Research Methodology: How the Angling-App Catch Studies Work, Their Data Boundaries and Where They Stand Now

Catch logs from a fishing app are plentiful and fine-grained in time and place, but they only record fish that were caught. We use them to test angling lore, such as how moon phase, air pressure, dawn and dusk or typhoon warnings relate to logged catches.

After the data integrity rules were updated, every study was recomputed on a single read-only point-in-time snapshot. For most studies the earlier conclusion no longer has enough support. That does not show the lore is wrong; the current data cannot yet decide, so those studies are marked inconclusive, and each page states what was measured and what is still missing. This page writes the shared approach down in one place so that anyone citing a study first knows the limits of the data.

Version 2026-09-30 · Data 2025-01-01 to 2026-09-25 · 中文版

The data are catches voluntarily logged by DudeFishinG app users. Each record carries a species (AI identification or the angler's choice), a time, a place (mapped to a water body classed as freshwater or marine), an optional length, whether the fish was released, the method, and the weather, air pressure and tide attached at logging time from the nearest station or forecast.

Studies count only records that pass the current eligibility rules: the record is usable for statistics, the species identity is unambiguous, and the capture time and place can be verified. Manually chosen times or places are left out of any analysis that uses time of day, date, place or water body. Each study has further exclusions (for example shrimp fishing, managed ponds and non-fish records) described in its method section, so samples differ between studies and are smaller than the site's full set of records.

What is missing matters as much: there are no blank trips, no trip start, end or duration, and no bait use. The data therefore have no catch per unit effort and cannot directly compare which time or condition yields more catches. This page and the studies publish only de-identified aggregates and methods, never individual records, accounts or coordinates.

• Inconclusive: the current data cannot support a conclusion and the earlier one is withdrawn; the study page lists what was measured and what data would be needed.
• Question reframed: once place is controlled for, the original question of whether gear selects fish becomes one about how method, place and species relate; the result is an observational association.

The numbers on each study page are frozen at the analysis date. At the start of each month we re-run the headline test of the studies still on the re-test list with newly accumulated data, to answer one question: with the new data, does the conclusion still hold? The rules are fixed in code in advance, and the re-test outputs only "holds" or "under review"; under review triggers a human re-analysis, and the text changes only if needed, with a note on the page. The monthly run uses fewer permutations, so the published numbers on each study page remain the citable ones; the re-test code and rules are in the repository.

Retired from 2026-09-30: 7 studies. Studies whose conclusions were withdrawn and marked inconclusive are no longer re-tested, and their re-test results are no longer shown: re-testing a withdrawn conclusion would treat it as if it still held.

The study still on the re-test list has been recomputed under the current rules, so its sample differs from the re-test baseline. Until the baseline is rebuilt on the new sample, no re-test result is listed here, so an old baseline is never shown beside new figures.

1. Write the question down before looking. Each study states what it compares and against which control before the analysis; a re-check keeps the original question and never switches questions to reach a conclusion.
2. Eligibility before analysis. Only records that pass the current record, species identity and time-and-place checks are used; analyses of time of day, date, place or water body leave out manually chosen times and places.
3. One read-only point-in-time snapshot. Each recomputation reads the same read-only snapshot, whose read time, counts and export program are archived; analyses never write production data.
4. No effort denominator, no claims about better fishing. Anglers do not log blank trips, so more records under a condition may only mean more people went out. The daily number of anglers logging a catch reflects both going out and catching, and cannot stand in for effort on its own.
5. Compare only with a control. Freshwater against marine, or comparisons within the same angler or water body, let the control carry the same bias; when the control cannot be rebuilt, no comparison is made.
6. Say "inconclusive" when the data fall short. Old figures are neither reused nor silently cut: when a conclusion is withdrawn, the page states what the earlier version said, what this version measured, why no decision is possible yet, and what data would be needed.
7. Measured and inferred kept apart. Pages list measured figures first and interpretation and limits after; readers may trust only the measured part.
8. Every figure recomputed independently. Page figures are computed from the snapshot by one program and recomputed line by line from the same snapshot by an independent program; any altered line is blocked.
9. Read-only analysis, aggregates only. Analyses never modify production data; what is published is aggregates and methods, not raw records or coordinates.
10. No silent changes. Every version has an analysis date and data period; revisions and withdrawals are noted on the page and recorded in the repository.
11. Re-test only conclusions that stand. Each month the headline test of conclusions that still stand is re-run on new data; withdrawn conclusions are no longer re-tested, so they are never treated as if they held.

These are not hidden weaknesses; they are where each study's design and re-check start. Read this section before citing any number.

• Only catches are logged. Blank trips are invisible, so there is no catch per unit effort and no reliable measure of effort.
• Manually chosen times and places are not necessarily when and where the fish was caught, for example fished by day and logged at night. Analyses of time of day, date, place or water body count only records whose capture time and location can be verified, so their samples are much smaller. (related study: Do logged catches differ between dawn, dusk and midday?)
• Most lengths are entered by anglers and only some are measured with a ruler; the region and length type of external length-at-maturity values have not been checked against the site's records. (related study: How many caught fish are below maturity?)
• Freshwater or marine is assigned from the water-body classification; misclassification shrinks the contrast between the groups.
• The number of users changes a lot over time, so comparisons across time cannot use a whole-period average; the daily number of anglers logging a catch also mixes going out with catching. (related study: How do logged catches change during typhoon warnings?)
• Length, release flag and photo are all optional; records that carry them lean toward fish the angler thought worth recording.
• Weather, air pressure and wave height attached to a catch come from the nearest station or forecast, not on-site measurement; verified pressure records cover only a few days. (related study: Do logged catches differ around a fall in air pressure?)
• The current snapshot has no non-native classification, so non-native versus native comparisons cannot be rebuilt yet. (related study: Do anglers keep non-native fish rather than release them?)

The "cannot answer" column is reviewed as data accumulate; when a question can be answered, a new version is released under a new analysis date and noted on this page.

1. Choose a question. Prefer questions that anglers actually debate and for which the data allow a control.
2. Write the question down. Decide what is compared and against which control before looking at the numbers.
3. Read-only snapshot. Read a single point-in-time snapshot from the production database, recording read time and counts, and write nothing back.
4. Apply eligibility. Count only records that pass the current data integrity rules.
5. Compute and recompute independently. One program produces the page content; a second, independent program rebuilds it line by line from the same snapshot and compares.
6. Decide. Write a conclusion only when the data can support one; otherwise mark it inconclusive and list what was measured and what is needed.
7. Publish and archive. Both language pages, the citation files and the dataset catalog come from the same content; evidence, code and change records are kept in the repository, and every page is checked after going live.

Every study page has citation files. Please keep the data period, sample size and analysis date when citing, and state whether the conclusion is inconclusive or a reframed question. This page is citable too: BibTeX · CSL-JSON · dataset catalog. 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.

What we can provide: Each study's analysis code and eligibility rules; the aggregate tables named in each study's method section; joint recomputation with your external data, with co-authorship.

What we most want: Length at maturity for local Taiwanese populations; fishing diaries that record effort and blank trips; catch records from other regions or communities whose time and place can be verified. Write to support@dudefishing.net.