Fact-Checking Fightnomics: Is the Pace Advantage real?

"Apparently it's not always the damage done when it comes to winning fights, it's the thought that counts."

These are the words of Reed Kuhn, a.k.a. Fightnomics, after coming to the conclusion that fighters with a higher pace - or rate of significant strike attempts per minute - have an advantage going into an MMA bout. His full write-up can be found online for free or you can pay money and read it as a section in his book. You probably prefer free so I'll wait for you to finish reading it.

Now that you're done, raise your hand if you read Mr. Kuhn's take on pace and thought, "Whaaaaaaaaaat? Attempting a high rate of significant strikes is what matters instead landing at a high rate?" The difference between the two measures is misses and we know that swinging and missing can help fighters win in certain situations, but are misses such an important component of pace that we should shift attention to strike attempts instead of strikes landed? Mr. Kuhn seems to think so as he notes, "We're also seeing the first hint that the rate of Significant Strikes landed may not be the key differentiator compared to the rate of attempts" (emphasis in original).

My fight brain has a fundamental problem with this result. As a guy who does sports analytics for a living, it's not that I don't believe in statistical findings. Rather, I don't believe in statistical findings when I suspect there's a problem with the underlying analysis leading to the result. And that's what this series is all about, putting the non-peer-reviewed, non-critiqued work of Fightnomics under the microscope to see what's legit and what might be a little fishy. Let the fact-checking begin.

Fightnomics' Chart

It took a few tries to figure out what Mr. Kuhn did in his analysis. I can't be 100% sure but it appears he took the list of UFC fighters who fought between Jan. 1, 2008 and some cutoff date in 2013 (I used Oct. 31, 2013 as his piece was published online in November). He then calculated each fighter's rate of significant strike attempts per minute based on their total number of fights in the sample period. Finally, he categorized each fighter as being low pace, average or high pace and computed an overall win percentage for each category. This is my best guess as to what Mr. Kuhn did.

Here's a re-creation of his chart using my FightMetric dataset and the methodology just described. Some things may be slightly different as we might've cleaned our data in different ways or possibly used a different cutoff date.

Has your fight world just been rocked by Science!!!? From this chart, Mr. Kuhn concludes that high-pace fighters have win rates 7% above baseline (≈50%) while low-pace fighters have win rates 11% below baseline (his low-pace number is 39%). He also believes it shows us "the first hint that the rate of Significant Strikes landed may not be the key differentiator compared to the rate of attempts" and that it's not always damage that matters, but the thought that counts.

This is particularly confusing when we look at a chart that wasn't included in his analysis. If we make the exact same chart using the pace of significant strikes landed instead of attempted, we get something even more striking - no pun intended.

There's a fundamental identity in MMA performance metrics.

Strikes Attempted = Strikes Landed + Strikes Missed

Mr. Kuhn could have chosen Strikes Landed but instead chose Strikes Attempted, implying that he believes Strikes Attempted is better. I don't really see it, do you? I also don't see how significant strikes landed may not be a key differentiator compared to the rate of attempts. It kind of looks better to use the pace of a fighter's significant strikes landed.

Is it possible he didn't compute these results? Or maybe we had almost identical results for strikes attempted but somehow got different results for strikes landed? I certainly hope this didn't have anything to do with trying to sell a book and getting better shock value by suggesting attempts matter more than damage.

When we examine the remaining component of attempts, strikes missed, we get something that looks almost worthless.

Average and high-pace missers do better than low-pace missers in this chart but there's no point spending time discussing it since the setup is strange to begin with.

In sports analytics, there are two ways of using data to examine winning and losing: trying to explain what has happened and trying to predict what will happen. The problem with Mr. Kuhn's original chart is it has nothing to do with using a pace metric to predict fights that haven't happened yet. He talks about its "predictive value" but uses a setup closer to that of explaining fights that have already happened - just done in a really weird way. I honestly don't know what this setup is useful for.

It's unclear why someone would analyze pace in this manner, especially when we know the individual matchups and can calculate and compare each fighter's pace before they enter the cage. That's what we'll be doing a little later.

Brains First

In what follows, pace will be examined in two ways today: first with our brains and second with data. Using our brains, we first recognize that fighting isn't like other sports which have alternating possessions. In basketball, teams essentially have an equal number of possessions throughout the course of a game (+/- 1 or 2 possessions) so winning boils down to a battle of efficiency. Who can get the most out of their offensive possessions while keeping the opponent from getting the most out of theirs?

Last I checked, unless there's a Gentlemen's Fighting Championships somewhere (John Nash, any help?), possessions don't alternate when you're locked in a cage with someone looking to rip your head off. Activity matters. It matters to judges looking for effective striking. It matters for knockouts and, to a lesser extent, submissions.

But the Pace Advantage isn't about what matters during a fight, it's about trying to predict fight outcomes before the cage doors lock. Can knowledge of a fighter's pace from prior bouts help us better predict the upcoming one? And which pace metric should we look at: attempts (strikes missed and landed) or only strikes landed?

My fight brain always thought that, from a striking perspective, MMA was about doing damage. Isn't that why we debate whether Michael Bisping really has "pillow fists" or if the two million fighters nicknamed "Hands of Stone" are truly deserving of it?

When we think of damage, while not perfect, it seems it should be pretty closely related to landing strikes. It also therefore seems like that would be the pace metric that matters. Sure you might miss at the beginning of a combination in order to land at the end. You might also miss entire combinations right now in order gauge your opponent's reaction and setup future shots. Against someone who's not a great counter striker, you might strike and miss just to keep them on the defensive, not attacking you. While missing is involved in all three of these scenarios, the end result is always about damaging your opponent in some way, shape or form, or avoiding damage yourself. This suggests that missed strikes only really matter if they lead to landed strikes (or avoiding opponent's landed strikes). In that case, we could just examine landed strikes directly.

