industry10 min read

WSOP Main Event Attendance Data: Methodology & Limitations (2018–2026)

Tom Sullivan·August 3, 2026

The data source

Every player count number in the article traces back to PokerNews' end-of-day chip count pages for the WSOP Main Event, 2017 through 2026. PokerNews is the event's live-reporting infrastructure: at the end of every Day 1 flight and every Day 2, it publishes a table of players — name, country, chip count, and in most cases a link to the player's PokerNews profile.

We captured those tables with a browser-based scraper that paginates each page and exports one CSV per flight — 52 source files across nine Series. Day 1 pages include players who busted during the day (listed at zero chips), which is what makes entry counting possible, not just survivor counting.

From pages to people

Raw pages list entries; the analysis needs people. Deduplication works in tiers:

  1. PokerNews profile URL — the primary key. A player's profile URL is stable across flights and across years. 48,364 of the 60,399 final records carry their own URL.
  2. Name + country resolved to a unique URL — for rows the scraper caught without a URL, if that exact normalized name and country maps to exactly one URL anywhere in the dataset, it inherits that identity (2,215 records).
  3. Name + country key — rows with no URL anywhere (9,813 records) are tracked by normalized name and country alone.
  4. Ambiguous names — a name+country combination that maps to more than one URL is never cross-year matched. Spot checks surfaced only a handful of these records, but that check isn't exhaustive, so we treat every ambiguous name as unmatchable — accepting the loss of a few players rather than risking false matches.

Net result: about 83% of all person-tracking runs on the URL key. Integrity checks across the 21,000+ unique URLs found zero cases of one URL carrying two different names within a year, and country changes across years are rare (about 0.1%).

The master dataset: 60,399 player-year records — year, name, country, entry day, and Day 1 outcome for every captured player.

Coverage: what we capture and what we miss

The chip count pages don't capture 100% of the field. Our captured counts run 77.5% to 84.7% of the official field in every analysis year, validated against WSOP's published field sizes. Two things about the misses:

  • The survivor lists are essentially complete. End-of-day bag counts are the pages' whole purpose; we validated them against official per-flight survivor numbers. What goes unlogged is a portion of same-day bust-outs.
  • 2018 is provably complete on entries: its Day 1 flight files sum exactly to the official 7,874.

Because misses concentrate among day-of bust-outs, we do not publish Day 1 survival rates — the captured data overstates them for everyone. Day 2 outcomes don't have this problem (see below).

Where the article reports estimated player counts rather than shares, captured counts are scaled to the official field size for that year ("coverage-adjusted").

External cross-check: WSOP published an official US/international registration split for 2026 (5,634 / 3,574, via Pokerfuse). Our independently derived estimate — 5,523 / 3,685 — lands within about 2% of it.

Definitions used in the article

  • Survivor census: players who entered on a Day 1 flight and bagged chips. Because survivor lists are near-complete, this is our cleanest apples-to-apples population for share calculations. US-share figures exclude Unknown-country rows from the denominator.
  • New player: no appearance in the prior four captured Series. We use a fixed four-year lookback so that "new" means the same thing in 2023 and 2026 — a necessity, because later years have more history available and would otherwise mechanically show fewer new players. (In this data, 88.6% of all returns happen within three years of a player's last appearance, so a four-year window catches nearly everyone.) Note the window counts captured Series: the 2020 online-only year is skipped, not counted as a gap year.
  • Returning player: the complement — appeared in at least one prior captured Series.
  • Cohort: all players whose first captured appearance was a given year, tracked at fixed checkpoints (returned within 1, 2, 3 years).
  • Late-reg: first appearance on a Day 2 page — the registration window extends into Day 2.
  • Day 2 outcome ("bagged Day 2"): appears on a Day 2 end-of-day page with chips greater than zero. About 34,400 player-records (2022–2026) were matched to a Day 2 outcome. Day 2 experience and nationality comparisons exclude Unknown-country rows, which can't be nationality-classified or matched to prior appearances.
  • Unknown-country handling: PokerNews lists some players without a country — 115 in 2018 growing to 698 in 2026. We resolved them in three tiers: cross-year name matching within the dataset (resolved ~41%; those players, mostly repeats, skew ~79% American), a Hendon Mob casher-list join (negligible yield), and a manual Hendon Mob sample of 2026's unresolved names under strict match rules (n=100, 70 matched, 58.6% American ±11.5) — statistically indistinguishable from the field at large. Conclusion: residual unknowns are approximately nationally representative, so aggregate US/international estimates redistribute them proportionally. Residual uncertainty on the US share is under one point.

