that scored fewer than the expected points based on field position, right? I'm not sure how much to make of a single game's results, especially since it's so divergent from the prior four weeks. I don't know what the binomial distribution would look like but I'm pretty sure you can still be a good offense if you're below average every 5th game. I think we get caught up in expecting the same thing, within reason, every week. There's more variance than that and it isn't necessarily that telling.
By The Numbers - Michigan State Recap
Any questions on how By The Numbers works, look here.
This was called out as a 3 point win for Michigan going in. It wasn't. State absolutely shut down the Michigan running game, the final number -4 for the game, but it was worse than that.
Minor was a -2.2 on 4 carries, none of them with positive values. Carlos Brown was -2.0 on 6 carries, with only 2 carries coming out positive. Shoelace posted a pair of negative runs. Odoms one carry was a negative. The only one who did anything on the ground was Forcier (+1.7) and most of his value was on scrambles. 6 of his 11 runs (sacks are excluded) went positive which is not great but for the rest of the teams performance, was far above average this week.
In the preview this was noted as the biggest disparity of any unit in the game. Excluding Stonum's fumble, the passing game was slightly negative (-0.6). The three sacks cost nearly 4 points on the game and the interception in OT was obviously a big deal.
Stonum, apart from the fumble, had a very nice game with 1 point on 3 catches plus over 5 points on the long TD which more than offset the lost value on his fumble.
Matthews was targeted 5 times, none of them complete. We'll have to wait for the UFR to see where the fault lies.
Hemingway picked up 3 catches for a meager +1.
Kogar was 1/4 on targeted balls, but snuck into the positives at +.6.
Odoms had a productive day, picking up nearly 3.5 points in 6 targets with 4 points coming on his 5 catches.
Roundtree's grab in the endzone was worth 3 points after missing on two targets that worth 1.5 points.
Saw this as a point or two disadvantage coming in, and that's about where it netted out, although in a strange way.
Jailbird Glen Winston was -8.6, even with his +3 TD run.
Larry Caper's game deciding run, just put him back to even on the game.
Running QB Keith Nichol ran only twice for little value.
Wow! Kirk Cousins, 5 rushes, all for positive value racking up an incredible +8 on the ground for the game. Wow!
What a bizarre path to an expected outcome.
This was the only segment that was a solid win for Michigan. Coming in, it was expected to be a 2 point disadvantage but it ended up being a 2 point advantage.
The two sacks netted a three point advantage for Michigan, even when taking out the benefit of the fumble.
Strangely enough, Keith Nichol (+2.4) added more value through the air than Kirk Cousins (-.4) did on the day.
Cunningham was the only receiver who managed to post more than a point or two of value for the game with a +7.3.
Even with the turnovers, Michigan failed to have an advantage in field position. The regulation numbers for field position where 24 exp points for MSU and 22 for Michigan. The first down at the 25 for OT is worth just over 4 points, so for the game, MSU scored 1 less point than expected (PAT is assumed) after being -4 in regulation while Michigan was -2 in regulation but -6 on the game.
Other the Zoltan audible, the special teams didn't provide any huge advantages for either side.
Olesnavage had another good game, going +1.9 on his two field goals while Swenson was +.6 on two chippies.
MSU had a +.5 advantage on kickoff teams with Michigan giving up a couple of good returns.
When Michigan was actually punting, the punt teams netted out with no real advantages and MSU had negligible advantage in returns.
A lot of value came Michigan's way via the yellow laundry on Saturday. NO CONSPIRACY!
When Michigan was on offense there was a net pickup of a point of value due to penalties. However, MSU's offense had the should have been killer penalty problems, costing them 5.5 points in value most of which coming on...
The Drive that spanned the globe
Thought I would add a little note on the drive that covered a ton of yardage and could not be stopped.
During the drive, Michigan State rushed 10 times for a value of 3.8, 3 of which coming from Cousins. The Spartans passed 7 times for a value of 5.9. That's 17 plays, adding nearly 10 points in value. Obviously, this was all to offset the ridiculous penalties being accumulated during the drive. Michigan State's 4 penalties cost them 4.3 points on the drive. After the second personal foul, Michigan State's 2nd and 25 put their drive expected points at 1.4, less that what they would have expected when they started the drive. A very strange drive.
You are correct, this was actually the first time Michigan has been below +6 in points vs expected. It feels to me like Michigan's total body of work right now is about right. Last week my full season projection had them 10-2 on the season which seemed quite optimistic. After the MSU game, they are pegged right at 8-4 seems like the appropriate number.
wrt talent and skill, though I'm also convinced Rodriguez is worth a game or two per season in coaching/scheming. That might be a little convenient thinking on my part.
why you cant use numbers to simulate real games...
at the end of the day, the game still comes down to exectuion and individual matchups... something math has a hard time computing...
"Matthews was targeted 5 times, none of them complete. We'll have to wait for the UFR to see where the fault lies."
you may or may not be able to figure out fault through the ufr...
the "fault" may be an injury, a miss comunication in the huddle, or a great play by a corner.. all of which probably wont show up in anyone's opinion.
