Awesomely Alliterative Answers, Ahoy!
to play football, not to play trumpet
The most persistently wrong thing I was wrong about amongst the many persistently wrong things I asserted about Rich Rodriguez and his bite-sized Era at Michigan was: "that turnover margin is going to be a lot closer to zero this year." Or words to that effect.
I should have been right, or at least in the general ballpark of right. Rodriguez's West Virginia teams were consistently in the black. Turnover margin is so weakly correlated from year to year that Phil Steele annually puts out a "turnovers = turnaround" post highlighting teams with double-digit swings in either direction so that he can predict against the teams with big numbers and in favor of the teams with little tiny ones. Here's your favorite team:
And here's how that worked out:
Amongst the many things that got Rodriguez fired, the persistently huge negative turnover margin is neck and neck with transferpalooza and program alum white blood cells for second place.
So these days, running across stats like these…
running the correlation between one year's turnover margin and the next, I found that the correlation was a mere 12%. That's still something, but it's clear that for most teams, the turnover margin they enjoy one year has virtually zero predictive value for the turnover margin they will enjoy the next year. That means that on average, teams with substantially positive margins will see major decline in margin the next year, and teams with substantially negative margins will see major improvement the next year. A team with a -10 turnover margin in 2009, for example, would have an expected turnover margin of -1.2 in 2010, an improvement of nearly a full turnover per game!
…make me want to bang my head against the wall. Of course Michigan would be as large of a chunk of that correlation as possible, and of course they would be on the negative side of things. Thump. Thump. (The only thing worse than defying this correlation is defying the correlation between turnovers and wins: GTP points out Georgia went from –16 to +10 and still finished two games worse than they did last year. That would seriously harsh my buzz if I was a Georgia fan.)
I have tried to make the world make sense and this is what I've come up with:
So while year-to-year correlation is low across college football, if you correct for experience—especially at QB—and maybe lack of prominent walk-ons/converted WRs in the secondary that 12% would be significantly larger. Michigan's program got so messed up that they stopped participating in the circle of turnovers*. Instead they laid at the bottom of the national rankings, a corpse dragged down by redzone interceptions.
Um… so… there's the above theory. And then there's Denard Robinson, Michigan's first returning starter at QB since Chad Henne. And then there's Greg Mattison and a defense that uses Craig Roh as a pass rusher instead of a clunky linebacker. There are a bunch of returning starters everywhere, including four guys on the offensive line.
I'm going to be the guy who puts his hat on a stick and pokes it out of a trench to see if there are any snipers around, again: Michigan's turnover margin should scream towards zero this year. They've got gravity on their side and many things besides. Also, Brady Hoke's miraculous digestive tract will move all that Tyler Sash wackiness to Ann Arbor.
This is the year Michigan has a mediocre turnover margin. Believe.
Awesomely Alliterative Answers, Ahoy!
Wrong site for this quote.
"I don't care what the numbers say"
So despite all the data you provide the career of Mike Hart as proof that years of data are garbage? I can listen to your argument if you provide me with something other than Mike Hart managed to hold onto the ball a lot other than when he was in Florida. While Hart was very good at ball protection let's not forget he was very for fortunate to not have several more turnovers in his career. You have some decent points, but how much weight do they have?
1. Ball security techniques- Mike Hart and early Tiki Barber were quite a difference but they were extreme outliers most backs are taught proper techinique and fumble about the same. ...5%
2. Play calling/offensive scheme- Other than option football which tends to have more fumbles I think this has been proven to have little effect...2%
3. Offensive Experience- More experienced qb's throw less picks, better offensive lines provide less pressure which equals fewer fumbles and ints. 15%
4. Defensive scheme- More aggressive defensive play calls can lead to more turnovers, but most teams do a similar amount of blitzing in the end..2%
5. Better teams- If you have a superior team you tend to generate more turnovers because you are ahead most of the game and are not forced into mistakes like the team with inferior players. 15%
6. Luck- 61%
Ok, let's just make up some numbers that help your argument.
The point is not the percentages. The point is that I believe luck is the overriding factor when discussing even teams and that he said despite all the data he rejects it.
