Winning the Match, Losing the Point Total: Where Table Tennis Lacks Data

Place a perfectly legal scoreline on the table: 11-9, 2-11, 11-9, 2-11, 11-9, 2-11, 11-9. The left side wins four games, the right side wins three. Added up, the winner scored 50 points, the loser scored 69.

I built this example by hand. It is not taken from any match. It is the extreme version of something that happens routinely in every table tennis event, from the Vietnamese national championship to a WTT Grand Smash: a game needs only a two-point margin, so the aggregate point total barely tells you who played better.

On my machine there is a file called blank_report.xlsx. It is the output of an analysis session whose input came back empty: no event name, no player, not a single data column. The only populated field was the sport label — table_tennis. I keep it as a reminder.

After seven years of watching footage and building indices, I have reached an uncomfortable conclusion: table tennis is among the most loved sports in Vietnam and the most poorly measured. What is missing is not the score. What is missing is the process that produces the score.

What is published and what is withheld

The International Table Tennis Federation (ITTF) and World Table Tennis (WTT) publish a fair amount: match results, game-by-game scores, world rankings, seed lists, schedules, and at some majors a handful of raw indicators such as the share of points won on own service.

What is not published is a much longer list: ball speed at contact, spin rate in revolutions per minute, placement maps on the table surface, rally-length distributions, foot and body position data, and most importantly the matrix of serve type against receive type.

For comparison: football already has an open metrics ecosystem with dozens of providers; tennis has placement data from camera systems; basketball has league-level motion tracking. Table tennis is close to blank.

Three technical reasons make table tennis data far more expensive to collect than most sports.

First, scale. The table is only 2.74 metres long and 1.525 metres wide. A topspin loop from a world-class male player can send the ball beyond 100 km/h, and the ball crosses from one edge to the other in under 0.3 seconds. A standard 60 to 120 frames per second camera does not sample enough to reconstruct the spin axis.

Second, spin. Table tennis is a sport where the decisive quantity is also the hardest to measure. A heavy backspin serve and a no-spin serve can share the same flight path for the first 0.2 seconds and differ only in the bounce after contact. Measuring that requires high-speed cameras set at an angle and a calibration model for each rubber type.

Third, a closed data culture. Associations keep training and head-to-head data internal. The Chinese national team is famously secretive here, and that secrecy is part of their competitive advantage.

In Vietnam the picture is thinner still. The national championship and youth events mostly publish scores and livestream footage. No placement tables, no rally statistics. When I asked about data from a national semi-final a few seasons back, the only surviving record was a photographed paper scoresheet.

The 11-point game was built to add noise

In 2026 the ITTF moved from 21-point games to 11-point games and cut the service rotation from five points to two. The stated goal was shortening match duration for television.

Statistically, that change carried a consequence few fans notice: it increased the variance of game outcomes.

I simulated 100,000 games with a simple binomial model, assuming independent points and a player with a 0.55 probability of winning each point. The result: the probability of winning a game to 21 lands near 0.70; the probability of winning a game to 11 lands near 0.63. My simulation error is about plus or minus 0.5 percentage points.

Put differently, the current format deliberately makes single-game results harder to predict. Good for television. Bad for analysts, because one game carries almost no signal.

The practical consequence is concrete: the minimum sample size for any statement about form must be a best-of-7 match, not a game. And any statement about level must rest on a run of five to ten matches in comparable conditions.

Every table tennis statistics table I have seen in Vietnamese media violates this to some degree. They narrate one game and call it form.

The most abused indicator

The share of points won on own service is the most quoted and most misleading metric in the sport.

Its definition is simple: points won while serving divided by total points served. Three problems sit inside that definition.

First, it merges three different things into one cell: serve quality, third-ball quality, and the opponent's receive error. A player with an average serve and a ferocious third ball will post a higher number than an excellent server with poor conversion.

Second, it depends on the opponent. Against a player with a weak backhand flick, the number looks good immediately. Against a player with a strong backhand flick, it collapses. So does the metric measure the player or the opponent?

Third, sample size. A best-of-7 match gives roughly 35 to 45 service points per side. With 40 observations, the standard error of a proportion sits around plus or minus 8 percentage points. A 5% gap between two players in a single match lives entirely inside the noise. It does not exist.

To use this honestly, split it into three layers: outright service winners, third-ball conversion rate, and the rate at which the serve is attacked first by the opponent. Those three layers sum to one metric, but each tells a different story about the same player.

The correct unit is the serve-receive pair

Current rules require two service points per side in alternation, and from 10-10, one service point in alternation. That structure produces something analysts routinely ignore: the service pattern repeats every four points, and at deuce each side serves exactly one point.

The tactical consequence is clear. A player needs at least two options per service point, because the opponent sees each one twice in a row. And at deuce, each serve is a single event with no repetition inside that game — which makes deuce the place where service data is most useless and also the place where matches are decided.

So the valuable data unit is not "how many points did I win on serve" but the matrix of serve type against receive type. That is a two-dimensional table, not a single cell.