Agree with you 100%. Not trying to predict games, just trying to look at things from a different standpoint, take old stats and try and make them more meaningful. The ability to predict the outcome of the game is pretty low, but you take what you know, you apply it over a lot of games over the course of the season and you can pull some good insight and have a good idea how things went and how they might go in the future.
mystery to their workings, especially if it's something near and dear to their hearts. even simple modeling is useful and informative and helps us make predictions, but the problem here is that we're not working with especially rigorous models or a ton of data. we had four games and no regression. folks said the same thing you're saying about baseball and the gains in understanding have been vast, to the point where they've changed the way organizations are run. pro football teams already do all the stuff the OP is doing (with added rigor of course) and have been since at least the 80's according to Pete Palmer.
but not to make predictions of individual games...
even in the most stat crazy sports like baseball, indivdual at bats are hard to predict as well as particular games...
not trying to discount stats. we use more stats than the average person in our porfession, however, we never use them to try to predict wins or losses, thats where the limits rest..
sports wrtires and magazines may try to predict wins and losses based on stats, but the more informed ones do so on important matchups, and other variables that math may not bare out.
are the ones that rarely use statistics properly. and i'm not saying predicting wins and losses is easy or that the margin of error isn't high (or, rather, that what we're undertaking here isn't way too simplistic for real money making), but the mere fact that you can use a Vegas spread to explain something like 80% of the variance in results is pretty telling. I'm sure Vegas is watching film and doing their scouting, but I'm sure they're crunching numbers as well. There wouldn't be a spread if games weren't predictable enough to make money on.
that the vegas guys are watching games. crunching numbers a little but some of the best football minds in the game, are invovled in vegas, i can promise you that...
the lines change becasue of injuries, starting lineup changes etc. becasue the matchups are the most important, obviously..
im trying to say you are right, numbers are cool and show some cool corilations, but im assuming that you still understand their limits...
nobody in the sports world, would say numbers are more important than matchups, nobody...
presumably, quantification is occurring to some degree. i wouldn't be all that surprised if scouting outstripped the granularity available to most quant analysis, if that's what you're trying to say.
You say not counting the Stonum fumble and not counting Zoltan's failed run attempt. Why do you do this? Are they not a part of the game?
I'm not trying to be a dick or anything, and I'm not huge into math and stats, just wonderin'....
Zoltan's was taken out because I was looking at the Michigan rushing offense. Zoltan's play was technically a rush but doesn't do much in evaluating Michigan's run game since it was such a one off play.
As for Stonum, I take out all lost fumble plays because everything I have ever read or seen indicates that fumbles in general are pretty random occurrences and don't necessarily reflect the quality of the defense or offense. Since I am trying to evaluate different units, including fumbles are huge value plays that I leave out because of their randomness and the major impact that they can have to skew the numbers. Stonum's fumble obviously had a negative impact on the game, but going forward, its more valuable to leave that play out when evaluating the Michigan passing game.
Do you have an explanation posted somewhere that explains how you arrive at your plus and minus figures? I gather that they have to do with performance relative to an average game, but I'd like to know more about the calculations involved.
Just added this:
from which he's calculated expected points based on down, distance and yard-line. from there you can tally the change in points expected from each play. example: 1st and 10 at your 20 is worth a certain number of points on average. if you run a play and find yourself at 2nd and 5 at the 25, the difference between the expected points before and after is the play's value.
i'm less familiar with how he does his adjustments for opponents, but it's some iteration of comparing performance across common foes.
M has just lost to Dantonio and MSU for the second straight year. This is quite ominous. When Tressell began his stint, this is how it started out. A couple of wins and a loss (2-1 in the first 3 years). We all thought no big deal, things will return to normal and we'll start beating up on OSU regularly again. 5 straight losses later, things are not looking good in that rivalry.
In many ways, Dantonio and Tressell are similar. From their hatred of Michigan, to designing specific game plans for Michigan all year long, to losing all their other big games, they are basically the same.
I'm afraid that we are about to start a long losing streak against MSU as well. Dantonia specifically prepares his team for the Michigan game, and for two straight years, they have stoned our offense. It is preposterous to see that RR doesn't seem capable of beating a really bad Sparty team. He clearly didn't prepare for Dantonia threw at him. This is epic failure from the coaching staff.
Unless RR cleans up his act, and starts game planning a bit more, we may be regularly losing to 2 of our 3 chief rivals. At least as long as Fat Charlie is at the helm at ND, we can count on one win. He goes, and we may even lose that one.
you are much smarter than me, i have no idea what half those words meant. no doubt your education trumps mine when it comes to mathmatics..
you couldnt possible expect me to know what you are talking about though, right? In which case im wondering what your motive is...if i miss interpret, i appologize..
regardless it doenst change what i know about athletic competetion..
matchups meaning, their DE's first step verses our OT's ability to kick step...
their fullbacks strength verses our lb's read steps..
their best db's ability to funnel recievers, verses our game plan... etc etc etc
those individual matchups, is what ultimately decides the outcomes of individual games...
not trying to show you up or something. i'm just not always clear.
anyway, i think i understand what you're talking about. scouting allows you to break down the game into individual matchups where the stats we're using here don't really reveal the relationships between the players that lead to a play's success. they merely record the end result: relative success or failure. as far as that goes, i was trying to agree with you. i'm not necessarily going to write off the possibility of some ubermathy technique to tease out interesting results, but i'm sure scouting is the best widely available means for making predictions about future games.
You should team up with the computer programmer on the old Spike show "The Deadliest Warrior." Sounds like your data could fit into his program and could attempt to predict a winner. His system could run the simulation of a game 1000 times (just like on the show) and predict the winner based on total win percentage. Of course, you would have to account for all sorts of things like Pass Offense A vs Pass Defense B and account for Pass Defense B's ability to perform based on season data. But, you are the math guy, I'm sure you'd be all over that. I am looking forward to seeing all your posts. Please keep linking your explanations for those of us, ok just me, who might have to read really slow to understand all the #s.