Better Lucky than Good? Best to be Both

I'm pretty proud of the fact that I have the only DIII football computer rating system that adjusts its preseason ratings for the amount of returning experience each team will have in the upcoming season. I'm fairly confident that returning starter adjustments were a significant factor in my model's success relative to other rating systems in the first few weeks of 2016.

Returning production is only one of the three factors that the best FBS rating systems use to augment their preseason (or in-season) ratings. One that is nigh-impossible for DIII would be recruiting rankings. The only thing I could even think of to do this would be skimming team's twitter accounts for names of recruits, and then checking all-state teams across the country. There's a lot of problems with this (incomplete information, unreliable predictability, etc.) and the time requirements make it completely unreasonable. The other factor though, turnovers luck, should be completely within my ability to calculate, but the NCAA Stats site kinda sucks. While finding cumulative team stats for turnovers is easy enough, compiling team-by-team quantities of total fumbles or total passes defended is practically impossible.

I imagine the concept of luck in turnovers is fairly controversial for a fair number of football fans & coaches. If turnovers are entirely due to luck, why practice avoiding them or forcing them? Why praise players who have a lot of interceptions? Quarterbacks who don't throw them? The simple answer is: the term "turnovers luck" doesn't mean that all turnovers are the result of luck.

Here's the concept, for which I'm borrowing a lot from other's people's work:

- On average each team, offense & defense, recovers 50% of all forced fumbles

- One average the defensive team intercepts one-third of passes defended (for FBS this value is 22%)

- Turnover luck is a function of the random bounce of fumbled balls & the trajectory of tipped passes

- Teams with much better or worse than average luck in one season will regress to the mean in the next season

I hope this removes a lot of the controversy surrounding this issue. It's hard to argue against the randomness of the direction a pointy ball bounces on the ground, in my opinion. A team's ability to force turnovers, therefore, is a function of their disruptive ability. Force more fumbles, and you'll probably recover more. Get your hands on more thrown balls, and you'll probably intercept more of them. I am not arguing against this basic concept of playing football. Part of the reason UW-Whitewater and Mount Union were able to reach so many consecutive Stagg Bowls is because they were (A) really, really good teams that (B) didn't fumble or throw errant passes, (C) got their hands on a lot of balls on defense, and (D) they were lucky after the ball was fumbled or tipped (I put this point last for a reason).

Because of the difficulty of compiling total fumbles & passes defended, I'm going to focus solely on my former conference, the IIAC, to demonstrate the importance of luck in turnovers. I went through the box scores for every conference game and tabulated the total number of fumbles, fumbles lost, fumbles recovered, passes defended, and interceptions for each team.

The first table below is for fumbles. Read the table like this: Dubuque's offense fumbled the ball eleven times, slightly more than average, but they only lost three of those fumbles, much lower than average. Dubuque's opponents fumbled thirteen times, and Dubuque recovered seven of those fumbles. The next column is the expectation--that each team should have recovered half of all fumbles. The Diff. column is the difference between the actual results and the expectation, given the number of total fumbles. The Luck column is the total number of extra fumbles that went Dubuque's way relative to a 50-50 expectation.

For passes defended, it's basically the same table, but instead of the expectation being that half of fumbles should be recovered by the offense, and half by the defense, the expectation is that one third of the total number of passes defended should be interceptions by the defense.

Again, Dubuque was the luckiest team in the conference. Between their offense's passes being intercepted at a lower rate than expectation, and their defense intercepting passes at a much higher rate than expected, they were a net positive of 12 interceptions on the season.

According to The Hidden Game of Football, every turnover is worth 4 points. This is a pretty simple way to look at things, but for my purposes, it should work out just fine. Obviously, field position, down & distance, time, and the current score should all play a role in the expected point value of a turnover, but I'm more interested in the macro view.

So what's the point of all this? Hopefully, an understanding of the role that luck plays in a team's turnover margin should help us get closer to determining a team's "true" talent level. Using 4 points per turnover, and eight conference games for the IIAC, finding the point per game value due to turnover luck is as simple as dividing the total number of turnovers gained over expectation by two.

To go back to my initial thoughts about preseason rankings, I can now combine three different factors to come up with, what will presumably be, more accurate ratings. The first factor is a four year regression of season ratings, next, an adjustment for the amount of returning talent, and finally, stripping away turnover luck from the points per game.

Without accounting for turnovers, Dubuque and Wartburg would project to be the top two teams in the conference, but since both teams had positive turnover luck for both fumbles and interceptions (as did NWU), their projections slip back to third and fourth in the conference. Surprising to me is that Coe, a team who absolutely was lucky to avoid the injury bug at all last season, had negative luck from turnovers, most notable in their pass defense. A team that gets their hands on 53 passes thrown obviously has talent, but apparently catching those balls after they get their hands on them wasn't one of those talents.

It should also be noted that some of the luck, as I've been referring to it, is also a function of talent. Other than Coe, for the most part, the team's with positive turnover luck were also the teams that dominated the all-conference lists. A year-over-year correlation probably exists for turnover luck, but it's probably a very weak one. Even if the correlation was 25% (which is much higher than most studies have shown), that means you could still give each turnover a point value of 3. For a team like Dubuque, while their projections in my model are going to look especially rosy next season (thanks to returning starters), don't be surprised if a few more turnovers go their opponents' way next season, but also don't be surprised if having two returning All-Conference DB's keeps their interception luck in the green.