Exclusions

  • 2017 is partial (thin legacy Day 1c data, roughly half the field captured). It appears on the decade trend line via official totals only and is excluded from all country, share, and retention analysis.
  • 2020 is excluded by design — the event ran online with split US/international brackets and isn't comparable.

Known limitations

Stated plainly, because they bound what the article claims:

  1. Return rates are floors. With ~80% annual capture, some true returns go unobserved. Capture rates are stable across years, so trends and between-group comparisons are robust even though every level is slightly understated.
  2. "New" means new since 2017. A player returning after a decade away counts as new. We consider that reasonable — a ten-year-lapsed player is functionally a new customer — but it's a definition, not a fact of nature.
  3. The 2021 cohort is slightly contaminated. Only three prior captured years existed to check it against, so some "2021 first-timers" were lapsed pre-2017 veterans. This inflates that cohort's measured loyalty; the article's load-bearing cohort comparison is therefore 2022 vs. 2023, which are measured almost identically.
  4. Day 2 survival measures outcomes, not skill. Bag rates are confounded by incoming stack sizes and by late-reg entrants starting short. The article says the international and veteran contingents outlast others — never that they outplay them — though reader judgment about what drives the gap is reasonable.
  5. International names are harder to match (transliteration variance across years), which most plausibly undercounts international returns. The measured international loyalty edge is, if anything, conservative.
  6. No satellite-qualifier flag exists in the data. The article's satellite-access section layers externally published qualifier counts (Pokerfuse, PokerNews, poker.org — see Sources below) onto our composition findings. The connection is correlation with a documented mechanism, not a traced causal chain, and the article frames it that way.
  7. Our source is one outlet's reporting. PokerNews' country attributions and name renderings are the ground truth available; systematic errors there would flow through. The coverage validation against official field sizes and the 2026 registration-split cross-check bound how large such errors could be.

Independent verification

Before publication, the entire pipeline was audited by an independent analysis session with access to the raw source files and a register of every claim — but not to our code or intermediate files. It rebuilt person identities, retention math, cohort curves, Day 2 outcomes, and late-reg profiles from scratch.

Outcome: the core findings reproduced independently, several to the exact digit. The audit caught one methodological inconsistency — an earlier full-window comparison had left the growing Unknown-country pool in the denominator, overstating the long-run American decline — and we corrected it before publication: American attendance over the full 2018→2026 window is essentially flat, and the American decline the article describes is specifically the fall from the 2023 peak (−15%). The audit's rebuild also matched slightly more player identities than our original pass; the article's cohort figures use those refreshed, better-matched numbers.

Sources

External sources cited in the article:

Data downloads

The aggregate tables behind every chart in the article:

The full player-level dataset (60,399 records) is available to researchers and media on request.

Questions about the methodology, or something that doesn't reproduce when you check it? Email [email protected] — we'll engage with it in the open.

Analysis and commentary based on publicly available tournament coverage, 2017–2026. Data processing and analysis assisted by Claude (Anthropic's Fable 5 model); all published figures re-derived in an independent audit pass. This is independent research from LiveHands, not affiliated with, endorsed by, or sponsored by the WSOP, NSUS Group, GG Poker, Caesars Entertainment, ESPN, PokerNews, or Poker.org.