BE Analytics: RoboJudge evaluates all 248 UFC decisions in 2014

RoboJudge is back to score every UFC fight that went to a decision last year. That means we’ve got 248 fights and 778 rounds to get through. Cancel all your appointments; we’ve got some serious work to do. On second thought, maybe we won’t get into detail on every single fight. Complete results will still be listed in a ginormous table for your viewing pleasure. Information overload? Perhaps. Here’s hoping some people will enjoy being able to check out any fight that piques their interest.

RoboJudge is MMA’s quantitative official. He uses FightMetric’s detailed performance statistics and is calibrated to the round-by-round scoring decisions of real-life MMA judges. By no means do I claim RoboJudge is perfect. Like any judge, he’s allowed to have some weird decisions and my sense is they’re more likely to happen when a fighter throws very few strikes while engaging in a lot of unthreatening grappling.

I look at RoboJudge as an additional piece of information. He combines the precision of FightMetric with the common sentiments of MMA judges to quantify which decisions appear consistent with the way typical judges tend to act and which seem unusual.

RoboJudge’s biggest advantage is in seeing every single thing that happens inside the Octagon. FightMetric’s official scorers view multiple camera angles, rewind and slow things down to frame-by-frame if need be. Since RoboJudge is an extension of FightMetric’s data, he sees what they see, which may very well include some things we missed. He uses 24 pieces of performance information to score each round and knows exactly what landed, where it landed, if it was powerful and what missed or was blocked.

RoboJudge’s biggest limitation is that he’s a computer program and, as such, he doesn’t have a human’s ability to examine the subtleties and nuances of a round. One could also argue that RoboJudge isn’t programmed with the scoring criteria of the Unified Rules, but I’d beg to differ. While he may not be directly programmed, his calibration to real-life MMA judges makes him indirectly programed with the Unified Rules to the extent an average judge tends to employ them.

In his current form, RoboJudge doesn’t do 10-8 scores or point deductions. Point deductions are extremely rare and 10-8’s make up only around 3.5 percent of all rounds scored, but keep this in the back of your mind just in case.

How RoboJudging Works

RoboJudge scores each round from 0 – 100 representing the probability a typical MMA judge would side with the fighter listed in the Winner column. I then use the round-by-round odds to calculate the probability of being declared the bout winner (at least 2-of-3 or 3-of-5 rounds). Finally, I calculate the probability a panel of three typical judges would side with the fighter in the Winner column (at least 2-of-3 judges).

If the RoboJudge panel gives the winner greater than 50 percent odds, they agree with the decision. The panel disagrees when the winning fighter’s odds are less than 50 percent, but the level of disagreement depends on the odds. Robbie Lawler taking the title from Johny Hendricks at UFC 181 with 40.5 percent odds isn’t nearly as controversial as Diego Sanchez beating Ross Pearson with 4.7 percent odds.

RoboJudging 2014

The table below contains the complete set of decisions last year. There was only one draw and it’s in bold. Norman Parke earned a majority draw against Leonardo Santos at UFC Fight Night Natal in March. RoboJudge doesn’t know that Parke had a point taken away for grabbing Santos’ shorts in the 2^nd round. Santos took the 1^st round while Parke won the 2^nd and 3^rd by a wide margin and was 99.7 percent to win the fight as a whole, if not for the point deduction.

The table’s fun to peruse but isn’t the most user-friendly resource for examining some of the tougher or more unusual decisions. Those are better examined with table listing the 34 fights where RoboJudge disagreed with the decision, in order of disagreement from the biggest outliers to the smallest.

In 503 bouts in 2014, the UFC had 248 decisions. RoboJudge agreed with 214 and disagreed with 34 of them for an 86.3 percent agreement rate. We could spend an ungodly amount of time dissecting each decision so I’m going to look at the top 10 most disputed UFC decisions of 2014 according to MMA Decisions, as well as Henderson/Thomson at UFC on FOX 10. The Hendricks/Lawler title fight series has already been examined in a prior article.

Of the top 10 disputed UFC fights, here are the eight where RoboJudge disagrees with the outcome. Click on the matchup title to see the MMA Decisions page for each fight.

