keep up this great work. thx
Through the first two weeks of the season, here is what my ballot looks like.
|Rank||Team||Conf||Preseason PAN||In Season PAN|
|4||Ohio St||Big Ten||11.8||13.6|
|9||Oklahoma St||Big XII||5.0||10.9|
|13||Texas Tech||Big XII||7.0||9.8|
|18||W Virginia||Big East||8.2||8.4|
Other Big 10 Teams
30 Notre Dame
34 Michigan St
36 Penn St
This is a purely data driven poll, there is no human intervention or manipulation. PAN stands for Points Above Normal. It equates directly to points scored Oregon at +20 PAN is 20 points better than the average team and about 12.5 points better than Michigan (+7.5 PAN) is projected at this time. The normal that it used is the average of the 120 FBS/1-A schools. For BCS conferences like the Big Ten, about 3/4 of the conference is usually above Normal.
Games against FCS/1-AA schools are not included in the calculation unless the FCS team wins, in which case a game score of –30 PAN is applied to the losing FBS team.
The pre-season metric is calculated based on team PAN’s for previous years. The In-season metric is calculated with each game played so far this year worth 25% and the remainder made up of the preseason number. Michigan is 50% actual and 50% preseason since they have played two qualifying games. Purdue is 75% preseason and 25% actual since they have only played one FBS opponent.
I will use this methodology for the first month of so of the season until all teams have played at least two FBS games to do a proper in season accounting of strength of schedule. Until then opponent strength will be accounted for based solely on the preseason rating.
UMass: 100% win likelihood
Bowling Green: 100%
At Indiana: 80%
Michigan St: 66%
At Penn St: 49%
At Purdue: 59%
At Ohio St: 23%
Putting the rest of the season together, yields an average win total of 8.7 wins which of course is impossible. Running a Monte Carlo simulation on these odds yields the following spread:
Nine wins (5-3 in conference) is the most likely scenario with about a 75% likelihood of 8-10 wins. This picture will change over the course of the next 2-3 weeks as we get more interplay between FBS teams and get a better gauge of team strength, but based on performance to date Michigan is obviously grading out very well.
keep up this great work. thx
If we go 5-0 thru Indiana, beat Illinois, and Purdue as expected, then we just need two more to hit 9. I think MSU and either Iowa or Penn State are the likely candidates. I cannot say that I had any hope of 9 wins before this Saturday. Sadly, I am already at a place where fewer than 9 will be a huge disappointment. Of course, all this depends on Shoelace staying healthy...
Per usual, really excellent stuff.
If I read this right, Oregon's current In Season PAN = 50% of its 2010 PAN + 50% of the Preseason PAN. That means its 2010 PAN is actually 31.1.
Obviously that number should come back down, but holy bejeebus, 30+ points better than the average FBS team is crazy good.
will drop significantly as they enter league play.
Thank you, my Product Marketing brain is much happier now.
I still don't grok PAN, but now with a chart I can compare Michigan's PAN to everyone else's PAN, and make jokes like this;
Hey Oregon, is that your PAN in your pocket, or are you just really happy to see cheerleaders again?
or, "when she tells you the size of your PAN doesn't matter, it matters"
also graphs are great, because now even a 5 year old can figure it out,
" Hey Timmy, how many wins would you pick for Michigan?"
Timmy - "Umm, 9?"
"why did you pick that Timmy?"
Timmy - "ahhh, cause it's the Big One"
very awesome, thank you
Can you do a side-by-side of your pre-season Michigan predictions and the ones in this diary? I'm just curious to know how much the match-up percentages have changed in two weeks.
Michigan's projection is up 1.2 wins vs preseason (7.5 to 8.7). Of the 1.2 increase 0.8 comes from starting 2-0. Bowling Green is up from 78% to 100%, Penn St is up from 32% to 49% and Purdue is up to 59% from 49%. Most of the other match-ups are within a couple points of their preseason marks.
Thanks a bunch!
