Is former UFC featherweight champion Jose Aldo still “Leg Kick” Aldo because his stats show him attempting roughly 20% more distance power strikes to the legs than average? Is Abel Trujillo’s lower body takedown defense really an abysmal 35% since he happened to run into a buzz saw who went 16-of-19 in a lower-body manhandling of Trujillo one night?
If Kobe Bryant were still in the NBA, would we care how he performed in 2007 when thinking about his upcoming games? Probably not at all. If you’re a die-hard Laker fan, then fine, maybe just a little – and go Clippers!
By my new way of calculating MMA statistics, Aldo actually strikes opponents’ legs at roughly half the rate of his traditional stats – well below average – and Trujillo’s lower body takedown defense is in fact a respectable 64% – 7% higher than an average lightweight.
MMA statistics are in their infancy compared to the major sports. Add in a lack of seasons and a low number of “games” played and the stats we see for fighters are usually lifetime averages or totals.
But why?
How much does Anderson Silva’s six-year destruction of the UFC middleweight division from 2006 to 2012 matter when thinking about his future matchups today? And should getting taken down 21 times by Khabib Nurmagomedov in a dreadful night at the office eight fights and four years ago still continue to weigh heavily against Trujillo?
Longer fights might improve the precision of certain statistics, but they can also drown out the effect of shorter fights. Think of which stats will rule the day when a fighter has a 25-minute battle and then a 3-minute finish. When we hear traditional statistics on takedown defense, striking accuracy, or striking volume, the 25-minute fight will probably overwhelm the 3-minute one even though what we really got was two fights worth of information.
If only there were another way… some “alternative stats” if you will.
By finally getting the second version of my MMA prediction model to a point where I’m reasonably comfortable, a nice side effect was the development of weights that give more credit to recent performances, yet don’t fully ignore the past.
If recent events are weighed too much, off nights and bad matchups will ruin the data. Too little weight and you’re basically back to traditional statistics that count every strike or minute equally, even the ones from 10 years ago. One could arbitrarily pick a recent number of fights to examine, but that’s the analyst imposing their guess on the data. In my method, fighters’ bouts were weighted over time based on a grid search for the best predictive accuracy in model training and validation.
Each week before big cards or major fights, I’ll use those weights to share the algorithm’s fight predictions and also put an alternative spin on traditional MMA statistics. The stats may look and sound like what you’ve read from me before, but remember the key difference is how they give more credit to more recent fights and are geared towards fight-by-fight information, instead of longer fights dominating the action.
The previous version of the model was overfit and while it seemed to test well, I felt from the beginning it needed to be improved. This version is a trained, validated, and tested machine learning prediction model that over a 3 ½ year period of backtesting (running a model live with only the information you would’ve had at each moment in time) performed just as well as picking favorites (65.5% vs. Vegas line of 65.4%). There could easily be variance at play so don’t expect that to continue, but even getting close to the accuracy of picking favorites should hopefully yield reasonable edge predictions.
Here’s how the model performed in betting with a hypothetical $10,000 starting bankroll.
Only 36.9% of bets were winners with a longest winning streak of four in a row for $3,769 and a longest losing streak of seven for -$1,794. Out of 108 events with bets, the model lost money 55 times, broke even three times, and made a profit 50 times, but the profitable events did better than the unprofitable ones on average. It seems to go through periods of relative stagnation that can last for over a year (and could even have a significant downswing) until the right matchups come along or the under-the-radar underdogs with the right statistical underpinnings do their thing on the grand stage.
93% of bets were on underdogs, which is why you don’t have to win with great frequency to come out ahead. But you do have to be selective. One of the main rules of gambling is: You don’t have to gamble. Applied to prediction: I don’t have to predict. In that vein, the model is limited only to matchups where both fighters have at least four prior documented bouts and the heavyweight division is completely ignored. Stipe Miocic can take his belt and go home for all the model cares.
Since training and testing used the closing money lines, an Alternative Stats piece will go out midweek with the weighted statistics I think are most interesting and the predicted winner. A piece with precise win probabilities and possible bets will go out as close to the event start time as possible, or on Twitter for smaller shows.
This is an experiment and entertainment. Do not bet on the fights using these numbers. Taking sensible gambles with an edge over time is similar to investing. But if I’ve made a mistake somewhere, those same gambles without an edge become sucker bets. Let me be the possible sucker. You’ve been warned.
With that, let’s get down to the Alternative Stats of UFC 214’s incredible night. Remember, what you’re about to read is derived from official UFC statistics, but calculated differently. See the notes at the bottom for specific statistic definitions.
Daniel Cormier vs. Jon Jones
The rematch we’ve all been waiting for yet car crashes, injuries, speed racing, and tainted pills kept snatching away from us. It’s always tough doing a statistical analysis of a rematch, since you mainly want to talk about what happened in the first fight and possible adjustments.
Interestingly enough, Cormier’s bout closeness measure is lower than Jones, meaning win or lose, he does it a bit more decisively.
Right off the bat, Cormier shows us why Alternative Stats might be important. His traditional metrics show him out-landing opponents with power strikes at distance to the tune of 11.5 strikes to 10.5 per five minutes (P5M), but his Alt Stats have him getting out-landed by over 9 power shots P5M (8.6 to his opponents’ 17.9).
