MMA Sabermetrics: FightMatrix Responds
by Nate Wilcox on Apr 16, 2009 9:15 PM EDT
in MMA News
Email
PrintBE stalwart Leland Roling put up a pretty epic post last night MMA Sabermetrics: Strength of Record. Not to be outgeeked, jcs of Fightmatrix fires back with his own take on how to generate the stat:
Although his efforts were more than noteworthy, after taking a closer look, I felt the combination of the two layers was a bit subjective in nature. Furthermore, the statistics included the entire careers of all fighters involved, which I felt was overkill.
I decided to take a modified version of this on. For starters, I only included the winning percentage of the opponents of a fighter’s opponents; that 2nd opponent layer. I concluded that only including a fighters’ opponents was too shallow of a measure, and if you were going to include both, you may as well only include the 2nd opponent layer, rather than weight the two opponent layers in an attempt to combine them. Also, I decided to utilize a dynamic 3-year sliding window, that would adapt itself to the bouts found at each layer.
Here’s the Top 10, if anyone is interested, I’ll post a more comprehensive list.
Fighter W L D Win% Urijah Faber 93 28 1 76.9% Katsuhisa Fujii 76 23 1 76.8% Yoshiro Maeda 156 49 6 76.1% Jens Pulver 110 35 1 75.9% Brian Foster 80 26 1 75.5% Toshiaki Kitada 89 30 14 74.8% John Gunderson 102 35 2 74.5% Kazuyuki Fujita 90 31 1 74.4% Tatsuya Kawajiri 123 43 2 74.1% Jeremy Stephens 80 28 1 74.1%
26 comments
|
0 recs |
Do you like this story?
Comments
Awesome
Fantastic to see that Leland’s posts are generating a discussion in the larger community. I was surprised by Jens Pulver’s high ranking—he’s dropped his last three fights!—in the Fightmatrix rating (perhaps he benefited from two fights with Faber and the “including opponents” layer?). Putting aside the vaguaries of statistics, to my eyes Leland’s ranking was a far better representation of the current MMA landscape that Fightlinker’s.
"It's like a flying knuckle sandwich." --Rogan
"And many men have eaten it." -- Goldy
yeah it probably cause of two fights with the #1 guy..
i dont know which is better, but i say, major props to leland for doing the shit he continually does for this site. :)
by Anton Tabuena on Apr 16, 2009 10:23 PM EDT up reply actions
I doubt it...
it’s two different ways of looking at the stat…which is how stats usually get going. And then eventually the best way is figured out.
Contributing Editor - BloodyElbow.com - SBNation's mixed martial arts headquarters.
http://CurseOfRonKarkovice.blogspot.com/
by Brent Brookhouse on Apr 16, 2009 9:40 PM EDT up reply actions
SoS has always been a hotly debated topic, and in all honesty, it’s not exactly a NEEDED stat, so the best method in determining it will likely never really be pushed. There are several other methods that I’m going to test out, and there is definitely one very interesting one that takes an entirely different spin. I just need to find the time.
Follow my analysis of all things MMA on BloodyElbow.com
by Leland Roling on Apr 17, 2009 12:21 AM EDT up reply actions
I prefer Leland’s stats for the following reason: I believe we all know who the best fighters (generally) are. It would stand to reason that taking a stat to show SoR should in some way indicate who the ‘best’ fighters are. As shown from above, you can use statistics and come up with some way to make certain fighters rise to the top. I.e. I’m guessing this stat could somehow be tweaked so that King Mo somehow became No. 1 on this list.
Given that I think that BJ Penn and GSP are among the top guys in the fight world, Leland’s rankings make more sense to me.
Out of interest – FM – Where do the top 5 on each metarankings list lie?
It just goes to show, my worth only goes so far when analyzing things. All of the analytical pieces of every intricate stat can be done in different ways and thought it different ways.
JM’s stat has some good things. It’s a sliding window, therefore eliminating some of the problems with past fights, but I still love the more historical look because it still includes those wins and losses from when they were relevant way back.
It’s easily up to the user of the stats to make assumptions from there. Unless you’re completely oblivious to the landscape of MMA today, you can easily see how a very, very high SoS on a guy who has taken on the likes of fighters who are on the downturn of their careers would be affected.
But… I think BJ Penn, Georges St. Pierre have the top spot overall. Just look at their records, and the records of those opponents.
This is a solid stat by FightMatrix though, as it works on a sliding window, which works well in what JCS_FM wants to put out.
Follow my analysis of all things MMA on BloodyElbow.com
straight up....
you could use any parameters to make certain fighters rise to the top
Yeah...
but a good stat shows how it is figured out and you can use your own judgment to determine its value.
Contributing Editor - BloodyElbow.com - SBNation's mixed martial arts headquarters.
http://CurseOfRonKarkovice.blogspot.com/
by Brent Brookhouse on Apr 17, 2009 10:06 AM EDT up reply actions
You’ll only do that if you’re intentionally trying to.
Follow my analysis of all things MMA on BloodyElbow.com
by Leland Roling on Apr 17, 2009 10:07 AM EDT up reply actions
I posted a Top 50 list... for those who wanted to see a larger one.
http://www.fightmatrix.com/2009/04/15/strength-of-schedule-with-a-twist/
Creator of the FightMatrix rating system.
