8th in the nation is about where Michigan rates every year (+/- 2) in number of D1 scholarship football players. It's usually around 30, although this year should top 45, even 50. Last year was a low - 24, I believe. So that sounds about right.
Brian --
I'm pretty sure that's how the guys over at Football Outsiders are doing their analysis (the value-added approach). It's a better measure of efficiency than what I use (drive efficiency), but I'm pretty lazy and never sat down to try and crip the FO stuff for college football.
Once merely a pipe-dream of mine to approach, the creation of cfbstats.com (we can all thank SMQ for the find) has made the possibility of creating this analysis much easier. The data is broken down by conference and team, with a host of statistical splits.
I think it's just a matter of finding a method to apply the data at this point.
I don't just answer the enternal question. I give hope to the hopeless! Or, in reality, hope to the people who pray for more than 8 wins.
Matt: wow, I'd never seen that site before. Great stuff.
Brian: the damnable confounding statistic in all of this is variance. Two teams that average 4.5/carry and 5.5/pass attempt may not be similar if one of them has a big play passing attack with a 50% completion rate and the other has a short passing game with a 65% passing rate. And the worst part of it is that one isn't necessarily better than the other.
this proposed system sounds also eerily similar to the one used over at protrade.com . It also talks about how touchdowns from the 1 yard line aren't as "valuable" ( much easier to achieve ), than say a 45-yard touchdown play.
That's the beauty of the value-added theory. It provides a vehicle to show that a team need not achieve balance in order to be efficient. That was the impetus for the Football Outsiders approach.
I think the hardest part of creating this methodology is the adjustment. Considering there are 119 D-1A teams playing a myriad of different schedules, finding the proper adjustment rate poses a problem. Simply putting together raw efficiency values is fine, but the real value lies in adjusting it to strength of schedule, etc.
Matt,
Is there a proper adjustment? How much more efficient a team is against the 100th defense than against the 10th defense isn't a constant ... it depends so much on the design of the offense, the players, the play-calling.
For those of us who follow teams that are usually good, we get into the habit of simply throwing away 1/2 the schedule (or more) saying "who cares how he/we did against Eastern Michigan?" Obviously, when you expand to all 119 teams you can't do that anymore.
I think it's best to regard non-conference games as mostly noise since there's no frame of reference, but in a relatively closed system like a conference (or an entirely closed one in the Pac-10's case) I think there's a lot of valuable data there. And you could make a case that ND could be statistically regarded as a member of the Big Ten with their three annuals plus Penn State. And the service acadamies.
Vijay,
I think we're actually approaching this from two different perspectives, and I don't think that's necessarily a bad thing.
The way I look at efficiency is not a scheme-specific or player-specific approach. In the end, whether a team chucks it or pounds it 40 times a game makes no difference; what is the crux here is whether a team is generating a point per 12 yards gained. The New England Patriots are a great example of this (and I think Aaron Schatz would agree).
Where I think Brian is looking to move (and correct me if I'm wrong, Brian), is a player value-added approach tying in player success/efficiency to team success/efficiency. While this is contingent on scheme and offensive design (and consequently independent of an opponent's strength as a team (for better or worse) is going to do what they're going to do), opponent strength does play a role. Maybe the role isn't as important as I originally thought before your response, but I think it plays a role nonetheless.
Notably, a weighted efficiency value based on the strength of an opponent mitigates "wild data." Clearly, Michigan is going to rough up Directional U regardless of what UM does on the football field (from a strategy perspective). With the weighted efficiency values, though, a clearer picture of how Michigan takes care of business independent of strategy is achieved.
In short, I suppose, scheme is vital to this approach, but it is so tied into an opponent's strength (such as Syracuse running the bejesus out of the building while playing Buffalo but having to pass all over the place against everyone else on its schedule because the Orange was so bad in 2005) that omitting a weighted value skews the statistic somewhat.
If I'm totally wrong on this (and there's a strong possibility that I am), then I guess ignoring my above blathering is in order. However, given that most of this stuff is new to football (especially college football), I think it's good that we're all exploring this a little.
Well, Brian just answered the inherent question here: what's the data pool.
