Belewe Moon: The Prodigal Sun

Submitted by MCalibur on August 23rd, 2013 at 8:08 AM

Everytime the moon shines I become alive. I’m a beast in the night. I’m on the prowl and I hope to find some light.

-Kid Cudi, Alive (Nightmare)

ProdigalSon We all have our on favorite canons, don't we? Once upon a diary in a dark and distant era I made reference to one of mine, that one by Pachelbel;  You know the one. It is inescapable. I mean, et tu Rap? Eminem, Jay Z, Tupac. But more favorite to me than the original is Blues Traveler’s riff on it, Hook. The song itself is kind of a slap in the face to the general public who’s taste in music is, apparently, so trite and unsophisticated that we don't realize that “when I’m stuck and need a buck, I don’t rely on luck” meaning musicians can just hijack Pachelbel and we’ll gobble it up anyway, even if we can’t put a finger on just why we like it.

“Hey pizza guy! Dough, sauce, cheese, and toppings? What kind of cookie cutter bullshit is this???” Relax, Popper, that [stuff] is delicious so what exactly is the problem here?

(Unfortunately for you all the combination of deep night, data induced madness, and alcohol overcomes the “this only fascinates me” levees in my mind and the inmates overrun the asylum so all apologies and thank you for indulging me.)

So why do I like the song even though Popper is being kind of a dick? Because it’s a brilliant monument of knowledge, understanding, wit, and self awareness. And irony. The song is a musical trope wrapped around lyrics consisting of lyrical analysis, mocks its performers, its industry, and its audience and STILL made hella loot. Kharma(n)* is a bitch though: Blues Traveler hasn’t had a song that popular ever since and it was this close to being their most popular single ever.If Penn & Teller did music, they would would do this.

For me, peeking behind the curtain piques my curiosity rather than diminishes my interest. It makes me think “hey, I can do that”  for a couple days until I remember “oh yeah, I lack skillz and talent.” And then I settle into true appreciation and fascination in watching people do things I cannot. The hook brings you back, man. Again and again.

My point in all that was that worthy synthesis is derivative of rigorous analysis. That is, a good place to begin creating something worthy of creation is by understanding things that are worthy of understandation. Follow? UFR, Mathletics, other stuff, they’re all the same—they break things down in order to build other things up. Even if the thing that is being built up is just context, that’s a thing worth building. If you know a bit about something and look at it upside down, sideways, and in a mirror, you might just see something kind of cool…in a nerdy sort of way. ‘Tis the Canon of MGoBlog.

Okay, dayenu.

*MCalidagger said “that’s Carmen!” to Lady of the Lake one time when he was three just seconds after she had stubbed her toe while  placing him in the purgatory of time-out. He was mad, she was mad, it was hilarious. Then I got in trouble for laughing. How is that my fault? Troof is troof, yo.

Here We Glow Again

BlueMoonWolf Another of my favorite cannons (remember: gun == Multivariate Least Squares Linear Regression Model) is what I call the Blue Moon Model. It really began as a basic assessment technique with which to project my team’s prospects with a few simple lower level assumptions. Three years on, I think its worth a remix. Previous foundations are laid here and here.

As a refresh the model takes a team’s Offensive Yards per Game, Defensive Yards per Game, and Turnover Margin per Game and converts that to an expected Win Percentage. IT IS A RETRODICTIVE MODEL so, it’s predictive value relies upon the validity and accuracy of the assumptions that are made. Even when you’re dead nuts on with those assumptions, you’ll be off by more than 1 win about 26% of the time. So, good luck guessing, then good luck winning. That’s the betrayal part of the name Belewe.  This is not a problem though, in the world of inductive logic even though conclusions are only probable, they are useful nonetheless. Also, being able to lay 3 to 1 odds is pretty good. And, guessing aint that hard when you know what you’re doing (upside down, sideways, mirrors…did I mention incense?).

It turns out, the model can be boiled down even further without sacrificing it’s accuracy by collapsing OYDS and DYDS into Net Yards per Game (NYDS). Voila, a 2 factor model with killer statistical significance. The intercept makes more sense too because it is unbiased. Say your team is average (OYDS = 375, DYD = 375, TOM = 0); why should you expect to win an extra 3% of your games? Trick question; You shouldn’t. The best application of this math is to make your assumptions about offense and defense, turn them into an average yardage differential, set TOM to 0* and proceed with your projection.