"Not so fast," you say. Maybe pure misses can influence those pesky judges?

My prior work has shown that judges tend to give credit to certain types of missed strikes, but I also found that these misses should only cause around 3.4% of rounds scored to change winners, or two scored rounds out of 60 on a typical event card (1 round = 3 scored rounds). This can help to explain certain judging decisions, like Nam Phan vs. Leonard Garcia at The TUF 12 Finale, but is it a strong enough effect to regularly change outcomes?

To change a fight's outcome the 3.4% would need to influence (1) a swing round - a round that changes the bout winner for the judge in question - (2) for a bout that makes it to a decision, and (3) for at least two of the three judges. Missed strikes would then be a factor that explained the fight's outcome. But for the purposes of the Pace Advantage, missed strikes also need to then have decent power predicting future fights - not an easy order to fill.

Numbers Second

Before we get into the numbers, let's start with some statistical basics. Wait, don't leave! It'll only take a second.

If a variable X is correlated with winning fights and we create a new variable Z = X + Y where Y is completely random, Z will also be correlated with winning fights, but to a slightly lesser extent than X. If we were to only examine Z, it might look like Z helps explain why people win fights, but the real power is actually coming from X.

Remember the striking identity? Here it is again.

Strikes Attempted = Strikes Landed + Strikes Missed

(Z)                =         (X)           +          (Y)

In the analyses below, if we see power with the pace of significant strikes attempted (Z), the next step is to figure out if this power is coming only from significant strikes landed (X) or if significant strikes missed (Y) also adds value.

For some breadth, we'll look at eight scenarios for each metric: four different time frames for two classes of fighter (UFC, WEC and Strikeforce and UFC only). I calculate each fighter's pace heading into each fight and exclude draws and no contests so every fight has a winner and a loser. If pace is a good predictor, then picking the fighter with the higher pace heading into each bout should lead to a long-run probability of success greater than a 50% coin flip. If the probability of success is high enough relative to the betting lines, we might also be able to earn a long-run gambling profit.

In the table below, each column utilizes a different measure of pace: significant strikes attempted per minute (Column 1), landed per minute (Column 2) and missed per minute (Column 3). For each column, we always pick the fighter with the highest pace according to that particular metric. The higher our success rate is above 50%, the better. Since we're doing statistics, we've got to watch out for randomness so I color-coded the chart to help us distinguish solid statistical results from potentially random ones. Green results are good-to-go, yellow is good, red is borderline and grey is "NO SOUP FOR YOU!!!"

What do you see? All numbers are above 50% but there's a big difference in their magnitude above 50 and reliability. The best predictions and strongest results come when pace is defined as significant strikes landed per minute. Call it the "Effective Pace." We get reasonably strong results when using attempts instead of strikes landed but we don't gain anything by doing it. It seems instead that the addition of strikes missed gets in the way a bit, with the driving predictive force being strikes landed.

For the gamblers out there, even though knowledge of fighters' Effective Pace (i.e., landing pace) puts us solidly over 50%, if we bet $1 on every higher-pace fighter, we'd lose money over the long run in most scenarios. In the rare scenarios where we make money, we can't be remotely confident it's not due to pure randomness.

Even though Column 3 is completely grey, every success rate is greater than 50% which would seem to suggest that some type of information's contained in the pace of significant strike misses; i.e., that it's not complete noise. It's possible the pace of misses contains two pieces of information: (1) information on fighters' landing pace and (2) some other useless crap.

It turns out that 70% of the time a fighter has a higher rate of significant strikes missed compared to his opponent, he also has a higher rate of significant strikes landed. Another way of putting this is that 70% of the time a fighter has a higher missing pace he also has a higher landing pace.

In column 3, when we only know a fighter's missing pace, the results aren't statistically reliable but they're not completely useless either. Knowing who has a higher missing pace gives insight into who probably has a higher landing pace - better than nothing.

For our last exercise, let's see if we can separate the signal from the noise in misses. Missing pace can be decomposed into a component that's related to strike landing and a completely unrelated component.

So the pace of significant strikes missed isn't completely useless after all. It has value when we can relate it to significant strikes landed. When we can't, it's pretty much noise.

The pace of significant strike attempts is just the pace of significant strikes landed plus extra information on missed strikes. Since landing and doing damage is what really seems to matter for predicting fight outcomes, when we already know a fighter's landing pace, adding misses to get his pace of strikes attempted is pretty much pointless.

Takeaways

If a fight dork is going to tell us a fighter's pace of swinging and catching air or forearm or shin has value, he better have a rock-solid analytical framework. Was Mr. Kuhn's framework rock solid? That's for you to decide.

The Pace Advantage as defined by Mr. Kuhn may be the rate of significant strikes attempted per minute, but Effective Pace as defined by Dr. Gift is a fighter's rate of significant strikes landed per minute. The only reason a Pace Advantage seems to exists is because Effective Pace is driving the result (Z = X + Y).

So the next time some nerd tries to tell us we should fear a fighter's pace of significant strikes missed and landed (and wants to charge us for the privilege), we'll know that our fight brains and MMA analytics say, "Hogwash!" We don't quake in our boots because someone throws a high pace of significant misses! It's the pace they set with significant strikes that actually crash into their target that's no joke.

For more, check out part 1 of the Fact-Checking Fightnomics series.

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.