-body nodes still contain different microhabitats, times and targets. The 21% retained association is neither a causal gear share nor an upper bound on one.
• Size results describe fish that were caught and logged, not fish that encountered the gear; pole anglers might log small fish more or less readily. The ruler-only and within-angler results agree, which softens but does not remove the concern.
• Fishing layer is selected and recorded by the angler. It is not a species trait or independent evidence for a selectivity mechanism.
• Without effort data we cannot say which method is more efficient; this study is about composition and size, not catch rate.

• Alós, J., Arlinghaus, R., Palmer, M., March, D. & Álvarez, I. (2009). The influence of type of natural bait on fish catches and hooking location in a mixed-species marine recreational fishery, with implications for management. Fisheries Research 97: 270–277. sourceMediterranean mixed-species recreational fishery: bait type shifts catch composition, size and hooking location; regulating bait is a simple compositional lever.
• Weltersbach, M.S., Strehlow, H.V., Ferter, K., Klefoth, T., de Graaf, M. & Dorow, M. (2019). Effect of lure and bait type on catch, size, hooking location, injury and bycatch in the western Baltic Sea recreational cod fishery. Fisheries Research 210: 121–130. sourceBaltic cod: length, legal-size harvest rate, injury and bycatch differ; cod caught per hour did not differ significantly.
• Tsuboi, J., Morita, K., Sahashi, G., Kuroki, M., Baba, S. & Arlinghaus, R. (2021). Species-specific vulnerability to angling and its size-selectivity in sympatric stream salmonids. Canadian Journal of Fisheries and Aquatic Sciences 78(10): 1470–1478. DOIAngling vulnerability and size-selectivity differ among co-occurring stream salmonids; selectivity must be read per species, not per gear alone.

1. Data. App catches captured and uploaded from 2025-01-01 to 2026-09-25 pass current source, moderation, time and water-body context rules and have an explicit resolvable species ID. Unresolved identity only excludes a record from species-composition estimates; it does not declare that catch invalid. Five methods meet the predeclared ≥300-record threshold, followed by species with ≥30 records. Shore casting and boat fishing fall below the method threshold. Final cohort: 11,364 records, 74 species and 2,837 water-body nodes.
2. Information. We compute I(species; method), I(species; water body), I(species; method | water body) and within-angler comparisons. Bias correction uses the mean of 1,999 permutations shuffling method within water body or angler. Permutation p is (1 + extreme permutations) / 2,000.
3. Selectivity index. Expected = Σ over water bodies of (method's catch count there × the water body's species share); index = (observed + 0.5) / (expected + 0.5).
4. Size. Log length regressed on method dummies with species, water-body and month fixed effects absorbed by alternating demeaning; angler-clustered robust standard errors. Ruler-only and angler-fixed-effects (angler additionally absorbed) as robustness. Per-species models absorb water body within species.

Read-only analysis; no production data modified. 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 with attribution and a link to the original page. Raw records are not available for download; bulk copying or redistribution of the dataset is not licensed. See the data use terms.

We are looking for collaborators. Estimating a causal gear effect would require within-water-body microhabitat, fishing effort or a controlled experiment. Researchers in gear selectivity or recreational fisheries can discuss use of de-identified aggregates with us. Write to support@dudefishing.net.