Does the Moon Affect Fishing Success? A Null Result from 33,704 Taiwan Angler Records

Solunar theory, in circulation since 1926, holds that lunar phase governs when fish feed. It has never been rigorously tested on recreational shore angling, and never in Taiwan. This dataset can answer it, with one structural advantage: the same anglers fish both freshwater and marine sites. Freshwater has no tide, which makes it a natural control group.

Finding: No lunar or spring-tide cycle is detectable. If an effect exists, per-trip catchability varies by less than 1.15x peak-to-trough, far below the doubling that solunar folklore claims.

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

Finding patterns after the fact is easy. We wrote down four falsifiable predictions first:

• H1 Solunar: catches peak near new and full moon (spring tides), producing a significant semi-lunar amplitude. → not supported
• H2 Tidal mechanism: the semi-lunar signal is stronger in marine than freshwater, and stronger on the west coast than the east. → not supported
• H3 Moonlight mechanism: night catches show a 29.5-day cycle, present in freshwater too. → not supported
• H0 Null: after controlling for day-of-week and month, no lunar structure remains. → consistent with the data

Each point is the mean catch count for that lunar day across 606 days, after removing day-of-week and month effects. Weekends already carry 1.86x the volume, which would otherwise swamp everything. The grey band is the amplitude range that circular-shift permutations produce by chance, a peak-to-trough ratio of about 1.13. Both lines wander inside it and neither tracks the new or full moon.

This is the standard objection to app-collected angler data and it deserves a direct answer. We see only trips that produced a catch. We estimate 46.7% of trips leave no record at all: of roughly 39,688 inferred trips, only 21,152 angler-days produced a logged catch.

But catches equal effort multiplied by catchability, so we asked the question by simulation: if true catchability had a lunar cycle, would we see it? We held daily effort at the observed pattern, imposed a cycle on catchability alone, and ran the same test to measure the detection rate.

We would see it. A 1.15x peak-to-trough cycle in catchability, meaning the best lunar day yields 15% more per trip than the worst, is detected 95.0% of the time; 1.2x and above is caught every time. The observed amplitude is only 0.0100, well under the 0.0617 threshold.

The one configuration that could conceal an effect is effort moving exactly opposite to catchability, with anglers fishing less precisely when fish bite more. Solunar believers do the opposite, so the bias amplifies rather than cancels. Even under perfect cancellation, a 2x effect is still detected every time.

The simulation's day-to-day variance (SD 0.478) exceeds the real series (0.423), so these power figures are conservative rather than optimistic.

A null result is only worth anything if its exclusion range is stated, and the literature makes that hard: some studies report gain relative to the mean, others peak-to-trough, others only variance explained. Mixing the scales badly distorts magnitudes. The widely quoted "+28% on a night full moon" is relative to the mean; as a peak-to-trough ratio it is 1.79x. The table below puts everything on one scale before comparing it with our detection floors:

The answer splits cleanly in two. Large effects we can exclude. The night full-moon ratio of 1.79x sits far above our night detection floor of 1.28x; an effect that size in Taiwan shore angling would have shown up, and the observed night amplitude is only 0.0203 against a 0.0589 threshold. The same goes for the 1.55x spring-tide effect and the 2x folklore claim.

Small effects we cannot exclude, and this has to be said plainly. The best effort-controlled recreational study to date (Stevenson & Millar 2013, New Zealand snapper, over 82,000 trips) estimates a peak-to-trough ratio of 1.137x, essentially sitting on our 1.15x detection floor. So the correct conclusion is not "the moon does nothing". It is that no lunar effect large enough to plan a fishing trip around exists in Taiwan shore angling. The kind of small effect that is statistically detectable but which its own authors call practically irrelevant is something this study can neither confirm nor refute.

The counter-example belongs here too. The geographically closest comparable study is at Yonaguni Island, about 110 km from Taiwan, where daytime trolling catch rates rose significantly around the full moon with supporting stomach-content evidence. That is boat fishing on a different species, but it is a reminder that this null is bounded by the gear and waters our data covers.

Marine-at-night alone holds only 3,070 records, pushing its detection floor to 1.50x. That cell is underpowered and we draw no conclusion from it.

An uncommon finding worth flagging. The dominant confounder in this literature is that anglers schedule trips around the lunar phase they believe in. Vinson & Angradi measured a lunar effect on effort itself of up to 36% at one lake, leaving catch and effort inseparable. Our logging-angler series has no lunar cycle either (amplitude 0.016, p = 0.80), so that confounder is absent from our data. That is both a methodological clean bill of health and an independent behavioural observation: these app users do not, in practice, plan trips by the moon.