*For the last time (yeah, right) predicting TOM is a fool’s errand and that's coming from a guy that LOVES trying to predict stuff. Go ahead and try but you’re wasting precious time that could be used to make more worthwhile assumptions.

  3 Factor Model 2 Factor Model

Model R-Square

0.7619 0.7617


0.5319 0.4995


0.0018 x


-0.0019 x


x 0.0019


0.1078 0.1082

I know this model is simple but that’s part of it charm: you can do this math in your head. Take your yardage differential, round by 5, divide by 5, move the decimal two spots to the left, add 50% and ADJUST BY 10 PERCENT (will never get over that) for each net turnover. I appreciate the sophistication of college level analysis but I was way smarter in elementary school. Arithmetic is where its at, homies.

I think there are two main applications of the model: expectation setting and benchmarking. This diary is long to I’ll split the benchmarking bit off into a different diary.

Expectation Setting

All fans want to know the same thing: how good are we going to be this year? Sensibly, we start at the end of last year then plug any holes left behind by attrition and arrive at an expectation of X because, naturally. I have no beef with that process because its a whole lot of fun, but you need to have the right starting point. BMM is handy for this. Here’s how local schools of interest did last year:


OYDS DYDS TOM NYDS 2012 Wins BMM Expect Delta Wins
Notre Dame 412.2 305.5 0.62 106.7 12 10.0 2
Ohio St. 423.8 359.6 0.25 64.2 12 7.8 4
Nebraska 460.8 360.6 -0.86 100.2 10 8.3 2
Northwestern 394.6 378.2 1.08 16.5 10 8.4 2
Wisconsin 393.3 322.6 0.21 70.7 8 9.2 -1
Penn St. 417.5 353.4 0.75 64.1 8 8.4 0
Michigan 383.1 320.0 -0.69 63.1 8 7.1 1
Michigan St. 359.3 274.4 0.15 84.9 7 8.8 -2
Purdue 402.7 416.2 -0.15 -13.5 6 6.0 0
Minnesota 321.4 358.6 -0.15 -37.2 6 5.4 1
Indiana 442.0 463.5 -0.25 -21.5 4 5.2 -1
Iowa 310.4 381.6 1.00 -71.2 4 5.7 -2
Illinois 296.7 387.6 -1.00 -90.9 2 2.6 -1


This year’s prize for Most Dissonant Record goes to: Ohio State. Plus 4, folks. Thirteen years of data has only seen that feat accomplished 8 times out of over 1500 total observations. Fun Fact: that is the third time OSU has managed to post a +4 during that period: 2002, 2003, 2012. In 2004 they posted a +3 followed by +2 in 2005 and 2006…wtf, man?  Tresselball, that’s wtf. Ball Control offense, good to great defense, low risk play calling. Jim Tressel hates math, Q.E.D.

I submit that the extended deviation is the offense’s “fault” because when you have good/great defense, you generate yardage differential by racking up yards on offense. What are you going to do, allow 0 yards per game? So I think the Tresselball offensive philosophy explains why Ohio State consistently defied the math for so long. Once the Buckeyes started stock piling national level talent and opened up their offense to leverage it a little more, their performance lined up with the model just fine. Until last year.


Look, I expect Ohio State to be a formidable opponent as usual but, #2 in the country they ain’t; at least not right now. Well, they are a deuce just a different kind of deuce, nameen? Shout out, to my local head start program. Anyway, Urban Meyer’s Florida teams leveled off at 450 ypg and I think OSU offense will be there this year. Yeah, yeah ESS EEE SEE defenses (!) but OSU’s roster isn’t Florida circa 2008 either. The typical B1G defense is OK all things considered, not great but not necessarily a pushover either. Braxton Miller is good but he has some work to do in his passing game, I need to see it first. I’m sticking at 450 OYDS. 475 is on the table but, show me.

