what's the difference between saying that you should expect to recover 50% of fumbles and saying they are random (coin flips)?
Turnover Analysis–2012 Preview
Random?? If turnover margin (TOM) is truly random, then it would seem a relatively pointless exercise to even track this statistic. So, as you might suspect, my analysis indicates turnovers are not primarily random. Here is Michigan's TOM for the past 14 years.
This does not appear to be simply a random result. And, if TOs are random then RichRod was undoubtedly the most unlucky coach in history! I find it impossible to believe that the worst turnover margin over 3 years was just coincidently the 3 worst years of Michigan football ever.
2011 Review: Michigan went from a turnover margin of –10 in 2010 (ranked #109) to a +7 in 2011 (ranked #25). But, what you might not know, the improvement in turnover margin was entirely due to fumbles and not interceptions. M lost 57% fewer fumbles and gained 186% more fumbles. Interceptions, uh, not so good. Based on the number of pass attempts, M threw 47% more interceptions and intercepted the opponent 18% less.
Fumble Recovery Rates: Michigan had unusually high (i.e. good) takeaway and giveaway % in 2011 which resulted in a net of 9 turnovers in advantage to Michigan versus the typical rates of around 50%. If these rates had been around the expected 50%, the overall TOM for the year would have been negative (-2).
We should expect the fumble recovery rates to return to around to 50% and this will negatively impact the turnover margin for this year. However, the interceptions thrown as well as the passes intercepted were very poor in 2011 and that should be good news for the turnover margin this year. The experience in the secondary should also result in better performance. Overall, I expect the turnover margin to remain very good next year and into the future. Very good teams have good turnover margins.
half a dozen on the other.
His supposition is probably that Turnover Margin is not random but fumble recovery is. If your team doesn't take care of the football, you will have a negative turnover margin, even if you manage to recover half of your fumbles. There are certainly elements of turnover margin that are not random. When most of your hand-offs are going to Mike Hart, you aren't going to be fumbling much.
Yeah, I understand. I just have beef with picking 1 team's data to make a proclaimation that TOM "just isn't random". Maybe we could pick better semantics (unpredictable, etc). A broader survey would demonstrate that having a large positive TOM in a given year doesn't really mean you'll enjoy a large positive TOM the next year. It's been done before. By many people. Including me. See this Diary.
Yes, there's a lot of variation but the central tendency of TOM is to correct back to neutral. Thing is, TOM is retrodictive. It goes a long way to understanding what has already happened but it's -- I'll say it, impossible to predict what will happen next based off of past TOM. Year to Year, TOM is random. That is not to say that the quality of a team/program is random which is what I think the above data set indicates more than anything.
Now, Here's a neat little diddy. Alabama's 2011 defense was, like, awesome, no? Yet, they 'forced' a touch below average turnovers (20 vs. 22.6). Nick Saban: why you no like turnovers? Bama has lost a lot on D, should we expect them to force fewer this year?
What about Oklahoma State, their D Forced 44(!) turnovers last year. That's over 3 per game. IN a typical game a team gets 13 drives. Yo. They have a lot of those guys coming back this year. As a team their TOM was +21. Where will they come in at this year? +15? +10? +5? -50?
Yes, the point is that fumbles recovery may be random -- but only over several years as the data proves! So, in any one year, recovery rates and/or interception % will vary significantly.
And, TOM does not appear to be random even over many years. In earlier posts I have looked at dozens of teams and a large majority of teams have very large TOMs year after year.
For example, USC had an average TOM of +10.2 for 10 years (1999-2008) and had 5 straight years of double digit TOMs (16, 18, 20, 19, 21) from 2000-2005.
Finally, my (our?) primary concern is Michigan football. So, if the TO analysis turns out to be valid for M, that is what is most important.
The USC example is a good example of what I was saying regarding Team/Program quality not being random. That stretch is what happens when you're a dominant power; you kill fools. Sorry brother, Michigan is no where near that level. I expect Michigan's offensive contribution to TOM will be average to maybe a little less than average - Denard passing record indicates a turnover liability. Defensively, I cant see us being anything more than last year, which benefitted from an abnormally high fumble recovery rate.
Michigan's +7 turnover margin last year was only 1 standard deviation from average...that's not a big deal. Go ahead and serve me up some crow pie if I'm wrong but I'm expecting Michigan's TOM to be lower than last year. Not necessarily bad, but lower. If it gets better, it'll be because Denard got way better, in which case...YAHTZEE.
So, right here, right now...Is Michigan's Net Turnover Margin going to be above or below +7?
If TOM is truly random, then there should be a 50% probability that M will have a Negative TOM this year -- and every other year!
