Big 12 Games Preview - 10/1/11

by Huck L Berry on Sep 26, 2011 1:46 PM CDT in Texas Longhorns and Other Football

I'm going to try to post a look at the following weekend's Big 12 games based on ratings and adjusted stats to date each week. So with that in mind it should be understood from the outset that personnel matchups, coaching decisions, injuries, etc. will not play into the numerical predictions and are all important information that will still need to be considered.

Some of you may be familiar with my matchup analyzer, which predicts a final score based on the power ratings of each team as well as specific stats for the game. What it doesn't yet have, but may be added soon, is a second final score prediction based on the adjusted stats it shows. As discussed in the Adjusted Stats Year in Review, regression analysis can be used to estimate points scored against rushing, passing, and turnovers. I have chosen to use 2011 stats to update that regression weekly based on the season so far. For 2011 that number is currently 2.84*TPPA + 2.48*TRPC - 94.55*TVPP, which is used for the below analyses.

Also note that my power rating predicted scores below will differ form the analyzer because I'm using a standard 3 point homefield advantage while the analyzer uses the season results to date.

Texas Tech (-7) @ Kansas
Power Rating Predicted Score - Texas Tech 54, Kansas 50
Stat Regression Predicted Score - Kansas 48, Texas Tech 47
Matchup Analyzer Link

You could say Kansas' defense has struggled so far this year, but the early numbers don't look much better for Tuberville's squad. I'm pretty curious to see what the total opens at for this game because if Tech is a seven point favorite then the oddsmakers must be counting on some points for the Jayhawks. Someone who has seen these teams play may have a good feel for this one, but I wouldn't touch this game. This early in the season with the line so close to both the predicted scores, the computer wouldn't recommend it either.

Texas A&M (-3) vs. Arkansas (@ Jerryworld)
Power Rating Predicted Score - Texas A&M 34, Arkansas 17
Stat Regression Predicted Score - Texas A&M 35, Arkansas 24
Matchup Analyzer Link

Last week the Aggies lost the first Top 5 matchup at Kyle Field since Generalíssimo Francisco Franco first died. In doing so they blew a 17-point halftime lead thanks in large part to some extremely questionable playcalling from Sherman and company. Meanwhile the Razorbacks got throttled by a powerful Alabama team. The computer definitely likes the Aggies to cover here, but the less than 9-point difference between the line and closest prediction doesn't rate a play this early in the year.

Baylor (-4.5) @ Kansas St.
Power Rating Predicted Score - Kansas St. 50, Baylor 40
Stat Regression Predicted Score - Baylor 42, Kansas St. 39
Matchup Analyzer Link

The Bears are led by the best player in the Big 12, but this game may come down to the Baylor defense against the Kansas State running game. Snyder will try to control the game on the ground but Griffin will still have his chances against the Wildcat defense. Last week's win over Miami was huge and made everyone in flyover country rise to their feet and chant "Big 12! Big 12! Big 12!" in unison. In fact, it felt like a personal victory for their conference mates because our Big 12 partners' success is how we derive our self esteem. No recommended play here either.

Texas (-10) @ Iowa St.
Power Rating Predicted Score - Iowa St. 34, Texas 26
Stat Regression Predicted Score - Texas 28, Iowa St. 16
Matchup Analyzer Link

Obviously a huge discrepancy between the power rating and stat regression predicted scores. This early in the season I would expect to see this fairly often before the numbers even out. Texas rates very highly in forcing turnovers so far this year which in and of itself is impressive from an adjusted stats standpoint because one of our opponents turned the ball over 7 times in another game (BYU against Utah). The real discrepancy here is that Iowa State has scored more points than their contributing stats would suggest, which is probably due in large part to the triple overtime game. However their defense has not suffered from the same impact, so I'd have to look into it a bit more to get a good feel for what's causing it. To show the discrepancy, Texas' predicted scores against an average opponent would be 28-21 power and 29-18 stats. The Cyclones' would be 42-23 power and 26-25 stats. Either way, with this large a discrepancy between predictions this is another no play although I look for Texas to win but not cover. The Longhorns' special teams remains a concern.

Ball St. (+38) @ Oklahoma
Power Rating Predicted Score - Oklahoma 43, Ball St. 13
Stat Regression Predicted Score - Oklahoma 41, Ball St. 21
Matchup Analyzer Link

The Sooners' OL suffered a loss this week as starting center Habern will be out with a broken arm. They will be forced to slide a guard over to cover and will have to train up another lineman on proper holding technique in order to remain effective. It's a somewhat significant loss but as long as edge contain defenders are miraculously unable to shed Oklahoma blocks despite properly taking them on with the outside shoulder and then trying to get away, they should be fine. I kid, I kid. Not really. Obviously no upset here and also no recommended play with the 8-point difference between the line and closest prediction.