When I prototyped this matrix on a set of domestic matches, a repeating pattern appeared: young Vietnamese players often post strong numbers with a short backspin serve to the middle of the table, then collapse when the opponent flicks backhand directly. No aggregate metric shows this. Only the matrix does.

Equipment and the shifting balance of power

In 2026 the ITTF raised the ball diameter from 38mm to 40mm. Surface area grew by roughly 11%, air resistance rose with it, the ball travelled slower and spin dropped in relative terms.

In 2026 celluloid balls were replaced by plastic balls. The plastic ball offers a different contact feel, reduces spin further, and bounces more consistently between production batches.

Both changes tilted the sport toward athleticism and power while reducing the absolute value of a purely spin-based serve. But this is where easy conclusions go wrong.

What actually happens after every equipment rule change is that nations with deep coaching systems re-adapt faster. China did not lose its position after 2026 or after 2026. They needed roughly 18 months to adjust, then kept winning. Smaller nations, which cannot run parallel experiments across enough players, take years longer — and the gap widens during that window.

This is why I never read an equipment change as a standalone event. It is a shock, and shocks are absorbed unevenly.

The WTT ranking is a schedule metric

The WTT ranking system uses a rolling 52-week window and counts the eight best results. That sounds reasonable. It produces three effects fans routinely misread.

First, ranking is a function of both level and schedule. Players who enter more events have more chances to bank points. Players out with injury not only fail to add points, they slowly lose old ones as they expire.

Second, points-defence pressure is real and it changes entry decisions. A top-10 player may choose a smaller event to protect points rather than rest before a major. That is a financial decision, not a sporting one.

Third, and this is the part that directly affects Vietnam: travel cost. A player outside the world top 100 who wants to accumulate points must fly to Europe or the Middle East several times a year. With current support levels for Southeast Asian national teams, the number of events a player can enter in one season is usually very small. By my calculation, the ranking of most Vietnamese players reflects a tight budget more than it reflects ability.

That is a systematic distortion, and it cannot be fixed by working harder.

China's depth is not in the top three

For two decades China has dominated the three biggest events: the Olympics, the world championships and the World Cup. But looking only at the three leading names misses the most important thing.

The important thing sits in the 4-to-20 band. China's provincial selection system produces enough players that an internal training session is close in quality to an international quarter-final. They also use sparring partners trained to replicate the exact playing style of foreign opponents.

Japan took a different road: extreme youth acceleration. Harimoto Tomokazu won the Japanese national title at 14, and his sister Miwa Harimoto joined the senior team very young. That is a trade: skip an intermediate cohort and concentrate resources on early standouts.

Europe took a third road: the club league system. Germany's Bundesliga gives young players week-after-week matches against adults. Brazil's Hugo Calderano is an example of an exception created by a league system rather than an academy.

Those three models cannot be compared with a single number. They are three ways of organising the same problem: how to give a young player 10,000 hours of high-quality opposition before the age of 20.

The risk surface

Three risks I track most closely in the current cycle.

Schedule density. The number of WTT events rose sharply after 2026. For players using a two-winged topspin game, the shoulder and wrist carry the heaviest load. Injuries in those two locations rarely cost a match; they cost a season.

Generational gap. After Ma Long's generation, many Asian teams face the same problem: the next cohort has enough talent but too few top-level matches. This is the kind of risk that does not appear in the rankings until it has already appeared.

Data risk. When data is not public, fans consume narrative instead of evidence. The information market misprices, and players who perform well without a story are undervalued for an entire career.

I run one control question before every analysis: what is the probability this is just background noise? If it is above 30%, I stop and write honestly about the noise instead of constructing a conclusion.

The counterintuitive angle

The player with the highest service-point win rate at a tournament is usually not the best server. They are the player with the strongest third ball, or the player who happened to meet the most weak receivers on that run. Those are different in nature, and only one of them is a repeatable skill.

The deeper blind spot runs the other way. Adding point-level data does not illuminate table tennis; it increases false precision. A table with three decimal places looks more trustworthy than a table with one, even when both are wrong. I once published a model like that and was corrected by my own readers.

Data is not wrong, the reader is — and I have been that reader.

The Tokyo 2026 Olympics were a natural experiment few have exploited. Table tennis was played in arenas without spectators. Elite players read spin partly from the sound of contact, and a silent hall changes how they handle a short serve. Data without context is only half the truth.

I am not saying table tennis cannot be measured. I am saying what currently gets measured is not what decides results. A 30% probability is not an excuse — it is a reminder that I am right only seven times out of ten.

Next-cycle signals

Three things I will track over the next 12 months.

One, whether WTT publishes match-level placement data. If it does, the entire table tennis analytics industry rewrites itself.

Two, average rally length at major events. If it keeps falling, the value of the serve rises again, and old metrics go stale in ways nobody announces.

Three, the entry-cost threshold for Southeast Asian players. This is the least discussed variable and the one that decides who appears in the rankings at all.

Table tennis does not live inside a spreadsheet — but a spreadsheet helps me see table tennis more clearly. And sometimes the clearest view comes from an empty file, reminding us that most of what we call analysis is still waiting to be measured.