Are the common beliefs true? What Data Oyakata found in sumo
These findings test a vast history of bouts using only data that was available before each basho began. They do not dismiss the fun of sumo. Instead, they offer a slightly mischievous look at the depth that numbers can reveal.
It was true. A rikishi at 7-7 on Day 14 wins 56.2% of the time on the final day, six points above an even 50%. By contrast, an 8-6 rikishi wins only 45.0%.
Read the results and methodologyThe chaotic day was the final day, not opening day. Upsets occurred on 35.8% of opening days, about average; Day 7 was the most orderly at 32.8%.
Read the results and methodologyIt was true. The day after Kinboshi, Hiramaku defeated Yokozuna, Rikishi's Win rate was 38.0%. This is 10 points lower than the same rikishi and the same Basho on other days (48.2%). I'd like to say it's a reaction to a big job, but I'd like to clarify something.
Read the results and methodologyThe numbers pointed the other way. Rikishi returning from a full absence won 59.2% in their comeback basho and 65.9% on opening day. The explanation is the banzuke: after a layoff, they often return at a lower rank and face lower-ranked opponents.
Read the results and methodologyIt was true. Kadoban Ozeki averaged 8.15 wins and achieved kachi-koshi 78.2% of the time, above the 75.3% rate for ordinary Ozeki. The data suggests that pressure can sharpen their focus.
Read the results and methodologyPushing style vs. Belt wrestling——We re-measured the ``compatibility of styles'' that fans value with only the first one, whose Strength index before the bout is almost even. The win rate on the Pushing style side is 49.7%. The Belt wrestling side has 50.3%, which is almost 50%.
Read the results and methodologyHalf true, half wrong. Touhou's rikishi averaged 7.59 wins, while Nishi's rikishi averaged 7.41 wins. It is true that the East is winning. However, when comparing the same status (for example, Maegashira 5) between East and West, the difference is almost zero.
Read the results and methodologyThe wall was clearly there. Rikishi, who was promoted to Sanyaku (Komusubi/Sekiwake) for the first time, averaged 6.16 wins. Only 27.7% can beat Win. It is clearly lower than Komusubi's overall winning record rate (42.3%).
Read the results and methodologyIt was true. Rikishi, who weighs over 158kg, stalled from 50.5% in the early stages (1st to 5th day) to 47.8% in the late stages (11th day onward). On the other hand, the lightweight Rikishi has increased from 48.3% to 50.7%.
Read the results and methodologyIt was true. And it's faster than I expected. Sorting Makuuchi's win rate by age, it's a beautiful downward slope: 52.8% in the early 20s → 50.2% in the late 20s → 47.6% in the early 30s → 43.7% in the 35 years or more.
Read the results and methodologyIt was true. rikishi's MakuuchiWin rate of 25% or more of the wins and earning with Technique (Hatakikomi, Hikiotoshi, etc.) is 46.2%. Rikishi with a pulling technique of 12% or less is 50.1%, a difference of 3.9 points.
Read the results and methodologyIt was a fantasy. The turbulence rate for local Bashos in Osaka, Nagoya, and Fukuoka is 35.7%, and 36.3% for Tokyo 3 Bashos—in fact, Tokyo is slightly more rough. Looking at each venue, the roughest one is Hatsu Basho (37.9%) during the New Year, and the hardest one is Natsu Basho (34.9%).
Read the results and methodologyUnfortunately, it was an illusion. The win rate of rikishi from Osaka, Nagoya, and Fukuoka area at local Basho was 47.3%. There is no difference from other venues (47.9%). As a trend, even among rikishi from the metropolitan area, rural Basho (50.6%) has a slight advantage over Tokyo (48.8%)—although this difference is too small to deny as an error.
Read the results and methodologyThe wave story is slightly true. The fluctuation of wins for each Basho is Pushing style 2.14 and Belt wrestling 2.05, and it is true that pushing is rougher. However, it is a close difference. What was more eye-catching was the difference in average wins.
Read the results and methodologyThe numbers will fall, but it is too early to conclude that this is a ``slack''. The win rate after reaching the 8th win was 52.7%, which is certainly a drop from the momentum (65.4%) up until that point. However, in the final stages, players with good stars are matched against each other, so after the winning record, the opponent becomes stronger.
Read the results and methodologyA little bit true. Rikishi, who is competing in Makuuchi for the first time, has an average of 7.42 wins, and is able to win just half of the time. This is slightly higher than the re-entering Baku group (7.23 wins) and the veteran Hiramaku (7.28 wins) with over 30 years of experience.
Read the results and methodologyA 100-year record of Winning techniques from 1927 to today. "Tsuridashi" (a great technique of lifting an opponent and carrying him out of the ring), which was commonly seen in the dohyo before the war, is 0.1% of all winning techniques in the 2020s - about 1/50th of that in the 1930s.
Read the results and methodology2003 Natsu Basho Onward Counting all weight classes before the bout, 46.7% have fought against the same opponent at least once in the past. Half or more was the first meeting. Even if there is a history of competition, if it's just a number, it's hard to tell whether it's ``compatibility'' or a coincidence.
Read the results and methodologyIf you drag the knob (tap is OK on smartphones), it will be replaced with the type Rankings of that era.
The length of the bar is the rank in that era (1st place is full). Please see the table below for the actual percentages by age.
Sample is only Winning techniques (excluding Fusensho, Hansoku, etc.). Years with little data are not displayed.
This is a story about the idea behind Predictions and its honest limitations.
52% of the bouts were bouts where the abilities were almost equal (the difference in strength was very small). Accuracy there was 55% -- just a hair's worth of flipping a coin. In other words, half of the fun of sumo is that no matter how much you analyze it, it is essentially impossible to read. The upset is not an accident, but the very design of sumo.
Banzuke is a "rating" determined by people based on the previous basho's record. But rikishi's true strength may change before Banzuke catches up. The "Strength index" on this site is a number that is updated daily by accumulating wins and losses alone. We will quickly show the ``growing rikishi'' who have not yet appeared in Banzuke, and the ``veterans waiting to return'' who remain capable even if their status has declined. In fact, when comparing Predictions using only the data before Basho started, Makuuchi Kunio's Strength index was more accurate than Banzuke's (approximately 60% vs. 56%). Banzuke's "hidden strength" that cannot be seen just by looking at his top and bottom - this is where the seeds of trouble lie. Please take a look at the "Strength index" and "Momentum" on the rikishi page.
For young, inexperienced players, their evaluations will vary greatly depending on their rank. Since the bottom is still not in sight, if they win, their ratings will go up all at once, and they will truly “become different.” On the other hand, veterans who have been studying for a long time have established skills and stable evaluations -- in other words, they are "easy to read." Data Oyakata also became more accurate when I created it so that younger users could quickly import new results. ``That new recruit has great momentum'' and ``That veteran has a sense of stability'' -- the impressions that fans felt were also reflected in the numbers. That's why you can't miss the opportunity for a young, energetic player to take the top spot.
Even after trying various things and refining how to measure strength, Kunio Makuuchi's accuracy has plateaued at about 60%. This is the ceiling that can be reached with the current data. The remaining 40% don't hit. But it also proves that the numbers have not yet caught up with Rikishi's efforts and moments in the ring.
*Predictions are based only on data known before the Basho started. Numbers are approximate based on the most recent Basho.
Data is updated daily and interpretations are reviewed from time to time.