Sometimes You Eat The Turnover Bar

Submitted by Brian on March 28th, 2012 at 1:24 PM

Turnover margin is a notoriously jittery stat that does not often repeat year to year. Turnovers are infrequent and hugely impactful, so they tend to wander all over the place without much rhyme or reason, slaying or rescuing seasons. Yes, there are obvious repeatable factors that correlate with good or bad turnover margin on a macro level. Get to the quarterback and he will explode in a confetti of bad decisions; allow the opponent to get to yours and watch the same thing occur.

On a micro level, sometimes you eat the bear… sometimes the bear eats you.

Michigan ate the bear last year, recovering around 75% of available fumbles. I know people want to believe there is a narrative that supports this model of football. When I returned from the Mattison coaching clinic presentation, one of the items I mentioned was that Michigan treats all incompletions as live balls in practice. I didn't think that was much of an explanation but a lot of commenters seized on it.

There does not have to be a reason that events transpire. It's not an Eastern thing. Sometimes you eat the bear. Sometimes the bear eats you.

Michigan ate the bear last year, and how. SBN's Bill Connolly put together a stat called "adjusted turnover margin" that assumes an NCAA-average fumble recovery rate (50.3% for the D) and NCAA-average PBU-to-INT ratio (21.9%), compares it to your actual turnovers gained, and calculates a points per game figure Connolly figures is just the bounce of the ball. Drum roll…

Five Teams Who Benefited Most From Turnovers Luck
1. Michigan (+3.97 points/game)
2. Maryland (+3.97)
3. N.C. State (+3.61)
4. South Carolina (+3.61)
5. Oklahoma State (+3.40)

I am Jack's utter lack of surprise. Michigan's overall fumble recovery rate of 75% was first in the nation by a whopping eight percent! (Maryland was #2 at 67%.) They were three standard deviations above the mean! They were a full standard deviation above the #2 team in the country! This aggression against regression to the mean will not stand!


This is the point where I talk about how lucky Michigan got last year and a lot of people say "nuh-uh." This gets a little frustrating on both ends. I get frustrated that something like that Sugar Bowl doesn't bring the point home; people who disagree with me get frustrated that I'm downplaying Michigan's success or being grim about next year.

They're not entirely wrong. I do think that if you replayed the 2011 season 1,000 times Michigan ends up 11-2 in relatively few of them. They were only sort of close in one of their losses*, won two-and-a-half nailbiters** and acquired 10 more turnovers than Connolly's model expects. Michigan also had the benefit of a soft schedule (no Wisconsin or Penn State) in a down Big Ten and an Ohio State team in shambles after tatgate. It was pretty uninspiring in terms of 11-2 years featuring wins over ND, OSU, and a BCS opponent despite undergoing massive transition costs and operating with a slap-dash, attrition-ravaged roster.

Which is to say: WOOOOOOOOOOO. Yes. Score.

But once we get past the woo and start talking about setting expectations for year two we should not base it off what Michigan did last year but what they should have done, what they lost to graduation and attrition, who they return and add, and who they play. We should start with the premise that Michigan was super lucky last year and probably won't be this year.

This doesn't mean their turnover margin is doomed. It just means their turnover margin is doomed unless Denard Robinson becomes a lot more responsible with the ball. Michigan got away with being –7 in interceptions because of the fumble surplus. That covered up a lot of blemishes last year.

What should we expect Michigan's turnover margin to be next year?


Happy Arguments

I am arguing it will be worse. I made similar arguments for much of the Rich Rodriguez era when Michigan was hugely negative every year and dammit it never changed. 

Experience at quarterback. This is a double whammy to the good for Michigan: they've got a senior starter entering his third year and—even more important—his second year in Al Borges's system. A number of Michigan's turrible interceptions a year ago came paired with hand-wavingly-open receivers Michigan's quarterbacks just missed, like this one Gardner chucked against Purdue:


The ball is in the air here, but it's going to the double-covered Gallon instead of the hand-wavingly open Junior Hemingway. This wasn't pressure—Gardner had all day—it was a huge coverage misread. In year two these things should significantly diminish.