Defensively, nothing has really changed at Ohio State. Coaches, recruiting, philosophy, nothing. Well, tatgate happened. During the tatgate era Ohio State’s defense was insane: 300 DYDS or better, often much better (275 or lower), every year between 2005 and 2010. Then, oops, back to typical (about 325 for them). I hereby grant them reasonable improvement on defense from last year out of the goodness of my heart and they end up at NYDS = 100. That’s 9 wins, with a shot at 10. I’d hate to see them get unlucky, truly.


Meanwhile, in Michigan

I’ll take the more straight forward part first, the defense. Not that its clear or easy just that, because of the reactive nature of defense, I think the best policy is to look at a program’s track record, give consideration to any systemic and roster issues that might exist, and call it a verse.

Rich Rodriguez era notwithstanding, Michigan’s Defense has been pretty consistent by the singular measure of DYDS. With competent coaching and a Michigan caliber roster, we typically hang out in the 300 – 350 zone; last couple of years we were at 325. Now Greg Mattison is pretty good but to start breaking through to the next level of defensive prowess and start heading toward elite, I think Michigan needs more experience and maybe a touch more raw talent. Jordan Kovacs will be missed but Heininger Certainty Principal, jack. I’m sticking with a base expectation of 350 – 325 for that side of the ball. Anything better than that would be kind of amazing.


Offense is trickier, especially with the loss of Darboh. Its no revelation to say that Michigan’s offense should take a step forward this year with more harmony between conductor and orchestra so it’s correct to expect more than average OYDS this year, but how much more? Since 2000, Al Borges has never called an offense better than about 425 (Auburn 2004). Indiana put up a 450-ish in the B1G last year so it’s possible and Michigan has better talent on its roster right this minute than Indiana does, but I don’t think we have an offensive philosophy like Kevin Wilson’s either. And we don't have the talent / experience overall to simply out-execute everybody like Alabama does(450 last year).  Let’s build it up from one more level down just to make sure.

I’m on record for Devin to pass for 225 to 250 ypg this season. The loss of Darboh gives pause, but I’m not backing off on that. So, getting to 425 means we need 175-200 ypg rushing from the backfield. That’s where we were last year with Denard featuring heavily in the run game. Fitz was a different back seemingly reverting back to 2011 form with Devin under center but then there was that leg thing. I’m going to forget about the leg thing and the questions re: the interior line (thou shalt not accuse me of not being generous) and give Fitz 1000 yds on the year leaving about a 100 - 125 ypg gap to get to the desired rushing target. Y’all think I’m crazy but I think we need to get about 50 ypg out of Devin on the ground to get to an offensive performance level that will keep us from freaking out unless one of the other RBs emerge to provide 600 – 700 yards on the year.

I can’t convince myself to go over Auburn 2004 and that’s being generous. What’s the Borges version of HCP? Even without the questions vis-a-vis the running game, going over 425 probably demands Devin the Monster and a Adrian Peterson level recovery from Fitz and Al Borges’s best offense ever. Again, things happen but I’d be kind of amazed if that happened. You can’t outrun your canons; you acquire new ones. That’s possible, but humans are some stubborn mofos. 400 – 425.

BMM says: 8 or 9 wins with a shot to win 10. If you think Michigan can get to a TOM of +0.7, shift your expectations up by 1 win, then go take your meds.

The Road to Indy

Legends division looks pretty tough this year. Nebraska has a lot coming back and has TOM mean reversion working for them. MSU got unlucky in close games and stands to see at least modest improvement on offense to compliment an elite defense returning virtually intact. Northwestern doesn’t really look as good as last year’s record to me and they had a nice TOM working for them last year; they’re on reversion watch. And their schedule is brutal. Still, the Wildcats are pesky.

The two most important games on the schedule occur November 2 and 9 (duh). Win both and we’re probably in the B1G title game. If we split those, we’re likely to be in a tie with Michigan State or Nebraska possibly both going into The Game which we will have to go all out to win.



August 23rd, 2013 at 11:37 AM ^

I don't suppose you could run year to year correlations between teams? I.e. how much better would your model get adding some number of previous years weighted for recency? Obviously this won't make Michigan look any better, but the question of the marginal value of knowing what happened, say, 5 years ago seems interesting to me. Also: obviously this is great and you are a lunatic in the best sense of the word.