1 and 118? So you are sayin' there's a chance! -LLoyd C. (Not that LLoyd)
here's one vote for "John Beilein's head in a Futurama jar"
With four weeks under our belts, all numbers move to in-season only. No preseason adjustments anywhere. Everything is currently opponent adjusted as well. It’s still early in the season so the next 2-3 weeks will still show some weird anomalies but based on history, four weeks is what it takes to start to get an accurate read on what teams will do this season.
Three biggest plays for Michigan:
1. Vincent Smith’s 32 yard run through to the 5, 10% added.
2. Denard Robinson 53 yards on the speed option, 8% added.
3. Ronnie Hillman’s first fumble, 7% added.
Three worst plays for Michigan:
1. Vincent Smith’s fumble, 7% lost.
2. Ryan Lindley 16 yard TD pass, 3% lost.
3. Ryan Lindley 9 yard pass on third down early in the second quarter, 3% lost.
Brian is in love with it, but how much was it worth? Punt from 48 gets to the 17. Team down 14 with the ball around the 17 with 2-3 minutes left in the first half win about 8.0% of the time. A successful conversion gives Michigan a 93.2% chance of victory where a failed attempt drops your chances to 88.2%. To break even, Michigan would need to have a confidence that they had about a 75% chance of conversion. National average on 3rd and 2 is about 58.5%. Michigan has been a top 25 level 3rd and short team so the decision was probably about a break even if you account for Michigan’s offense.
This case is a bit closer than I expected, but if you believe our offense was bound to score, which it obviously did, a 21 point half time lead is good for a 97.1% chance of victory. Even if Michigan can get a field goal and run out the clock, an average conversion rate makes the decision break even.
Ultimately, the decision in and of itself may not have been a true EV+ but it opens the door of optimism that Coach Hoke is willing to go EV+ and close the close door that he EV-.
Win odds on specific situations like this are fairly accurate but still prone to some sample size issues. Typical sample size on these situations was around 120 games, not small but a swing of a game or two can distort the numbers.
My top 5: 1- LSU 2-Wisconsin 3-Georgia Tech 4-Oklahoma St. 5-Alabama
Big Ten Rankings: 10-Illinois 21-Nebraska 29-Michigan 32-Penn St 34-Michigan St 38-Iowa 83(!)-Ohio St 102-Purdue 103-Minnesota 108-Indiana 110-Northwestern
Big Ten: Not a good start to the year as the Big Ten finds four teams in the bottom 20 and only 2 in the top 20.
Minnesota’s numbers are a little weak right now. I don’t include FCS opponents which would knock down the Gophers further but the strong performance against USC is also excluded as its one of two games this year without a play by play available from the NCAA.
Michigan: +5, 7th Nationally (Georgia Tech), 1st Big Ten
Minnesota: –1, 84th, 9th
Denard Robinson: +4 rushing, 7th among QBs rushing (Collin Klein, Kansas St)
Fitzgerald Toussaint: +1, 64th RB (Orwin Smith, GT)
Michigan: +0, 62nd (Boise), 7th (Wisconsin)
Minnesota: –6, 117th, 12th
Denard Robinson: +0 passing, 83rd (Kellen Moore, Boise St)
Michigan: +2, 30th (Louisville), 4th (Illinois)
Minnesota: –2, 89th, 11th
MarQueis Gray: +5 rushing, 2nd among QBs
Duane Bennett: –3, 183 out of 191 rated RBs, last in Big Ten
Michigan: +0, 63rd (UCF), 7th (Illinois)
Minnesota: +0, 60th, 6th
MarQueis Gray: +4 passing, 43rd
Da’Jon McKnight: +6 receiving, 45th
Michigan: –.46 pts/drive, 112th (Boise), 12th (Ohio St)
Minnesota: +.27 pts/drive, 26th, 4th
Michigan: –1.2, 93rd (Auburn), 12th (Purdue)
Minnesota: –0.3, 79th, 9th
Michigan: +6, 8th (Rutgers), 1st
Minnesota: +2, 39th, 7th
Prediction: With an active QB and a dangerous receiver, Minnesota has just enough weapons to but a bit of doubt, but Michigan holds a huge advantage with the ball. Michigan by a couple TD’s, 35-17
There is a still a lot of volatility in the numbers right now, but based on year to date performance, Michigan is currently projected at just over 9 wins on the season. I expect a fair number of game odds to swing but as of now, here are the remaining games and the current projections of Michigan winning.
At Northwestern: 94%
At Michigan St: 35%
At Iowa: 39%
At Illinois: 16%
Ohio St: 88%
Some of those are clearly out of whack right now, but the numbers love Illinois’ defense and are projecting Ohio St to go 5-7 and miss a bowl based on current performance. Odds of going 12-0 are currently at 1 in 118.
