Short Changed Hero

[ed: bump]

I can't see where you’re comin' from / but I know just what you’re runnin' from / And what matters ain't the who's baddest / but the ones who stop you fallin' from your ladder.

-The Heavy

THIS AINT NO PLACE FOR NO HERO TO CALL HOME

For a little over four years now I’ve had a summer time hobby of trying to predict plausible performance levels from various QBs for the upcoming football season. I have tried to root these projections as deeply into the bedrock of reality as is possible for a figment of one’s imagination and at this point there is a codex of sorts in the diary archives describing my methods. It’s fun to go back and see what worked and learn from what didn’t. There’s something there, man.

For Devin Gardner 2013 I laid out two stat lines hinging on two sets of assumptions—a reasonable/prudent set, and a ‘sexy’ set. The reasonable prediction: Gardner would complete 225 of 360 passes for 2900 yards, 23 TDs, and 10 INTs. In reality he went 208 of 345 for 2960 yards, 21 TDs, and 11 INTs. There’s a HEAVY dose a good fortune involved there but, hot damn, that’s pretty good. The assumptions here were basically looking at only QB stats and nothing else Devin had shown enough in his 5 QB starts during the 2012 season to perform at the “seasoned veteran QB” level which I think of as an incumbent with 2 years of experience in tow. That's a brutal benchmark, IMO but that's what I measure guys up against. That's what we want them to be.

Anyway, the sexy set of assumptions were:

1. Devin has elite talent. I believe this one held. More on that later.
2. The O-line would be fine despite the possibility of being “a touch weaker than last year (2012).” Eh boy…
3. The offensive scheme would be well tailored to Gardner’s skill set and that of the support around him. This was sometimes true but not consistently often enough for Borges to keep his job.

Ok, so the necessary assumptions for DG to be the second coming of Vince Young vanished into the ether. But those last two assumptions about the support and scheme are really kind of baked into the reasonable prediction too. For my money, the fact that DG put up the numbers he was able to in spite of the glaring flaws of the team is a testament to just how good he can be if the conditions are reasonable.

The fact that there are so many straight-faced questions being asked about Devin Gardner’s incumbency status is ludicrous. Sure, numbers don’t tell the whole story but they tell a good part of it. DG went from being one of the darlings of the 2013 Manning Passing Academy to needing to prove his talent simply because he couldn't compensate for all of the flaws around him last season. He did as well as could reasonably be expected without adjusting for other very real headwinds.

[After THE JUMP: Gardner under the microscope.]

Sandcastles Are Destined for Destruction

Having said all that, there is obviously significant room for improvement. Let us decompose the white rainbow:

One of my fundamental hypotheses is that Completion Percentage and Interception Rate are mostly attributable to the QB’s Skill level whereas Yards per Attempt and Touchdown Rate are attributable to System and Support in addition to player skill. So, while Devin had a very good overall passer rating in 2013, it was bolstered by a abnormally strong YPA. The other three components lagged the rating and the two most attributable to Devin’s skill (according to me, let’s talk about it) were way off of where we would expect them to be given 11 years worth of data to compare against.

I find all of this somewhat confusing and basically contradictory to the things I actually believe: that Devin is capable of elite play if he were privy to resources (supporting performances, sound offensive system) that would enable him to perform that way. Yet last years data show the things he is in sole control of (decision making and accuracy) weren’t where they “should have been.” What gives? Enter another hypothesis: player skill does not regress; you don’t suddenly forget the things you’ve learned.

Comparing 2013 to 2012 reveals that Devin in fact improved his skill level as quantified by Cmp % and Int % where as the Support and System around him (YPA, TD%) declined dramatically. We know this by simply having watched the games, but its nice when the objective measurements validate our subjective observations. Reconciling all of the available data suggests, at least for me, that Devin did in fact improve despite lacking a running game and dependable pass protection but the interaction between system and support attenuated our ability to detect that improvement.

Once More Unto the Breach

My rubric for establishing an expectation for future QB performance hinges on 4 main factors: Skill, System, Support, and Statistics. Also schedule. Five, there are 5 chief weapons I use to project QB performance… /Spanish inquisition. Where does Devin stand going into 2014?

Skill

If its not yet clear, I think Devin is every bit as talented as he needs to be in order to be a absolute monster. Notre Dame can testify. With renewed health, additional experience, and actual technical development with a position coach and everything, I expect to see the best Devin yet from a skills standpoint in 2014.

System

Borges got fired, so that says something. But, different isn’t always better. The thing that frustrated me and a lot of others lat year was the seeming incoherence of it all. When it worked (Notre Dame, Ohio State) it was awesome, otherwise we had nothing. Installing new schemes or whatever seemingly from week to week worked for a minute and then it didn't. At the very least I think it’s reasonable to expect that Nussmeier will at least bring coherence. But its hard to expect the team to master a new system the first time out.

The system also benefits from the quality of the players in it which brings us to…

Support

I break this up into 2 subgroups*: Skill Support (RB, WRs) and Shield Support (the line). 