Perhaps somebody could clarify something for me. Those name in the list. Is that simply ranking the fighters based on how difficult their opponents have been?
If so, how is it that Mark Hunt isnt in that list? In 10 fights he’s fought Manhoef, Overeem, Barnett, Fedor, TK, Crocop, Wanderlei, Yoshida, Bobish and Nishijima.
I dont know of any fighter who, on average, has fought such tough opponents.
MMA-ELO has SoS rankings that seem to be more consistent with my expectations.
http://www.mma-elo.com/me_hist.php?fighter=30
Their top 10 SoS looks like this:
1. Mark Hunt
2. BJ Penn
3. GSP
4. Chuck Liddell
5. Josh Barnett
6. Koscheck
7. Couture
8. Arlovski
9. Hendo
10. Florian
I must be missing something because I dont see how John Gunderson has a higher SoS than GSP or BJ.
I read your explanation but was a little confused. So whilst in the last 3 years, GSP has fought BJ, Fitch, Serra, Hughes, and Koshceck, he doesnt factor in your top 50 because of the winning percentage that his opponents and his opponents opponents have had in those 3 years. Is that right?
What are your thoughts on an ELO ratings system and the ELO SoS I linked to above. It seems to provide a more useful and accurate output that can also be varied using a few tweaks.
I dunno, I guess I just have trouble determining the significance or value of a top 10 SoS list that includes Katsuhisa Fujii at number 2, whose strongest opponent was probably Sinosic or Sang Soo Lee and Brian Foster at number 5, whose strongest opponent would have been Radach, but excludes GSP and Chuck Liddell from the top 50.
If Im missing the point or failing to interpret the number correctly please let me know.
The site you provided
Based their strength of schedule on ELO ratings, a rating that is far too basic for MMA.
I did so on Win/Loss/Draw records, which is much more objective.
Creator of the FightMatrix rating system.
What complexities does your system take into consideration that an ELO ratings system doesnt? Also, from what I read, the ELO system is entirely objective. What makes your technique more objective or useful.
(Im not trying to be an antagonist, just trying to understand your system better as at face value to me, the ELO system provides a more meaningful output.)
ELO
Puts the same value on KO as it does split decision. That alone leads me to believe that its too basic for combat sports.
Every objective measure has some subjectivity behind it. Nothing can ever be truly subjective.
Creator of the FightMatrix rating system.
I would view awarding KO, submission, unan and split decisions differently as a very subjective measure. I mean, you look at Bisping v Leben in a unanimous decision. Bisping said that was his goal. So he achieved that. Should he be penalized due to not finishing the fight as opposed to Anderson who got a (t)ko over Cote? Once you start measuring the performance based on how the fight took place you start going down the slippery slope of subjectivity and the objectivity of a system can become compromised. Personally, put a lot of weight on guys who end fights as opposed to going to decision but in attempting to be as objective as possible, how can you decide on which winning method is more significant?
Predictability of the outcomes of similar fighters at similar points in their career. Its pretty obvious that a KO is more impressive, and more representative of a skill difference than an SD is. If you can agree that a win is worth more than a draw, you should also agree that a TKO is worth more than an SD.
If you don’t finish your opponent, you leave open the opportunity of him finishing you, or winning a decision. So yes, a KO should be worth more than a decision of any kind. If you can’t finish your opponent it not only says something about you, but that of your opponent.
Creator of the FightMatrix rating system.
I would be more impressed with someone that stood 5 rounds with Anderson Silva and beat him at his own game in a decision than someone who came in and caught him off guard early in the first and won by TKO. Serra v GSP is another example. Whilst Serra was able to get the TKO, I think Serra getting a decision win against GSP would be a more difficult and impressive accomplishment. So I guess I disagree that a (T)KO is always indicative of skill difference and should always be scored higher in an objective system.
Unless you can objectively measure the skills/weaknesses of fighters, its hard to really measure how impressive a particular type of win is against a particular opponent. I mean, submission against Aoki is more impressive than a KO. A KO of Brian Stann is more impressive than a submission and a decision win against GSP, imo, is more impressive than a KO. Objectively measuring all that is a hard task though.
Having said all that, I do see the argument for a SD to be scored less though as with SDs there is often uncertainty about who really won the fight.
From his explanation:
For starters, I only included the winning percentage of the opponents of a fighter’s opponents; that 2nd opponent layer. I concluded that only including a fighters’ opponents was too shallow of a measure, and if you were going to include both, you may as well only include the 2nd opponent layer, rather than weight the two opponent layers in an attempt to combine them. Also, I decided to utilize a dynamic 3-year sliding window, that would adapt itself to the bouts found at each layer.
Contributing Editor - BloodyElbow.com - SBNation's mixed martial arts headquarters.
http://CurseOfRonKarkovice.blogspot.com/
by Brent Brookhouse on Apr 17, 2009 9:01 PM EDT up reply actions
Comments For This Post Are Closed
Secondary Sidebar
User Tools
Related FanPosts on Bloody Elbow
FanPosts
Community blog posts and discussion.
Recommended FanPosts
Recent FanPosts
MMA Rankings