OTHER GAMES OF INTEREST

Georgia Tech (-12) @ North Carolina St.
Power Rating Predicted Score - Georgia Tech 54, North Carolina St. 8
Stat Regression Predicted Score - Georgia Tech 64, North Carolina St. 16
Matchup Analyzer Link

Uh, not sure what to say here. I'm very comfortable saying other people should put their money on Georgia Tech. But I'm not comfortable enough to say I'm going to put my own on them. I don't trust that schizophrenic team despite the fact my computer is actively trying to hack into my bank and online accounts in order to place a bet. Mostly because I obviously don't think the discrepancy between these two teams is nearly that big. The line moved quickly from its open at -10.5, but still.

New Mexico St. (-1.5) @ New Mexico
Power Rating Predicted Score - New Mexico St. 37, New Mexico 7
Stat Regression Predicted Score - New Mexico St. 39, New Mexico 20
Matchup Analyzer Link

The Lobos have just been awful but they just fired their head coach so this could be a rally the troops type game for them. Combine that with the fact that betting on New Mexico State in a football game just never sounds like a good idea and I'll just put this one out there for discussion.

Air Force (+3) @ Navy
Power Rating Predicted Score - Navy 33, Air Force 14
Stat Regression Predicted Score - Navy 51, Air Force 33
Matchup Analyzer Link

That huge difference between total points can be attributed to the style of play of these two teams. With both teams keeping the ball on the ground and grinding the clock, we would expect the score to be closer to the power number. Recall that pace is not a part of the stat regression prediction. Either way both models look for a final margin more than two touchdowns greater than the current line.

Michigan St. (+3) @ Ohio St.
Power Rating Predicted Score - Michigan St. 25, Ohio St. 12
Stat Regression Predicted Score - Michigan St. 19, Ohio St. 8
Matchup Analyzer Link

The computer is not a big fan of the Buckeyes to date. And to be honest, there's no reason it should be as they lost to the Hurricanes badly who in turn got beat by Kansas State. And Ohio State struggled with Toledo at home before beating up on an unimpressive Colorado squad. On the other side of the coin, Michigan St. got beat soundly by Notre Dame despite winning the turnover battle and outgaining the Irish 358-275. 12 penalties and poor special teams play plus the fact it seems my computer is Catholic leads to the predictions above.

That's it for this week.

Tweet Comment 21 comments | 0 recs |

Do you like this story?

Comments

In this case, “Other games of interest” should simply be “Other games”.

by Abe Lemons on Sep 26, 2011 2:00 PM CDT reply actions

Admittedly the only interesting thing about most of them is the line/prediction difference.

by Huckleberry on Sep 26, 2011 2:02 PM CDT reply actions

Re: Iowa St – why not remove the OT data? An offense starting 3 drives at the 25 yd line would artificially skew the data, right? Wondering what the models would show were OT stats wiped out…

by Jeff Peel on Sep 26, 2011 2:14 PM CDT reply actions

I base the stat regression model on yards per play number, but you’re correct that the points aren’t removed from the power model. That isn’t done only for the sake of consistency as I don’t have the data for previous seasons to do that.

by Huckleberry on Sep 26, 2011 2:15 PM CDT reply actions

In fact, it felt like a personal victory for their conference mates because our Big 12 partners’ success is how we derive our self esteem.

Finally, someone who truly understands an Aggie.

by Fried Rice on Sep 26, 2011 2:44 PM CDT reply actions

so when is the computer gonna be ready to tell us when to make this money?

do you have your hit rate from previous seasons available?

by mattdubya on Sep 26, 2011 3:08 PM CDT reply actions

I ran a hypothetical of LSU vs LSU, and a really strange video popped up – I believe it’s John Wayne Gacy in a knife fight with the Twilight Sparkle My Little Pony. I couldn’t watch much because my brain was exploding but it looked like to me that Sparkle was winning.

by CrazyJoeDavola on Sep 26, 2011 3:14 PM CDT reply actions

That probably isn’t too different than what you’d see if you traveled into Les Miles’ head, ala Being John Malcovich.

by nordberg on Sep 26, 2011 3:27 PM CDT reply actions

matt – I used to have back data for seasons starting with 2008 but lost a random assortment of it during server changes. 2008 was a kickass year, 2009 wasn’t, and 2010 started off bad and then got better.

