How would you sum up the FightMatrix rating system?
This software rating system takes an unbiased, objective look at all available professional MMA results from day 1 to the present. The rankings are a hybrid of boxing rankings and traditional MMA rankings in that they represent the “right now” like a boxing system, but fighters don’t “fall” as fast due to a loss.
Furthermore, there are also elements of foreshadowing and prediction-based penalties in an attempt to properly compensate for scenarios such as inactivity, bottom-feeding, and impressive debuting fighters.
Before the site’s creation, it was decided upon that displaying output of a purely statistical (ELO or equivalent) would make no sense for MMA. ELO methods are too basic, and an “accurate” statistical model, would’ve been finalized using past trends as benchmarks, which when used alone, is a poor methodology for the ever changing picture of MMA. Furthermore, these models can produce wildly anomalous results.
This system needed to be able to factor in special considerations like impressive debuting fighters, “fluke” wins, inactivity, etc, so the decision was made to study respected rankings, and make a software system that properly place the keystone fighters in the divisions, but take bias out of the equation by ignoring overly hyped fighters, giving credit to underrated fighters, and being able to pull the cord on fighters who are well past their prime. In addition, the system also excels by being much more scalable then your typical Top 10s and 15s, by analyzing even the most minor shows.
In summary, FightMatrix is a computerized rating system that attempts to capture un-biased “popular thought” and combine it with prediction-based analysis. FightMatrix’s system is truly one of a kind.
When do you update?
We usually update every Sunday or Monday.
When a fighter is dropped for being inactive, what happens to his/her rating?
At this point, the inactivity penalty has already kicked in. As time goes on, the penalty will continue to decrease their rating, at a progressively higher rate. When the fighter becomes active again, their new rating will depend on their old rating, time of inactivity, and most recent performance. As of recently, a modification was made that allows these penalized inactivity fighters to regain a large (but not total) portion of their lost points in their first two performances back.
Fighter A is ranked in the Top 10 at Lightweight.. but they only had one fight there!
The rating system is division-adjusted, not division-specific. In other words, fighters carry their previous achievements with them to their new divisions. If Joe Smith moves from Welterweight to Lightweight and has the 5th most points at Lightweight, he’s not necessarily the 5th best Lightweight, he is the 5th best fighter currently fighting at Lightweight. This stands true as long as Joe beat a semi-legitimate opponent with the last 3 years in the division he is currently ranked. If not, some points are temporarily removed until he does so, or returns to his previous division.
Hey, this guy is rated in the wrong division? How can I let you know?
Fighters are almost always ranked in the division in which they last fought. However, there are rare exceptions. One exception is when we have reason to believe that a fighter stepped outside of his division only for a single bout, and in addition, accomplished nothing. If you feel a fighter is ranked in the wrong division, please leave us a note in our forum.
Is the FightMatrix rating system similar to an ELO rating system?
Sort of. At its core, the system depends on a formula which is similar to that used in an ELO system. However, unlike ELO, the FightMatrix rating system used here puts more emphasis on the most recent event and contains a massive wrapper of code around the core formulas that are necessary to properly consider different types of bout outcomes, the effects of inactivity, impressive debuting fighters, etc.
Why is Fighter A ranked higher than Fighter B?
Because that is how the ratings ended up. Seriously, these questions are difficult to answer, because so many variables go into the rating of a certain fighter. If there is a ridiculous result (like Eric Esch being ranked above Fedor), then there is likely a database problem, and we invite you to please inform us of this. However, if the fighters in question are only a few points apart, then it is because they are at approximately the same level.
Why isn’t your system perfect?
I always use the weather forecast analogy when our results conflict with some opinions. How many times has the weather forecast in your area been wrong — the same system that has hundreds, if not thousands of different projection systems, that likely have millions of highly qualified manhours invested into them? Not to mention, the consistent, well thought-out data points and related delivery systems that stream highly accurate data on a frequent, consistent basis. Mixed martial arts is an imperfect sport, with imperfect judges and even an more imperfect matchmaking system. Oh yeah, don’t forget the questionable data reporting to boot. We do the best we can with what we are given.
Why is this fighter’s W-L-D record different than their record at (other source here)?
The reasons for this are endless. Missing bouts, incorrect results, faulty reporting, undetermined pro/am status.
How can I help??
There are two major ways that you can help.
1. Take a look at the “Unknown Division” rankings to help us determine which division these fighters should be assigned to.
2. Submit any missing or incorrect results to Sherdog, our main bout source.
What do the Mouse Over statistics refer to?
Last 3 Years – Win-Loss-Draw record for the three years prior to the ranking date.
“Big League″ Record – The Win-Loss-Draw record for the fighter in the organizations considered the “Big Leagues”. This “Big League refers to the top organizations all-time based on a statistical analysis we performed (UFC/PRIDE/WEC, DREAM/K-1, Shooto/Shooto Americas, WVR, Strike Force, BoDog, IFL, EliteXC, Bellator, GCM, TKO, Cage Rage, MFC, DEEP, Pancrase, Cage Warriors (UK), Jungle Fight, KOTC (US), Legacy (Texas), M-1 Mix-FC/Global, WSOF/PFL, KSW, Rizin, TitanFC, RFA, InvictaFC, OneFC, Absolute Akhmat/Berkut). The analysis implements minimum bouts per time period and average fighter ranks. Some organizations were counted as one, due to mergers, etc.
