Greetings from the 10th Annual MIT Sloan Sports Analytics Conference. The nerdiness at the Boston Convention Center is palpable. You can feel it in the air as sports dorks from armchair quants to professors to team executives and staff to MBA students looking to land a job gather to talk shop.
MMA is noticeably absent from the panels again this year but that won’t stop us from doing some MMA analytics here. The datasets for UFC 195 and UFC Fight Night 81 were recently finalized which means one thing: We get to sick RoboJudge on Lawler vs. Condit and Dillashaw vs. Cruz!
The first two UFC title fights of the year were each in their own way barnburners with controversial endings that may not have reached robbery status but left many fans and media members fired up. You’re probably still pissed off if your name’s T.J. or Carlos…or you’re a cool baby with sunglasses enjoying life.
We’re not going to resolve any feelings of anger today as much as run the fight data through a model calibrated to round-by-round scoring decisions to get a probabilistic view of the outcomes. Remember, RoboJudge isn’t perfect and can’t see certain nuances of a fight. But he also easily sees every detail through slow motion, frame-by-frame analysis that can be difficult to discern in real-time with the naked eye.
Don’t think of RoboJudge as the truth or infallible, just another perspective. Besides, there is no true outcome when probabilities aren’t zero or 100%, only higher likelihoods and edges.
In the immortal words of DJ Quik, now let’s get down to business…
T.J. Dillashaw vs. Dominick Cruz
I looked forward to this fight more than Lawler-Condit or any other early-2016 title fight, and boy did it not disappoint. Watching Cruz or Dillashaw alone puts a smile on my face. Put the two together and make it free, and I’m Carlton Banks.
The official scorecards were all over the place – the perfect representation of this truly being one of those fights where almost every round is reasonably debatable, especially for cageside judges watching in real-time with all the potential viewing angle problems, dead zones, poles in the way, refs in the way, switching from the fight to the monitor, etc.
The RoboJudge report shows that the biggest edge for any single round was Dillashaw’s 4th at 70.1%. That may seem pretty far from a 50% pure toss-up, but try scoring live fights cageside sometime and you’ll quickly get a newfound appreciation for the type of round where three people out of 10 may disagree.
The other rounds were even closer. I’d guess you know at least one person for each of the 1st, 2nd, 3rd, and 5th rounds who’d argue with you over their scoring, possibly even the 4th.
For many, the fight came down to how they scored round 3 – the closest by RoboJudge standards. A typical judge in a round with those particular performance statistics tends to give the fighter in Dominick Cruz’s shorts the judging nod 56.3% of the time. Cruz had a five-strike edge in the all-important power head shot category, which judges tend to weigh almost twice as much as power to the body and legs where Dillashaw had his edge. Dillashaw had more volume, but a lot of his misses where whiffs after Cruz was long gone. Volume and activity with missed strikes can sometimes be confused for effective striking, but pure whiffs at distance are the easiest to discern from cageside.
Even though Cruz has a slight edge in the 3rd, the round was still legitimately debatable. The whole fight was legitimately debatable as anecdotally evidenced by the 12 writers who scored the fight for Cruz and 10 writers who gave the nod to Dillashaw.
While RoboJudge gives Cruz an edge in the majority of rounds, at the end of the day the fighter with Dillashaw’s performance should walk away the winner roughly 1-in-3 times. But that leaves 2-in-3 times for our new, returning, in a weird way defending, and in my opinion, rightful bantamweight champion Dominick Cruz. And what do you know, while they took dramatically different paths to get there, 2-of-3 cageside judges agreed.
Robbie Lawler vs. Carlos Condit
It flat-out sucks that someone had to lose this fight. Unlike Dillashaw-Cruz above, the official scoring for Lawler-Condit was pretty consistent, save for one round. The judges and most observers had the 1st for Condit, 2nd for Lawler, 4th for Condit, and 5th for Lawler. Looking back on it all, the fight came down to the 3rd.
But remember, judging odds are about the likelihood of being declared the winner after five rounds of action but before we know how the judges scored any rounds. In nerd speak, it’s ex post to the fight and ex ante to the reveal of scorecards.
Round 3 was close with Condit landing 18 power strikes to Lawler’s 11 but losing the head battle. The second minute of that round was clearly Lawler while the other four were Condit or neutral. When Lawler landed it looked more powerful (reminiscent of Bendo-Edgar fights) and that’s where an argument could be made about statistically analyzing a single fight. A landed power strike from Lawler probably isn’t the same as a power shot from Condit. But RoboJudge also has better vision than you and I and the real-life judges, clearly documenting every little detail.
Condit turned it on in the 4th round to the tune of 43 power shots landed to Lawler’s six, wobbling Lawler for good measure in the final minute. The 5th was a glorious gift from the MMA gods with Condit’s 55 power strikes landed to Lawler’s 49 plus one bleeding Condit face.
I always run two versions of RoboJudge, one where judges are free to score points for missed strikes if supported by the data and another where I force missed strikes to be valued at zero. Most of the time these two versions are so similar there’s no point in showing both (e.g., with Dillashaw-Cruz above) because one fighter’s swinging and missing volume doesn’t utterly dominate the other’s.
That gets thrown out the window with Condit.
Condit throws on so much damn volume with probing kicks and 2-, 3-, and 4-strike combinations that he’s a bit anomalous to the model in terms of missed strikes. Judges tend to, on average, give points for missed jabs to the head and power shots to the head and body, and Condit’s got those in spades. Looking at power head strikes alone, Condit missed 38, 38, 38, 35, and 76 in rounds 1-5, respectively, to Lawler’s 6, 12, 14, 10, and 33.
You begin to see why “The Natural Born Killer” is a natural born missing anomaly – in a good way, not a Leonard Garcia kind of way.
Here are the two RoboJudge reports, first where misses can have value and next where they can’t.
If we value Condit’s misses, he blows Lawler out of the water. But he’d probably blow any fighter out of the water.
If we go to the perfect world where misses aren’t scored at all, Condit had a strong 1st round at 83.4% and owned Lawler in the 4th at 99.9% (in a non-10-8 kind of way). Those two rounds alone put Condit in a pretty strong statistical position. Even average showings in two of the three other rounds would give him a pretty decent edge. According to RoboJudge, Condit’s 3rd round was worth 69.3% and the crazy ass 5th was basically 50/50.
At the end of the day, Condit comes out at 92.3% to be scored the winner over Lawler to take home the UFC welterweight strap. If you think it’s hard to discern some of Condit’s misses, make it higher. If you think Lawler’s power strikes should count more, make it lower. Either way, it’s a big number that matches well with the anecdotal evidence of 15-of-18 media members who sided with Condit (excluding draws).
Were either of these decisions mind-blowing robberies? Not really. But even though some would say, ex post, that it all came down to how you scored the 3rd round, Carlos Condit now has statistical reason calibrated to real-life MMA judges to be a lot more pissed off than T.J. Dillashaw. And the UFC has a statistical justification for booking a rematch if it so chooses.
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.
About the author