With that, an adjustment would seem pretty obscene and useless.
I'm an idiot.
Matt,
Let me be clearer on what I'm saying ...
Take the example of Michigan. Our top two backs are Mike Hart and Kevin Grady, neither of whom has breakaway speed. Neither is ever really a threat to go 80 yards for the TD.
Look at Minnesota, with Laurence Maroney. That kid could go 80 yards on command.
It's at least plausible to suggest that Maroney, with his game breaking ability, would make Minnesota's offense react more strongly to the strength of the competition than Michigan's would. Where Maroney might rip off 3 long runs including 2 TDs against EMU, Michigan is *still* driving the ball down the field in 8 plays.
So forget the play by play or the player particulars, when you look at the end data, you see Minnesota's yards/carry stats being wildly inflated by Maroney's big runs against EMU. If you come up with a deflation factor that works for Minnesota it doesn't work for Michigan and vice versa.
That's why I'm asking (haven't seen the data, so I'm only asking) is there really an adjustment factor?
Vijay,
I think that's much clearer (at least to me).
Unless it is a game-by-game adjustment (which is impossible (in terms of consistency of application) as football provides a myriad of factors that are inherently unadjustable -- weather, stadium, etc.) creating a uniform adjustment appears to be at best difficult and at worst pointless.
With that in hand, I think the true command of a "value-added" analysis proposed by Brian is that it destroys the Maroney-esque skewing of perception. Clearly, weighting the value is not the dispositive vehicle for the chosen end; rather, by taking an efficiency analysis away from accumulated stats and applying it to smaller instances of value -- emphasizing situation and context -- I think a truer understanding of value is discoverable.
As you said, though, working from a purely theoretical approach here isn't solving any of the questions we may have. Only after some data is created will the variance issue you posed be answered (or at least manageable).
For Ohio to produce 78 players to Michigan's 50 is a very significant difference; I'm not sure exactly why you downplay that. Ohio is the seventh most populous state, but ranks fifth in NFL player production; I'd say that IS a pretty impressive feat.
Being ranked eighth isn't so bad for Michigan (actually, it's what you'd expect given that it is the eighth most populous state) but the gap in actual player production between our state and Ohio is pretty shocking - they've produced 56% more players even though their population is only 15% higher (11.4 million to 9.9 million). Given that they have only one BCS conference school and that Michigan has two, it's remarkable that we have such an edge in the all-time series (57-39-6).
As people said, Football Outsiders has a somewhat similar project going. If you want to go through with this though, I recommend looking at Run Expectancy and Win Probability Added (baseball stats). They are both very similar. For instance, with the bases loaded, no outs a team is 'expected' to score ~2.4 runs the rest of the inning. A grandslam would produce 4 runs + .555 expected runs (bases empty, no outs) - 2.4 = 2.1 runs created (not to be confused with the other stat... runs created). A double play would produce 1 run + .387 expected runs - 2.4 expected runs = -1.013 runs created. Using another table, these can also be applied to a win probability matrix depending on the inning, number of outs, lead difference and whether the team is home or away (that table is just a bit complicated).
Given that they have only one BCS conference school and that Michigan has two, it's remarkable that we have such an edge in the all-time series (57-39-6).
I'm not sure anyone has done a study on Ohio and Michigan high school football from the 1890-1918 era, but your all-time advantage in the series comes primarily from that time frame (and MFing John Cooper). But I suspect you knew that already.
Like it or not University of Cincinnati is a BCS team but only so much as any other Big East team is a BCS team...meaning I just felt like being an asshole and saying something about that.
:)
True, Tom. But even if we're only about .500 over the last half-century (which I believe is the case), that would still seem to be overachievement given the two state's HS talent pools. Obviously, we've made up for the difference by recruiting very well nationally.
Hey Hal Mumme was the man. Made a eveyr game of a lackluster Kentucky team worth watching. He never punted on 4th down if he could get away with it. He went for it like 80% on 4th down.
Mumme's former offensive coordinator gets paid thousands of dollars by rich high school football programs to teach his offense.
Commenting by HaloScan