References

• Vinson, M.R. & Angradi, T.R. (2014). Muskie Lunacy: Does the Lunar Cycle Influence Angler Catch of Muskellunge (Esox masquinongy)? PLOS ONE 9(5): e98046. DOI341,959 angler reports using near-identical methods; the authors' own creel cross-check found the lunar cycle did not explain significant variation once catch per unit effort was used.
• Quigley, C.N., Gonzalez Murcia, J.D. & Kauwe, J.S.K. (2023). SN Applied Sciences 5. DOIThe only peer-reviewed test of commercial solunar tables; null, but based on only 361 trips.
• Stevenson, B.C. & Millar, R.B. (2013). Promising the moon? Environmental and Ecological Statistics 20(4): 591–608. DOINew Zealand snapper with effort controlled; the lunar effect was statistically significant but the authors judged its magnitude too small to matter in practice.
• Pepperell, J.G., Smart, J., Wolfe, B.W., Williams, S.M. & Tracey, S.R. (2025). Abundance indices for black marlin derived from 50 years of diarized daily catch-effort data. ICES Journal of Marine Science 82(10): fsaf182. DOI54 years and 11,613 fishing days; catches 30-40% lower at new and full moon, which the authors attribute to spring-tide turbidity rather than moonlight.
• Shimose, T., Yokawa, K. & Tachihara, K. (2013). Higher Catch Rates Around the Full Moon for Blue Marlin in a Diurnal Trolling Fishery. Bulletin of Marine Science 89(3): 759-765. DOIYonaguni Island, about 110 km from Taiwan and the closest comparable study; daytime trolling catch rates rose significantly around the full moon, supported by stomach-content data. It is a counter-example to our finding and we list it as such.
• Kuparinen, A., Klefoth, T. & Arlinghaus, R. (2010). Fisheries Research 105(2): 111–117. DOIThe strongest counter-evidence: a lunar effect in pike that survived with true catch per unit effort.

A method with no power returns zero for everything. The day-of-week effect is an uncontroversial real signal, and the same residualisation and harmonic fit recovers it cleanly:

Saturday 1.45x, Sunday 1.54x, weekends 1.86x weekdays. The lunar null is not a blind instrument.

Amplitudes are in log units; peak-to-trough ratio equals exp(2A). Amplitudes below 0.06 cannot be distinguished from noise. P-values come from one-sided circular-shift permutation tests.

Measured

• Lunar amplitude: 0.010 overall, 0.031 marine, 0.015 freshwater; all p greater than 0.6.
• Semi-lunar amplitude: 0.027 overall, 0.015 marine, 0.038 freshwater; p from 0.20 to 0.63.
• Logging-angler counts show no lunar cycle either (amplitude 0.016, p = 0.80), which rules out the "more fish bite, more anglers go, effects cancel" explanation.
• Fish length, which is independent of effort, standardised within 121 species over 23,380 records: lunar amplitude 1.0% overall and 2.2% marine (p = 0.22). At most 1 to 2%.
• Rising tide 52.2% versus falling 47.8% (marine, EXIF timestamps, 3,186 records, p = 0.015); concentrated in estuaries at 53.1%, with open coast flat at 50.7%.

Inferred, to be read separately

• The two-point rising-tide gap cannot separate fish behaviour from angler behaviour. Anglers widely believe rising tides fish better and schedule accordingly; we count catch records, not catch per unit effort.
• That the signal appears only in estuaries may reflect asymmetric rise and fall durations at river mouths rather than fish behaviour.
• The null applies to recorded catches, which is not identical to fish feeding activity. The power analysis narrows that gap substantially but does not close it.
• Our west-versus-east tidal-range proxy (interquantile spread of stored tide heights) failed to separate the coasts, so that natural experiment reduces to geographic grouping.

1. Data. 33,704 catch records joined to water-environment classifications for 9,493 water bodies. Filtering to records after 2025-01-01 with a lunar age and excluding non-fish leaves 30,216: 12,540 marine, 17,399 freshwater, 277 unknown.
2. Lunar age. Computed independently at local noon from a 2025-01-29 new-moon epoch; median absolute difference from the stored field is 0.57 days.
3. Model. Daily log(catches + 1) residualised on day-of-week (6 dummies) and month fixed effects (20), then regressed on cos θ, sin θ, cos 2θ, sin 2θ where θ is the lunar phase angle. Logging-angler counts and catches-per-angler follow the same specification.
4. Significance. Circular-shift permutation: the lunar series is rotated by k days (15 ≤ k ≤ N−15), preserving autocorrelation and seasonal structure while breaking the true phase correspondence. 3,000 permutations (1,000 to 1,500 for smaller subsets).
5. Timestamp provenance. 38.9% of capture times come from photo EXIF. Lunar cycles are insensitive to errors of a few hours; the tidal-phase analysis uses EXIF-timed records only.

The analysis was read-only and modified no production data.

• No direct effort measure: trip duration and blank trips are unobserved. Simulation substitutes for measurement but does not replace it.
• The sample is app users, not a random sample of Taiwanese anglers; the most active 1% contribute about 18% of records.
• Tide-height occupancy requires a tidal model to build a null distribution, so it is described but not tested here.
• No single-species analysis: most species have only a few hundred records, which is underpowered.
• Hypotheses were written before testing but within one analysis session, without public pre-registration. Out-of-sample replication remains outstanding.

The aggregates, figures and methods on this page are free to read and cite. Please retain the data period, sample size and analysis date shown here. 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. The clearest gap in this study is the absence of blank-trip records. If you work in fisheries science, ecology or applied statistics, we can provide de-identified daily aggregates, co-design a pre-registered replication, or add trip-level effort logging to the app so that genuine catch-per-unit-effort becomes available. Write to support@dudefishing.net.