Is Skill Linear in Sumo?

Sumo is the national sport of Japan. Wrestlers (Rikishi) are ranked in a ladder format. If you win more than you lose, you move up the ladder (Banzuke). At the top of the ladder sits the Yokozuna. This is the highest level, with only 73 ever in the long history of Sumo.

For lower ranked wrestlers, beating a Yokozuna is a major accomplishment. Fans cheer and throw their seat cushions towards the ring. There is even a cash prize, depending on the wrestlers rank. Intuitively you would think that those highest on the ladder should be the ones to defeat the Yokozuna most often. Is that true?

I looked at the match history from sumodb for the last 20 years, or 2582 Yokozuna matches. Here is what the win % looks like for the next 8 ranks below the top. Note that it is very rare for very low ranked wrestlers to face a Yokozuna, so I stopped when sample size became too low (below M5).

Stats by Rank

The closest ranking (Ozeki) does have the best win percentage, although it is only 27%. What is curious is the reverse trend between Komusubi (13%) and M1 (16%). Note that the cash prize mentioned earlier is only from M1 and under... perhaps the cash incentive is enough for that extra effort.

Stats by Yokozuna

Now consider that not all Yokozuna are equal. Hakuho is widely considered the best of all time, and this analysis supports that claim. I've split out the dataset by vs. Hakuho, and vs. all other Yokozuna. The story of his dominance is quite clear.

It's easy to see the difference in win rates versus Hakuho. Noticable as well is the absence of the M1 spike. Win rates verse Hakuho steadily decrease as we go down the ladder, except for M5 which has a small sample size of only 21 bouts (4 of which he lossed).

So... no it doesn't seem linear. The spike in M1 win rates compared to the higher ranked Komusubi is too significant to ignore. Is the moral of the story just that a little cash goes a long way?