Diego Sanchez over Ross Pearson

Sean Strickland over Luke Barnatt

Kiichi Kunimoto over Richard Walsh

Lipeng Zhang over Wang Sai

Joe Ellenberger over James Moontasri

Warlley Alves over Alan Jouban

John Moraga over Dustin Ortiz

Andrei Arlovski over Brendan Schaub

Here are the two fights where RoboJudge agrees with the outcome:

Russell Doane over Marcus Brimage

Alan Patrick over John Makdessi

Doane/Brimage is interesting because all but one media member had it 29-28 for Brimage with Doane taking the 1^st and Brimage the 2^nd and 3^rd. MMA Mania described the 2^nd round pretty well, “10-9… Brimage, but it’s close.”

RoboJudge gives the 1^st solidly to Doane (98.6 percent) and the 2^nd and 3^rd both to Brimage (42.8 and 23.3 percent), but the overall fight odds are 58.2 percent for Doane. The basic idea is that Doane had a lock on the 1^st round and therefore only needed to win one more. While Brimage had the edge in the other two rounds, it wasn’t a large enough edge to make it sufficiently likely he’d win at least two rounds.

Patrick/Makdessi may be the fight where the largest number of people disagree with RoboJudge. Most media members scored it 29-28 Makdessi with Patrick taking the 1^st and Makdessi taking the final two rounds. RoboJudge agrees with the 1^st and the 2^nd rounds, but not necessarily the 3^rd. Here are the relevant stats for the round:

Note: Clinch control is having the opponent pressed against the cage. Ground control is being on top, but the positions can vary. Miscellaneous ground control is any controlling position other than guard, half guard, side control, mount or having the back (e.g., controlling a turtled opponent, controlling on the ground with the opponent pressed against the cage, standing while the opponent lies on the ground, etc.).

While Makdessi connected a total of nine times more than Patrick, many of his strikes were jabs. Meanwhile, Patrick had a huge advantage in power to the head and legs, along with a little bit of clinch control. RoboJudge valued the power over the volume.

Diego Sanchez over Ross Pearson

For a comparison of my approach to Pearson/Sanchez versus Fightnomics’ approach, see yesterday’s Fact-Checking Fightnomics article.

Pearson and Sanchez threw down at UFC Fight Night Albuquerque in June and started a firestorm of controversy after a split decision victory for Sanchez. Every media member but one scored the fight 30-27 for Pearson (the other was 29-28) and RoboJudge is in complete agreement. Here’s how the fight played out in each round:

Overall, Diego Sanchez had no better than a 30 percent chance of winning any particular round and only a 4.7 percent chance of winning the fight. But luck intervened and kept Sanchez from a 3-fight losing streak while telling Pearson, “No win bonus for you!”

Benson Henderson over Josh Thomson

Remember this fight? It went down in January at UFC on FOX 10 in Chicago, Illinois. It’s not in the top 10 most disputed UFC decisions, but it gets people riled nevertheless. It’s where we became certain that Bendo was judging kryptonite.

Most judges and media agreed that the 1^st went to Thomson and the 3^rd went to Henderson. Scores for the 2^nd, 4^th, and 5^th rounds are all over the place. Some gave them all to Henderson while others gave all to Thomson. The most common media scenario was Henderson taking the 2^nd round and the 4^th and 5^th going to Thomson.

RoboJudge gave all three of these rounds to Bendo, with the highest odds at 73.7 and the lowest at 59.9. The main tradeoff in the 2^nd round was Bendo’s striking advantage and Thomson’s 31 seconds of back control. If he were able to be more active with the back, we might’ve had a different story.

The 4^th round was another striking/grappling tradeoff, with Thomson’s Achilles heel of low volume possibly biting him in the ass yet again. The 5^th was mostly striking with Bendo’s power winning out for RoboJudge.

Without spending an inordinate amount of time examining decision after decision, that’s the 2014 judging year in review. Overall, I’m pretty happy with RJ’s performance. Is there anything you notice in the tables that stands out (good or bad)? Let me know in the comments.

Paul is Bloody Elbow’s analytics writer. All mistakes are his own and they’ve been known to happen sometimes. Follow him @MMAanalytics. Fight data provided by FightMetric.