Numbers are better than cookies. I like the odds of 8-10 wins, make them higher, I say!
better than a cookie? No way.
Do your likelihood of win percentages take into account home field at all? If so how? Also, do you have PAN numbers for every FBS school? Do you have a website where you publish them? If not you should post them weekly on the site as a "rest of the FBS" segment after you put up your top 25.
Homefield is accounted for in the win projections. Homefield is worth about 3 points per game. If otherwise equal teams meet, the home team will win just under 60% of the time. I do have numbers for all teams and a website, unfortunately I haven't had time to get the website up and going so it will probably be sometime in the offseason before I have anything publicly available.
They are not very good. However we have lost to them in the past 2 years. Is there something about the way Purdue is built that hurts us?
Remember, Purdue's numbers are actually based 75% on expectations. What those expectations are derived from, I'm not certain. Maybe one of Mathlete's previous diaries? But I do know that Purdue was expected to be better than they have shown thus far in the season.
This is a game that will most likely prove more winnable as the season progresses and more data is accumulated.
I hope we can win at least 9 games. It's looking a lot more like anything less could be considered a disappointment!
Looking more and more like we could start 6-0, then pick up two more wins from P-U and illinois. Then hopefully take out one of Iowa, PSU or Wisky.
First of all, I think this is very cool considering this is pretty much what I feel like our likelihood of winning these games is.
I'm not sure if it is possible with your model, but do you think that you could project how the outcome of the ND-MSU game could affect our projections for the season? If ND wins, that increases their PAN and, in turn, our PAN while simultaneously decreasing MSU's PAN, which would increase our percentages of winning against our future opponents, particularly, MSU, correct? The opposite should also be true, so I'm just curious how much a big ND win, close game either way, or big MSU win would affect the projections of our season.
Typically awesome work! Thumbs up to you!
the pre-season enough. but i can't remember if you looked at how well current-season games predict rest of season results versus weighted past seasons in a previous post, so maybe i'm off.
Prior to the season I had our Maize and Blue at 8-4. After the first two games I still have us at 8-4 with a decent chance of 9-3 depending on the Penn State, Iowa, and Wiscy outcomes. I'm sure anyone following this team as in depth as we all do would take 8-4 or 9-3 in a heart beat.
Am I the only one that is very nervous about seeing "100% chance to win" some games?
Were more likely to finish with 11 wins than 6? Ill take it
Though it is not ridiculous. Before the season I guessed 7-9 wins depending on luck. I believe your predictions were very similar. Now Crist not being able to play for most of the first half is a clear +1 on the luck factor so I feel that firmly puts Michigan in the 8 win category. My math is super simple so 7 base wins +1 for a lucky outcome in a game = 8 wins. Your number crunching here says 9 and I like that number plenty so you'll get no argument from me. So my only question for you Mr. Mathlete is: where does your PAN rank UConn?
UConn is 66 but started the season at 46 and saw a big drop after getting smoked by Michigan. Their second game was against a FCS team so it did not count.
The good news is there are only 5 Big Ten teams (other than Michigan) ranked higher than that preseason UConn ranking. Of course Michigan plays them all - no Minnesota or Northwestern on the schedule. So if the UConn PAN is accurate (pre-Michigan) then there is reason to believe Michigan will win 7 games: UConn (fait accompli), ND (fait accompli), UMass, BGSU, IU, Illinois, Purdue.
Michigan's comparables (below Michigan but above UConn and close to ND) appear to be MSU and PSU. Sparty is at home while the Nits are on the road. I feel like beating Sparty at home seems more likely than beating PSU on the road. Of course I'd like to win both but we'll have to see how the season plays out. If ND pounds Sparty then that game feels much more secure. If Sparty wins then that likely becomes a tougher game than I anticipate.
Great analysis. One question though: did you assume that the result of each game is independent to each other? Because they are surely not. And therefore you will have to make predictions every week like those ESPN analysts.
If I would have known about cool stuff like this I would have paid attention more in my methods classes. It is very interesting how stats can be used to show any number of things