1 and 118? So you are sayin' there's a chance! -LLoyd C. (Not that LLoyd)
Say we had punted to the 17...what was the probability that they would score in that scenario. If they did score and it was 14-7 at the half, what was the probability of a win?
16.7% Chance of a touchdown, 73% half time odds.
7.8% Chance of a field goal, 85.5% half time odds.
Thanks. This is my favorite diary each week because I'll be at the games thinking about how big plays affect our chances of winning.
I think it would be cool for you to come back to these game-changing decisions and evaluate whether they were the correct decisions statistically, based on our end of year stats.
Also, these numbers give us less than 4% chance to win 7 or fewer games, almost 90% chance to win 8-10 (favoring 10 over 8), and about a 9% chance to win 11-12 for anyone else wondering.
I realize this is only four games in so, as you said, the data still has a lot of volatility. But in your experience, have you ever had games where the numbers stated a team had a 100% chance of winning, but then lost? Just curious after reading the predictions for the rest of the season. Thanks and very interesting stuff as usual.
Boo maize on yellow!
How about Smith?
Why does the regression factor drop out after week four?
The numbers really rank Illinois' defense highly. My question is how much does someone like Denard affect this. In other words, Illinois might win almost all their games, but do the rules change when they face a great offense with a great gamechanger like Denard?
I'm thinking about this because of a link I read (from Brian yesterday) talking about how Tressel was going to win the Big 10 Championship almost every year because of superior athletes and defense, but would lose when facing Bowl teams having an equally high caliber of athletes (i.e., LSU, Florida, Alabama, etc.) This makes me wonder if Illinois, while truly having a solid defense, still won't be able to cope with Denard's 4.3 speed, especially if:
If these things happen, I could see our offense doing very well, even against Iowa, Illinois, Nebraska, and OSU.
This weekend's Michigian State / Ohio State game should be Very Telling.
If Michigan State wins, it will be very telling.
If Ohio State wins, it will be very telling.
I think they are both mediocre. I am not sure it will tell us much unless someone is blown out. State's got problems up front and I was never convinced they were that good to begin with. We know OSU has some issues. More interested to see how Nebraska does.
I'm surprised none of Denard's picks show up as one of the top 3 negative plays.
We were up 21-0 in the 3rd quarter. Everything is based off stats, so there really can't be that many teams that come back after being down 21 points, even after the opposing QB throws an interception.
Why is it that FEI has us so much higher than your analysis? Any idea? From FEI's rankings, we would certainly have to be highly favored against MSU and at least somewhat favored against Illinois, whereas you have us very likely to lose to MSU (which I very much doubt will happen) and almost a lock to lose to Illinois. From what I have read, you and FEI seem to be based on statistical analysis. Any idea why FEI's is so different from yours?
Does the first score of the game have extra weight or something? Why would Smith's long, non-touchdown run be worth more than Denard's long, touchdown run?
It's all situational. Vincent's run pushed the game from tie to 7 points (almost, I know he didn't actually score himself). Where Denard's TD took the lead from 7-14 which isn't as valuable as breaking open the tie.
By us scoring we ended up at with a 97% halftime odds of winning but by us punting they could have gotten to 73% chance of winning. So that's like a 10 times greater chance of winning on the upside and that's what Brian was talking about by putting a boot to their throat.
Also I think there's something strange about saying which plays added the most percent as a way to measure the most postive plays. Isn't a play that take you from a 1% chance off winning to a 11% a hell of a lot more important or positive than a play that takes you from 50% chance to 60% chance? The first play improves your chance of winning by roughly 11 times and the second by 1.2 times.
I guess my point is it seems like it would be a worthy exercise to point out the big +'s when when the game is competitive or we are losing and the big -'s when we shuttin be putting the game away.
Excellent diary! Would love to know how to calculate this info but I'm sure that would take too much time
Do you have winning percentages given a pregame spread (i.e., if a road team is a 5 pt favorite vs. the home team, how often do they win)? If you wouldn't mind sharing that info, that would be most appreciated. Not sure if road vs. home favorite would skew the results.
Would love that data for up to a 17 point favorite for each home and road team.
I get the biases generated by allowing "reputation" to creep into rankings, blinding us to actual performance, but in the case of the Team From Ohio don't we have what the judgment psychologists call a "broken-leg cue" - that is, a factor that the better knows will affect the future even when it does not show up in past performance?
Horse-racing origins of this phrase: Statistic model predicts Horse X will win with 88% probability, but bettor Y knows that X broke his leg, so 88% is probably too high.
Wolverine-OSU analog: model based on OSU's first four games predicts 88 % chance of Blue victorious, but we know OSU gets some key players back before then.
So, how shoud this affect your prediction of 88%??
Mathlete, where are you? Need update of future predicted wins...