Fitzgerald Toussaint could be Mike Hart-like. IIRC Toussaint has not fumbled as a Michigan ballcarrier. As carries move to him from other sources—largely the fumble-prone Denard—Michigan should reduce the number of fumbles that can go against them. Fumble prevention/extraction is a skill.

The defense should be sack happy. Michigan finished 29th last year without getting great production out of its three-tech or weakside defensive end. Will Heininger had one sack last year; Craig Roh and Jibreel Black combined for 5.5. If the moves of Roh and Black inside upgrade the pass rush at three positions, the blitz-mad Mattison D will be in QBs' faces even more than they were last year.

Complicating factor: Mike Martin only had 3.5 sacks last year but his disruption opened things up for other people.

Protection should be good if the line is healthy. Lewan is an all Big Ten left tackle (at least) and Schofield is a touted recruit with a year of quality playing time under his belt with all the tools to pass protect on the edge. Wicked blind-side hits on Denard should be rare.

Sad Arguments

Denard is just turnover-prone. This has been a fact by air and ground ever since he hit the field. While he's going to improve with experience, sometimes players never have that light bulb pop on. Toussaint will shift some carries to his five points of contact but Denard will still get a bunch of carries, and he'll cough the ball up some.

Chucking sure interceptions up to Hemingway will result in interceptions because Hemingway is gone. Unless these are going to Gardner.

Hello massive reversion to the mean on fumble recoveries. If Michigan recovers over 70% of available fumbles this year I'll eat a lemon. Probably at the Rose Bowl.

If a tackle goes down, yeesh. Everyone's assumption is that this would see Kyle Kalis step in at right tackle. Mega-hyped recruit… and a true freshman.

Seriously, Denard is walking variance. I think Michigan will preserve its fairly positive TO margin. If they don't, we will all be sad about Denard's inability to shake the turnover bug. I can't predict he'll be better or worse until we see him play.

There's a reason a couple departing seniors picked Robinson—who was an All-American as a sophomore, remember—as a "breakout player" in that Rothstein article from yesterday($). He could break out. He could run in place, and not know which it will be makes predictions here even more useless than they have been in the past.

You may now return to thinking about Taylor Lewan on a tandem bike.

*[Even if Michigan does score against Iowa they have to get a two-point conversion and then win in OT, which is like a 20% shot.]

**[OSU should not have been since there was no reason to overturn the Toussaint TD that would have ended it.]



March 28th, 2012 at 4:31 PM ^

Last season's turn around was amazing, and under normal circumstances we should not use that to set this season's expectations. What's normal? 

It's certainly true that team 133 could improve in all aspects of play and still drop four games. If that happens we will find joy in their improvement regardless of the number of wins. 

The coaches and the players have told us that their expectation is to improve at every position. We've been told that their expectation is the Big Ten Championship. Maybe early in 2011 that seemed like "coach speak", but after last season we know that Hoke & co mean every word. 

Maybe a normalized expectation is for 8 or 9 wins, but  I must admit that I do expect them to win ten or more games this year.

First, since his arrival Borges has maintained that quarterbacks always show a dramatic improvement in their second year in the system. If he was not meeting that expectation I can't imagine that Borges would be telling us that Denard is the starting QB. My expectation is that, barring injury, Denard will have the break out year we've seen flashes of all along.

Second, while some key players have moved on there is plenty of talent left and these coaches have already demonstrated the ability to adjust thier schemes to maximize the skills and talents that their players bring to the field. I expect Mattison's defense to be better than last year. If they regress in fumble recoveries, they'll improve elsewhere enough to more than compensate. Mattison will still have to set his defense based on the offense he expects to face and he may get out guessed like he was in the Ohio game, but just as last year that means an ugly win not a loss. 

Third, these players know they can win. Last year Mike Martin said "We didn't know how a championship team functioned". Now they know, they've experienced it and believe in themselves each other and their coaches. Martin doesn't have to drive over to someone's house to pull them out for a voluntary pratice, they all show up on their own. 