A much more effective method would be to take 1 player and use him to disprove all of the data on the subject. Right?
Never said that. I just happen to agree that turnovers are not random... they just depend on so many variables that they might as well be random.
I agree with your theory on why turnovers happen, and without watching a full season on a particular team, you can't determine what percentage luck had to do with the previous season's numbers.
But, I do think you can take a previous season's turnover margin and put in context with what else you know and generally figure out if there is some sort of regression in the future. For instance, if it was a very young team, especially at QB, it is likely that the team will have better "technique/strength/experience/etc." the next season and thus a better TO margin. Or if a team won a ton of close games and also had a very high positive TO margin, their TO margin may have had more to do with luck than their "technique/strength/experience/etc." and a regression might be in the future.
The reason college football sees "randomness" in TO margin from year to year is that your first two categories are so fluid from year to year. Very few teams maintain the same levels of skill and experience from year to year, thus your TO margin adjusts accordingly. Throw in the handful of teams that do have an inordinate amount of luck, one direction or the other, as a whole, TO margin will generally mean little as an indicator for the next year's team.
I think getting more pressure on the QB definitely will help out the TO margain, and hopefully Mattison is getting the guys up front ready for that task. I honestly can barely remember what it feels like to jump up and cheer for a sack by a Michigan defensive lineman/linebacker. It's been far too long.
I devoted a large portion of a diary I wrote late last summer on turnovers and mean regressions and so on. I too was left with a sooty face and singed eyebrows.
Oh well, this time it will be different!
Nicely done and I really like your list of explanations. Some additional data that could be interesting:
It'd be great to see some context for that 0.12 correlation. I wonder what the within-program year-to-year correlations look like for, say, sacks or rushing yards per carry. I suspect that they're all pretty low for some of the reasons that you mention here (about the natural change in programs from one year to the next). It makes sense, though, that the turnover correlation would be lower than most of the others.
Also, seeing the difference between the interception and fumble correlations might be nice. I could imagine some features of programs that consistently lead them to having high or low interception numbers (e.g., type of offense run, preferred risk aversion of QBs recruited, etc.), but it's harder to tell a non-randomness story for fumbles.
Check out the diary I linked to above, you might dig it.
San Diego State was -10? What does that say?
I hear they've got a new coach this year, so we should be able to take advantage of that situation when we play them.
"Denard is the first returning starting QB since Chad Henne"
go crosseyed, so this discussion is interesting, as are the barbs between you when arguing about the analyses. Here's my prediction: Michigan has a better chance of winning if they reduce their turnovers. You can quote me on that. /s
Without a doubt denard was counted on to run way too much last year. And even though he is a great athlete in terrific shape, you could see he was getting tired on some drives, this may have also affected some of his bad decisions. Hopefully this year that wont be a problem.
SDSU went from bad to worse in TO margin in Hoke's first year, for whatever that's worth.
of why i won't be buying phil steele this year. 7 point font, single spaced, wordart, clipart and every single gimmick employed by middle school lunch ladies to jazz up their menus, you know, for the kids.
/i will probably still buy it
Statistics are like end of the world predictions. Eventually one day they'll be spot on, but not today.
But, this Saturday, I'll be wearing my tin foil hat. And in the remote event we get past Saturday, then we have the end of the Mayan cycle on 12/12/12. And in the highly unlikely event we get past THAT, we've got global warming! What's a Chicken Little to do???
engineering theorems and stats can't always predict football outcomes!
Science should NEVER fail, but I love it when it does. Just watch the games. Enjoy them. Revel in the fact that nerds consistently fail at predictions and that bookies consistently succceed.
Its sposed to be FUN, not math homework. Put a spike into that wall you bang your collective heads on---that would at least make the attempt colorful.
all know. Brain knows. He even touched on it a little bit. Defensive pressure on the QB is huge. My friends and I watched in horror the last 2 years as the D let the opposing QB sit in the pocket for hours. I couldve carved up Gergs D. That will change. Along with the returning starter at QB, I think the t.o. margin is almost a lock to be zero or positive.