Skill Support: Normally I get worried when we lose a player as productive as Jeremy Gallon was but, I kinda think we’re stacked at receiver. Funchess, Darboh, Chesson, Norfleet, Canteen and friends. I’m eager to see Chesson and Funchess blocking downfield for Norfleet on a little slot screen deal. That sounds like fun.

As for the running game, I’m pretty stoked to hear that Deveon Smith is distinguishing himself as the guy at RB. I’ve had a diary in draft for while that tries to objectively compare running back performances in a similar way as passer rating does for quarterbacks. That work suggests Smith performed better than Green last year. I’ll try to get that diary out in the not too distant future so you can see what I’m seeing but let it suffice to say that I’m buying the Smith hype. This is not to say that Green can’t or wont be good, just that I suspected Smith was ahead of him based on the numbers.

Michigan is better off in this area in 2014 than it was in 2013.

Shield Support: This would be the $64,000 question. I’ll admit that this is an area I know little about. The quality of offensive linemen quality is hard to measure. I've done a some work that shows that, in general, having a lot of returning starts on the Oline is a good thing (duh); the problem is that Michigan doesn’t have a lot of returning starts on the Oline this year. Its actually worse off than it was as a group going into 2013. Then there’s the issue of losing two NFL worthy tackles, one of which was an elite prospect.

Its hard to make that situation look good, but the argument I’ll put forward is that replacing two starters on the line is better than replacing three. Additionally, the replacements last year were mostly redshirt freshmen and had zero in game experience, the stables are stronger this year.  From this view we are better off than last year. I’ll admit this is largely wishful thinking but I think its at least plausible. I’m going to go ahead and say that the Oline will improve from terrible to mediocre despite the loss of Lewan and Schofield. Hooray?

*Six! Our six principal weapons are….

Statistics

This is just another word starting with the correct letter so that I can keep my little S thing going but I’m referring to regression to the mean here. Its pretty basic stuff if you believe in math but I go about applying the concept in a way that violates what I was taught was proper in school. So here’s another breakdown:

These three charts show the progression (as you proceed in clockwise fashion) of what happens to the data as I manipulate it to get to the point where I pick off my adjustment factor. The top left is the raw data for a bunch of returning players and how they’re rating changed from where they were the previous season. There is a nice trend but its still pretty weak (R_squared = 0.31) for the purposes of prediction. In the second chart I have binned the data along the x-axis which has no real effect on the trend (R_squared = 0.30)  but the magic happens when progressing to the third chart. There I have averaged each bin to reveal its true central tendency which allows the strength of the trend pops (R_squared = 0.94).

Here’s the issue that I’m attempting to address. In the first chart, each data point has equal influence on the least squares regression. The problem is that the points that fall way off the trend have a lot more leverage than the points that are close to the it even though the high leverage observations are vastly outnumbered. We all know how important sample size is to data analysis so why would we ignore this knowledge?

A different way to explain the logic at play here is that each bin represents a different normal distribution of real observations at a particular point along the trend. The changes that occur for players that post bad ratings manifest for different reasons than those that occur for players that post good ratings. The bin averages allow us to connect the peaks of each bin distribution.

I’m not a mathematician so I can’t provide a proof that this is a legitimate manipulation of the data. Maybe one of you is smart/educated enough to prove/disprove the concept. I am not.

Anyway, the final equation is  y = -0.7449x + 97.22 and Devin’s current x is 146.1. So based on reversion alone we’d expect his 2014 rating to drop 11.6 points to 135.5.

One last not about this finding: basic algebra shows that 130 is the Mendoza line for passer rating.

Schedule

This is diary unto itself but looking at the defenses Michigan will be facing in 2014 there is 1 scary defense (MSU), 2 decent ones (PSU, Utah), and 8 defenses that mediocre or worse. Some will go up others will go down but its hard to expect that we will end up facing a defensive slate that is all that difficult. This should be a favorable schedule to put up good offensive numbers overall.

All told I think the Devin’s rating will tread water but he should be able to post a “legit” 145. That means:

The improvement in interception rate is way too aggressive so I’m overriding that to be a more modest 0.028. The YPA seems low to me as I expect Nussmeier to leverage the receiving corps, a presumed strength, to protect the offensive line, the presumed weakness. So I think YPA will be a touch higher, say, 8.3 ish. Quarterbacks under Nussmeier as OC have averaged about 28 attempts per game which translates into about 365 attempts over 13 games.

The critical assumptions here are:

1. The receiving corps will be a strength
2. Devin will show modest improvement in accuracy (completion percentage and interception rate.
3. Nussmeier will leverage assumptions 1 and 2 to create a pass happy offense.
4. The offensive line will be good enough to allow 2 and 3 to manifest.

I still think Devin has a the potential to be an absolute monster (160+) but these numbers are already very aggressive and the team needs to overcome significant issues to make them real. Let’s hope the ones keeping Devin on his ladder are up to the task.

And you feel like you feelin' now / Doin' things just to please your crowd / Well I love you like the way I love you / and I suffer but I ain't gon’ cut you 'cause…