Last, this coaching staff is committed to getting each player to achieve their absolute best and have proven that they know how to do that and accept nothing less.

Last August, I expected to win 8 or perhaps 9 games but I was unable to pick which games I thought we'd lose. Some were clearly going to be tougher but not a single game seemed unwinnable. That is how I see this season. We face some tough games, having Alabama as our opener defines a tough start, but can any of us say that we view that game as unwinnable? We will undoubtedly be underdogs, but it's Alabama's opener also. Any reasonable assessment, at least at this point, must view that as a winnable game. If that game is winnable every game this season is too. 

Is the expectation that we win at least 10 games this season unrealistic? In 1997 we were ranked 14th in the preseason by the AP. Various pundits were projecting 9 wins in part because of a very tough season. 

This is Michigan fergodsake!

Go Blue!

Blue boy johnson

March 28th, 2012 at 6:00 PM ^

I am always amused that such a mundane subject, (regressing to the mean in regards to turnovers), or whatever the fuck it is we are talking about here, is such a completely riveting topic to Brian. Never in all my life did I expect someone to dedicate 27 articles to this sleep inducing subject. Hey, whatever floats your boat.


March 28th, 2012 at 7:47 PM ^

Cannot wait to see a more confident Shoelace throwing the ball around this year.  If Michigan can get the same amount of takeaways and cut down on the interceptions this year, they are gonna be scary good!


March 28th, 2012 at 8:27 PM ^

offensive fumbles are a bit of a crap shoot, but I think that defensively you can point at hoke's emphasis on fundamentals to explain the higher recovery rate.  It was in one of the videos of the coaches clinic, if you're pounding it in to every kid's head that he has to keep separation from the defender and an eye on ball, that's going to improve your odds of recovery. 


It also helps if you've got the ball carrier completely surrounded.  Pictures like this were not rare last year. 


March 28th, 2012 at 10:15 PM ^

Repeating myself here, but guys who I have talked to that played football at the college level laugh hysterically at the idea that forcing or giving up turnovers is luck.

Their reasoning makes sense: On each play in a real football game there are multiple instances of man to man violence and critical technique happening. The angles and impacts are subtle to basically unnoticeable to all but coaches and players. This mostly unquantifiable minutiae is what breeds turnovers for and against. 

Guys get bigger and more mature and this can help. Guys get more experienced and the game slows down for them and this can help. Superior coaches with a greater attention to detail, scheme, and technique take over and this can help.

Brian knows the game better than any non-participant or coach I have ever seen, but there is a lot numbers just can't account for in the game of football.

El Jeffe

March 29th, 2012 at 10:17 AM ^

It's a reasonable point; however, I think the argument is not so much that in a given year a team will always regress to the mean in terms of causing or preventing turnovers, because some dudes or groups of dudes are just better than other dudes or groups of dudes (see: Woodson, C.). Rather, the main argument, I think, is that the serial correlation of turnover margin is notoriously low.

So what may very well happen is that next year M may have a totally pedestrian turnover margin and some jackholes on the internet will say things like "Wow, I guess Mattison is starting to lose his touch," or "Wow, I guess Mike Martin and RVB were more important than we thought." Both of which may be true, but they ignore the Occam's Razor explanation, that turnover margin simply has a large stochastic component to it, and therefore does not correlate well from year to year.

And then we'll get three more posts from Brian about this.


March 29th, 2012 at 11:41 AM ^

The term that should be used is "not repeatable", which isn't quite the same thing. And even more specifically, it's "not repeatable at the D1 level."

If you have Michigan play against a high school team, I'd very much so expect UM to recover 70%+ of fumbles. And if you had UM play against an NFL team, I'd expect the ratio to be reversed. It's a skill. I don't even think Brian disagrees with that. However, it's a skill everyone who plays at the D1 level is exceptionally good at, so there's a bit of a standstill. 

If every D1 football team played 162 games a year, you wouldn't need to regress turnover rate quite so strongly. However, since turnovers are rare, and everyone UM plays against is pretty good at recovering them, it's hard to tell much from which way 30 or so bounces went.