But, you say you expect M to have a good TOM this year. That seems to indicate TOM is not random.
My conclusion has always been that TOM is a reflection of performance. Good teams have better TOM -- poor teams have worse TOM. I even believe this applies to individual games -- the better team will tend to have the better TOM (obviously there are games here and there that are exceptions but, overall, the data supports this).
Like I said, I dont really like the semantics, here. I think "random" is a bad word choice for what we're discussing. But ultimately we agree, I guess. TOM is indeed a reflection of performance. Rephraseing, it is not a predictor of performance, it is a consequence of it. Feel me?
By default, a winning record correleates to manageable TOM and, most likely, a positive one (as in greater than 0). I think Michigan should have a fine season but not a great one, I'm thinking 9 wins and a division title, all we have to do is be good enough to beat MSU at home and we're there. It would be hard to get there if our end of year TOM were -12, no? A great season would have us in the Rose Bowl or better, hard to get there with our team's current outlook without a little gernosity from our opponents, wouldn't you say?
So, gimme a end of year TOM of -4 or better, and I think you're fine. At -7 or so, we've probably cost ourselves a game. BY -10...you're probably a bad football team on one or both sides of the ball. To have a great year, with the information we know right now...M need to have a TOM up around +10.
M will have a very good team this year and DRob will show tremendous improvement in interceptions. Fumble recovery rates will be closer to 50%.
End result: M will have more than a +7 TOM in 2012.
I actually agree that we should always expect turnovers to return to the mean, I just thought that your original critique was unfair. His statement of recovering 50% of fumbles doesn't necessarily invalidate the rest of what he said (even though I agree with you, as I said).
Go back thorough Greg Mattison's coaching career and look at the turnover margin of every single team he has coached. You'll see a lot of plusses, and no minuses (at least none as far back as I looked). There's not even many small plusses (plus two or something).
I for one believe that turnovers are not random, but are in fact mainly a result of coaching style. Take Mattison's: Very aggressive, very blitzy.
How does this result in his positive turnover margin? Well if you're sending more people on the blitz you're probably putting more pressure on the quarterback and forcing him to make reads faster and under more strain. Thus, a higher possibility of a duck that results in an interception.
Heck, I'd even suggest that an aggressive style can result in more fumble recoveries. If you have a blitz on, you probably have more players in the box than normal (or at least, more players close to the LOS). This can give you a numbers advantage in covering the ball. Sure, the football may bounce random, but if you have 6 players near the ball when it comes loose and the other team has 5, you have a higher chance of getting it.
That's not even to mention how I'd postulate that better defenses in general, regardless of playing style, are more likely to have a + turnover margin (just like better offenses are more likely to not turn it over). If you have cornerbacks that can blanket a reciever, you're more likely to get interceptions. Better defensive line? Hitting the running back in the backfield forces more passing downs, and is also probably more of a surprising hit for the back resulting in a higher chance of fumble. Then, of course, sacks can cause fumbles a lot of the time and a better line will probably get you more sacks. And so forth.
Looking at the FBS as a whole might be random, but that's akin to saying yards per game are random because they all average out to be around 250 yards per game and anything over or under that is just dumb (un)luck.
You bring up good points, and I dont refute any of it. But, defense is only half of the equation. Greg Mattsion can't do nothing for Denard's accuracy issues. Maybe Denard gets better, but it wont be because of how often Mattison blitzes. Taking it to an unfair extreme, why dont we blitz all the time, wouldn;t we like to force a turnovers on every play?
You're right in that you need to do the study properly. You need to look at what are known as matched pairs: how one team compares to itself, from one season to the next. Otherwise, you're not studying what you think you're studying. What I'm saying is that when you do that across the board no trend exists. In fact, the only trend that shows up consistently is that of mean reversion (high numbers tend to fall, low numbers tend to rise).
Sound football (offensively and defensively) results in good TOM. Simple as that. Is Greg Mattison good at getting his defenses to play soundly? Sure, I'd go with that. But you give him Denard 2010 throwing the ball and he's not going to give you a positive TOM no matter how much he blitzes. To get a big number in TOM (+12) you need a good team and a little luck.
MIchigan's +7 from last year isn't that high. But there's some dubious stuff embedded there. First, We recovered fumbles at an atypical rate. You wanna bank that we'll repeat that recovery rate? Be my guest, but I'm expecting 50-50. Second, Denard Robinson still has accuracy problems until he proves otherwise. Those two things, indicate that our TOM is likely to be lower this year. Will it be -7? I'm not going that far. But +7 is a ceiling moreso than a base expectation. Anything in between is ho-hum, no big deal.