Author’s Note - This is the third in my string of summer stats columns, regular on Thursdays. Last week, I used spread and standard deviation to highlight the most interesting divisions in college football. This week, I offer some theoretical commentary on the nature of turnovers to lay a foundation for further analysis.
As the summer continues, I’ll wander through some thoughts about football and analytics generally - “state of the game” kinds of things - and some more specific previews looking ahead to this fall, both in the Big 12 and around the NCAA. I’m always open to suggestions about college football data projects, so feel free to reach out in the comments.
A growing sentiment in the football analytics world is the notion that while wins depend on efficiency and success (which is undoubtedly true), the role of turnovers is merely random fluctuation along the way. In his well-thought-out manuscript Football Study Hall, football stats pioneer Bill Connelly discusses the turnover battle as a crucial element in the decision of win and loss in a college game. Buy his book. In a brief online summary, linked here, Connelly laments the transcendent freedom turnovers have, carrying out their own whims despite the best effort of teams:
”...[T]he impact of turnovers is obvious.... Coaches know this, obviously, and there’s almost nothing they can do about it. A team really isn’t in control of this aspect of the game. Sure, quality and style can impact the situation a bit. A better quarterback might be less likely to get his passes tipped. A better defense is slightly more likely to get its hands on opposing passes. An option offense tends to lay the ball on the ground a few more times over the course of a season. But this is only slightly in the control of the player or team.”
Emphasis mine: “Slightly in the control of the player or team.” Remember that phrase.
Connelly comes back to explain the lack of correlation between fumbles forced year to year and concludes that, when it comes to fumbles, “there isn’t much you can do about it.” Now, I’m not here to drag Bill, or even to naysay his work. What I am here to do is to squabble about matters of degree. So, humor me for a moment. Turnovers are only “slightly” under a team’s control. How much is slightly? What percentage of turnover “luck” can teams control with effort, skill, or other unobservable characteristics? If that controllable aspect of turnovers is ~5%, ok, I’ll grant you that it’s probably not worth parsing out. But what if a team could account for 15% of turnover luck? 45%? The role of team philosophy, strategy, skill, and effort in turnover margins warrants further investigation as a potentially unexploited source of marginal efficiency.
While any casual observer admits the stochastic (Author’s Note: Fancy way to say “random” while signaling that I’ve taken a lot of statistics classes in school) element to turnovers, things are less random than they appear. Words like “luck”, “fifty-fifty ball”, and “bad bounce” shape the way we currently think of turnovers. These all, of course, contribute to that stochastic element of fumbles and interceptions; if you’ve held a football, you know its a fickle thing, prone to bounce and roll two a thousand different ways as it pleases. I offer a competing paradigm, but to do so, I have to take a quick baseball detour. Bear with me.
Batting Average on Balls in Play (BABIP) is one of the wonkier stats you’ll find in the sabermetrics world, not to mention that its pronunciation is a mouthful. From the ever-useful Fangraphs Sabermetrics Library:
”Batting Average on Balls In Play (BABIP) measures how often a ball in play goes for a hit. A ball is “in play” when the plate appearance ends in something other than a strikeout, walk, hit batter, catcher’s interference, sacrifice bunt, or home run. In other words, the batter put the ball in play and it didn’t clear the outfield fence. Typically around 30% of all balls in play fall for hits, but there are several variables that can affect BABIP rates for individual players, such as defense, luck, and talent level.”
Cross sport analogies get a little murky for me, especially as I have some convoluted thoughts about other football and baseball metrics. For example: Bill Connelly touts his IsoPPP as a football version of slugging percentage, but in reality, its closer to BABIP. So, I want to set aside the mechanics of the stat, and focus more importantly on its interpretation and what it tells us.
Fangraphs identifies three parts of BABIP: opponent ability, own ability, and luck. We can apply that framework to turnovers. When a player has a high or low BABIP (relative to league average or the player’s career), the usual refrain from analytic folk around the league is “that’s unsustainable.” So the first analytical look at a player’s BABIP is to look for outliers. This applies to turnovers in college football. We look at a teams turnovers compared to league average or year-to-year team rates, and we can pass judgement: this is unsustainable, they are really unlucky, etc.
That is a fine process, especially for a passing glance. The year-to-year comparisons are tricky, as you have to keep in mind the Siddhartha-esque turnover of college football: each year the teams are different, but the same. But as players trickle in and out, coaches may stay, and their attitudes, practice drills, and coaching philosophy certainly can provide a standard for comparison. The next level of turnover analysis Involves controlling for coaches and styles.
As a working example (and I’ll get to TCU next week, I promise) let’s look at TCU’s upcoming non-conference foe Ohio State. I apologize for invoking them; they are the perfect example of a successful team under two coaches with different philosophies.
So, looking at these two groups (and admitting of course these minuscule sample sizes), we can see that Tressel’s Buckeyes averaged about .3 more turnovers a game than Meyers. That .3 can be attributed to the three aforementioned factors of opponent strength, own talent, and luck. But, I’m not willing to concede that .3 is all random bounces - some unobservable characteristic of Tressel and Meyer account for that discrepancy; maybe it’s recruiting better talent, or emphasis on offense versus defense, or maybe it’s the draw of the schedule, getting better defenses at home, etc. It is nearly impossible in broad strokes to tie down all of the variation, but the variation exists, and it’s not because Urban Meyer is luckier than Jim Tressel.
In terms of analysis, the coach control is an important benchmark: If you used just recent history, and saw Meyer’s 2012 1.5 turnovers, you might think, “hey, he’s doing all right, relatively” in looking at Ohio State’s averages. But, when you account for the coach, 1.5 turnovers a game is actually Meyer’s second worst season. The year to year component, while far from precisely predictive, in fact can inform what kind of year a team is having, turnovers wise, and what you should expect from a team, more or less, going forward.
Beyond a passing glance, we have to examine context of turnovers. In baseball, that next level of analysis is quality of contact. Someone may have an oddly high BABIP, but its just because they are consistently hitting line drives or placing the ball where defenders aren’t (speaking to you, Aaron Judge) or they may have an oddly low BABIP because they are getting shifted against their tendencies (speaking to you, Matt Carpenter). While there is some element of predictive regression in BABIP, there are also causal factors driving variation in BABIP beyond good or bad luck. The same can be said for turnovers in college football. Some teams may pass more, and have more turnovers as a result. Some teams may emphasize a pitch or a read, which leads to the ball being more vulnerable.
This week was an exercise in asking the right questions: To what extent can college football teams influence the variability of turnovers via style of play and talent, and to what degree are they random? What do turnover rates tell us about a team’s season performance?What I’ll do next week is compile TCU’s 2017 turnovers and examine them in context of Gary Patterson’s defenses and the plays in which they happened, to continue to parse out some of that causal variation in turnovers.