Last Quality Performance – The date on which the system feels that the fighter registered his or her most recent quality performance. A quality performance is a fight in which the fighter fought an acceptable opponent and substantiated their current rating (at the time) based on the outcome.
Quality Performance % – The percentage of career fights in which the fighter registered a quality performance.
540 Opponent Metric – This is an experimental metric that attempts to reflect a fighter’s recent opposition quality. The metric goes back 1,080 days, and looks at all the fighter’s opponents’ W-L-D records within a window of 540 days before, and 540 days after their respective bout(s) with the fighter. All of the records are added up and the cumulative win percentage is presented by this metric in a “strength of schedule” format.
Win Finish % – The percentage of a fighter’s wins that do not reach a decision.
Combat Age – One can equate “combat age” to “dog years” in that it will mirror the basic human lifecycle. With this new computation, we expect fighters to hit their respective “peak” in the years that your typical adult “peaks” in life, somewhere between 25 and 40. We will expect most fighters to start their descent at some point in their 40s and be far removed from their peak by 60.
Rank State – A fighter’s rank (US) state is assigned based on the state that the fighter most frequently fought in professionally, within the past three years. If there is a tie, the most recent state is used.
What are the differences in the Elo rating systems in use?
“Standard” Elo (ELO K-170)
The goal in the “standard” version is to keep the system as simple as possible, with prediction rates as high as possible for the entire database. A basic, straightforward mathematical system.
- Starter rating is 1000.
- K-Factor is 170.
- Uses the standard “algorithm of 400” Elo equation.
- Draws have a match impact of 0.5 for each fighter. Wins have a match impact of 1.0 for the winner, 0.0 for the loser.
The goal in the “modified” version is to build on the standard version — make enhancements to better handle common scenarios, but not become complex enough to where this system can’t be easily explained. Priority for prediction rates are for fights in which both fighters are beyond their third pro fight. A basic, straightforward mathematical system, with minimal, yet practical tweaks.
- Starter rating is 1000.
- K-Factor is 275 for fighters in their first three fights, 155 for everyone else.
- Uses the standard “algorithm of 400” Elo equation.
- Fighter that holds a continent-level home advantage is given a 15-point adjustment in the core formula. Ratings must be treated as “neutral ground”.
- Draws have a match impact of 0.5 for each fighter. Split decisions have a 0.667 for the winner, 0.333 for the loser. Majority decisions have a 0.833 for the winner, 0.167 for the loser. Other wins have a match impact of 1.0 for the winner, 0.0 for the loser.
- Disqualifications and 1-round technical draws are ignored.
The goal in introducing Glicko is to stay as pure to the Glicko system as possible. Prediction rates optimized for fighters with 5 or more pro fights. A slightly advanced mathematical system with tweaks that are equally practical and analytical.
- Starter rating is 1500. As a result, the displayed ratings have been reduced by 500 points to better match the Elo output.
- Starter/maximum RD (rating deviation) is 230.
- Inactivity never affects rating, but can affect RD. Inactivity equal to or less than 180 days does not affect RD. Days beyond 180 utilizes a C value of 87 to increase RD up to a maximum of 230. No contests and disqualifications are completely ignored and do not “reset” inactivity.
- A “rating period” for inactivity is a day, while a “rating period” for rating calculation is a fight.
- Draws have a match impact of 0.5 for each fighter. Split decisions have a 0.55 for the winner, 0.45 for the loser. Majority decisions have a 0.61 for the winner, 0.39 for the loser. Unanimous decisions have a 0.91 for the winner, 0.09 for the loser. Other wins have a match impact of 1.0 for the winner, 0.0 for the loser.
“Whole History Ratings” is an Elo-based system that depends on multiple iterations to produce rating output. The standard Elo/Glicko systems update ratings incrementally, run one time and go directly from start-> finish, whereas WHR uses information gathered in previous iterations to build what essentially equates to a multi-dimensional moving average where the pre- AND post-bout results of a fighter’s opponent, or their opponents and on and on, can have an impact. Prediction rates optimized for recent UFC fights. A moderately advanced mathematical system with purely analytical tweaks.
- Starter rating is 0 and can go positive or negative. After the rating runs are complete, ratings are converted to an Elo-like output and 1000 is added for display purposes so that the top fighter’s rating falls roughly in line with Elo/Glicko systems.
- The w^2 value = 42.
- Iterations conclude once the convergence drops below 0.00001. This takes roughly 70 iterations.
- Non-unanimous decisions are treated as draws.
- No contests, 1-round technical draws and disqualifications are ignored.
My question has not been answered in the FAQ. How do I contact you?
If you don’t see your question answered anywhere, please leave a comment!