The Empty Column in a Table Tennis Scorecard: Why 250 Points Proves Nothing

At 10:40 p.m., the statistics file for a national-league table tennis match opens at 4,180 rows. The column I need most — points won on serve — is blank. The file is not corrupted. The recorder logged set scores only, because set scores were the only thing requested. I am used to opening a football match and holding 92 pages of numbers: pass counts, touch maps, pressing indices. For table tennis I hold one A5 sheet and seven scores. People ask me whether women watch football. I once answered with 92 pages of data. For table tennis, I have no pages to offer. Table tennis carries the densest decision load in ball sports. A match runs at most seven games of eleven points, so the total rarely exceeds 90 points. Each point lasts three to five seconds of live ball. Inside that window there is one serve, usually one return, and occasionally one exchange. The entire match fits into under ten minutes of actual play. The statistical consequence is plain: every point is a near-independent mini-match, yet the number of observations inside a single match is tiny. Since 2026, the WTT system run by the ITTF has published point-level statistics at major events. Viewers can look up serve-point win rates, win rates in rallies of five shots or more, and performance at deciding points for players such as Fan Zhendong, Wang Chuqin, Sun Yingsha and Tomokazu Harimoto. The ITTF ranking operates on a rolling 52-week window, so every result has an expiry date and every point has a price. At Vietnam's national level, that data layer barely exists. Spectators receive scores. They do not receive the process that produced the scores. Match reports after each round therefore circle around lines like “this player is finding form” or “this opponent has an awkward serve”, with nothing to verify them beyond the writer's memory. Three indicators are what I want before any report. Serve-point win rate. It does not measure a pretty serve. It measures the ability to convert a serve into an advantage on the third ball — the attack immediately after the opponent's return. A player with heavy spin but no third ball will score lower here than a player with an ordinary serve who takes initiative. Win rate in rallies of five shots or more. This is the stamina-and-control indicator, the thing that separates a player who can sustain long matches from one who lives on the first three shots. Performance at 9-9 and 10-10. It measures nerve, and it is also the most inflated indicator of the lot. This is where I want to pause longer than usual, because it is the part few reports mention. A single table tennis match generates only about 40 to 50 service points. At that sample size, a player winning 68% of service points has a 95% confidence interval stretching from 54% to 82%. In other words, after one match we cannot distinguish a genuinely strong server from an average server on a lucky night. Pool three matches and service points reach roughly 120 to 140. The interval narrows to 60% to 76%. Still far too wide to declare anyone dominant on serve. A figure solid enough to trust needs about 400 service points, close to ten matches — a stretch over which form, injury and opposition will all have shifted at least once. That is why I avoid lines like “this player has the best serve in the tournament”. I can write: over the last three matches, this player's serve-point win rate is 68%, on a sample of 128 points. The second sentence is longer, uglier, and more accurate. The common industry belief is that data equals truth. Table tennis shows the reverse face of that belief. Players with high serve-point win rates do win matches. Most of that relationship runs through a third variable: overall technique. A good server is usually also good at controlling tempo, choosing position, reading spin. The serve is not an independent cause of victory; it is one expression of the same underlying ability. Reading a correlational indicator and concluding causation is the most common error in sports analysis, and table tennis is the easiest place to make it, because every indicator rests on a small sample. There is also a selection problem. Point-level data exists only for televised matches. Those matches usually feature strong players, deep in the draw, against opponents they are favoured to beat. The dataset we hold is therefore filtered in exactly the direction that makes every player look better than reality in the hard phases. In 2026, when European football paused, I compared 120 matches with crowds against 98 without and found completed passes up 7.3%. The lesson I carried into table tennis sits elsewhere: every indicator depends on measurement conditions, and measurement conditions are usually hidden from the reader. In 2026, when I wrote that Germany generated only 0.48 xG against South Korea, a group of readers called it a woman guessing. I did not argue. I posted the 40-page raw dataset alongside. The lesson repeats: people do not object to data, they object to having to read it. More dangerous than a wrong table is a beautiful one. An empty table forces me back to the video. A full table stops me going back at all. Data does not lie. Only the reader has not been honest enough. Progress does not arrive with analytics software. It arrives with recording every point: who served, who won, how many shots the rally lasted. A national league that logs that for two seasons will own a foundation money cannot buy. There are evenings I sit with numbers longer than with people, and I have never felt alone. If someone published the serve-point win rate for the entire league tomorrow, would we have the courage to print the sample size beside it?