Overall the picks are running to the good at about a 0.590 clip on a unit basis, which is actually pretty good.

by Huckleberry on Sep 26, 2011 3:29 PM CDT reply actions

nice. anything over 55% seems decent to me. maybe i’ve missed it in previous posts but do you have confidence ratings added to these picks down the road?

by mattdubya on Sep 26, 2011 3:33 PM CDT reply actions

Huck -

How significant do you think removing OT yardage would be? On one hand, the offense is limited in that it cannot gain more than 25 yards (although an infinite # is theoretically possible, I guess, due to penalties). OTOH, the likelihood of the offense scoring is much, much greater.

I know you don’t have the data, but academically speaking, do you feel that the model would be more accurate if OT data were removed? And if so, were I to obtain such data, how many tote bags would that net me?

by Jeff Peel on Sep 26, 2011 4:12 PM CDT reply actions

Yeah, regulation-only data would certainly make the power model more accurate although the effect on the stat regression model would be minimal if any.

You could have as many imaginary tote bags as you want at that point. Of course, it’s still a lot of work from there but data acquisition is a pain in the butt.

by Huckleberry on Sep 26, 2011 4:23 PM CDT reply actions

Interesting OU’s center broke his arm. Was it his snapping hand ? I wonder if they will just flip flop center guard or something to keep him on the field in a cast. It is Bob Stoops after all. A broken arm is no cause for not playing. Two broken legs, maybe, assuming both of them were compound fractures and the player is back by the Texas game.

by roach on Sep 26, 2011 4:27 PM CDT reply actions

That’s what she said.

by Jeff Peel on Sep 26, 2011 4:29 PM CDT reply actions

Couple questions from someone who’s just now taking an interest in this kind of stuff:

1. Do your models consider time? My guess is that some teams change considerably during the course of a year, particularly young teams such as Texas, so an early performance should be given less weight relative to a recent performance.

2. Is there any way to account for specific match-ups? I’m thinking of some kind of factor analysis that would identify attributes that aren’t measured directly. It would be sweet if these variables were available (i.e., measured directly) and maybe it is, if you’ve got the results of each player (e.g., Team X makes a lot of yards from their TEs and Team Y gives up a lot of yards to TEs).

by MarkW on Sep 26, 2011 4:57 PM CDT reply actions

checking out your analyzer, lots of fun, im sure it will even out a few more games down the road. all the books have Texas a solid 9-10 point favorite, thats blowout status for a road game. Texas Defense is 2nd in the country in td’s allowed with 3. Iowa St has allowed 8tds . Texas is 5th in total yards allowed this year with 776. Iowa is 28th with 1093 yards. points allowed, Texas allows 15 ppg, 16th. Iowa st 26.7ppg,74th in rank. Texas defense is better than Iowa st’s D. Garrett Gilbert may have skewed the stats of the offense by a bunch. im giving the points and taking Texas. A weak Iowa State defense and a Texas offense without a history of game film gives Texas a heavy advantage i believe.

by MONTY on Sep 26, 2011 5:45 PM CDT reply actions

Does the analyzer tell you which one of nm/nms are the lobos?

by Castle AAARGHH!!!!! on Sep 26, 2011 6:46 PM CDT reply actions

I think you may have misinterpreted the part you’re trying to poke fun at.

No worries, I’ll try to program a help-our-readers-read-anator before the end of the year.

by Huckleberry on Sep 26, 2011 7:28 PM CDT reply actions

Bet the house on GT, got it.

by redfoot on Sep 27, 2011 6:55 AM CDT reply actions

Huck- what are your thoughts on Washington St +3.5 @ CU? The computer model has it as WSU 57 CU 24. That is a 36.5 point difference (currently the highest spread/prediction difference of all the games this week).

by bigdmullet on Sep 27, 2011 11:44 AM CDT reply actions

bigdmullet -

You can check out the new model prediction tracker here. I’m trying out a new theory where I base the recommended plays on both of the two models. Each model makes a prediction and then each possible bet is analyzed based on the less confident of the two models.

So the final output only recommends bets if the less confident of the two models still thinks it’s a good enough play. Please note that I have done zero backhistory on this method so there will still be some tweaking in the weeks ahead based on number of bets it’s recommending. I’m completely open with my results so I figured I’d just put this out there and we can all follow along.

As for the Wazzu @ Colorado game specifically, you will see that it is a non-recommended play this week. The reason is that, while the power model has that huge blowout, the stats model has it as 41-28. It’s still a decent play but the computer says that at +3 the play is 71.8% likely to hit based on the less confident stat regression model, which just isn’t high enough this early in the season.

by Huckleberry on Sep 27, 2011 1:35 PM CDT reply actions