DC used to statistically piece-up his Strikeforce and early UFC opponents with power at distance, but that hasn’t quite been the case lately. His last five opponents outstruck him in that realm to the head and body, including Jones in their first encounter.
DC shows a power advantage in Alt Stat world with 2% more rounds with knockdowns and almost 3x the knockdown rate. But the clinch is where he does his best work, the only problem being that’s Jones’ zone as well. Jones throws more variation in his clinch attacks with an extra 3.2 power shots to the body P5M than DC.
For a former Olympic wrestler, DC’s traditional clinch takedown defense is a slightly above average 55%. His Alt Stats improve it to 73%, but once again Jones gets the better of him, going from 92% clinch takedown defense (traditional) to 97% (Alt Stats). Just like in their first fight, the numbers suggest that, if anyone, Jones should be able to drive the location of the fight.
Both guys are usually in control on the ground throwing nasty damaging blows and both standup ASAP the rare times they’re on bottom, 5.5x (Jones) to 12x (DC) as often as an average light heavyweight.
The pick is Jones, but here’s rooting for one hell of a fight no matter who walks away with the undisputed light heavyweight strap.
Tyron Woodley vs. Demian Maia
What if I told you the prediction was Maia by KO/TKO? Yeah, that’d be crazy.
We pretty much know what this fight’s about. The welterweight champion Woodley has most every striking advantage at distance and gets a knockdown in almost ¼ of his rounds (22.9% traditional). In Alt Stat’s “what have you done for me lately” world, it shoots up to a whopping 38.2%, over 4x the welterweight average.
Let’s not even talk about Maia’s distance striking. He’s got to get this thing to the ground and Woodley’s about an average takedown shot defender lately thanks to the last attempt being a successful one from Rory MacDonald.
Maia’s lower body takedowns in the clinch have been improving with an Alt Stat success rate of 62% (compared to 47% traditional and 44% welterweight average), but his real killer is volume at distance and in the clinch.
He shoots 1,047% more takedowns and lands 1,390% more than average at distance with the numbers coming down to a slightly less insane 170% more attempts with 240% more landed than average in clinch takedowns. And once he gets you on the canvas, he’s on top 88% of the time with half guard or better 2/3 of his time in control.
Opponents standup on Maia 53% less than average and he has a submission finishing rate of 53% in Alt Stat world. When Maia doesn’t finish you, he’s throwing dominant position submissions roughly 80% of the time so he’s less likely to lose his advantage even if your defenses happen to hold up.
It’s a close one, but the pick is Maia.
Cristiane Santos vs. Tonya Evinger Germaine De Randamie
Since Evinger has zero documented bouts, I figured why not dwell on what could’ve been. The model had Cyborg at 76.0% to take out the former… kind of… maybe champion of the Cyborg division, Germaine De Randamie.
Robbie Lawler vs. Donald Cerrone
Who cares who wins. Just let it be a glorious display of martial arts skill and regulated violence.
First off, Lawler has still never attempted a documented submission and it’s probably a good bet it’s not coming tomorrow.
The knockdown advantage goes to… Cowboy, in traditional and Alt Stats. Neither guy has good head defense from power strikes at distance, which fans will love. Neither guy out-lands their opponents with power at distance, and they barely out-land with jabs to the head.
Maybe I meant martial arts “will” when it comes to taking power shots from distance with these guys. They may eat their share of fists and shins, but they’re both pretty well above average in making their opponents crash to the canvas at some point and, on certain metrics, making their opponents leak – as Big John McCarthy and Jerin Valel like to call it – colored sweat.
With Lawler and Cerrone spending 3:52 and 3:55 of every five minutes, respectively, in open space at distance, Cerrone throwing roughly typical volume (Lawler less), their knockdown and damaging power, and an extreme lack of takedown attempts, put the kids to bed early and strap in for a show. Gentlemen, we salute you and thank you in advance.
The pick is Lawler in another close one.
Ricardo Lamas vs. Jason Knight
Making new sheets of 955 different statistics per matchup (no joke) took a lot longer than expected. Alt Stats will have to call it a day here.
The pick is Knight.
Renan Barao vs. Aljamain Sterling
The pick is Sterling.
Josh Burkman vs. Drew Dober
The pick is Dober.
Notes: Strike attempts are for an entire five minute round in each position (P5M) and are categorized as jab or power. A jab is just a non-power strike. Strikes are documented based on where they land or are targeted (head, body, legs), not the type that is thrown (punch, elbow, kick, knee). Visible damage rate is per five minutes the fighter is not on his back. It’s hard to bust up someone’s face while lying on your back. Damage percentage is per power head strike and distance head jab landed. Knockdown rate is per five minutes at distance or in the clinch off the cage. Knockdown percentage is per power head strike landed while standing. It’s really hard to knock someone down if they’re already on the ground. Clinch control is having the opponent pressed against the cage. Ground control is having top position. Submission attempts are per five minutes of ground control minus time spent in the opponent’s guard plus time spent with the opponent in guard. A bout closeness measure towards zero means a fighter is in blowouts (win or lose) and towards 100 means he is in very close fights.
Paul is an economics professor who covers MMA analytics and business for Bloody Elbow and is a licensed judge for the California Amateur Mixed Martial Arts Organization (CAMO). Follow him @MMAanalytics. Fight data provided by FightMetric.
About the author
