An updated prediction for the 2009 season using a solid and proven mathematical model*
This is a follow up to a diary I wrote before the season started, where I aggregated all of the preseason polls I could find and tried to turn their individual uselessness into something that might be useful. I thought that perhaps that:
AVERAGE(Moron1:MoronN) +/- arbitrary fudge factor = something to kill time before kickoff
I got enough positive comments that I decided to update the prediction based on the first two games and take it a little further into implied point spreads.
My initial model predicted 6.28 wins which seemed about right at the time given the little we knew about the new roster and the other teams in the Big Ten. However, I closed the diary by saying the following:
“The first two games should tell us everything we need to know. If M starts out 2-0 then look for an 8 or 9 win season as the Illinois and Wisconsin games would start to look much more favorable. A 1-1 start means 5 or 6 wins as it means we have lost our first tossup game and our ranking around 50 is probably correct. 0-2 means a season of 3 or 4 wins is coming as we have dropped one of our 5 likely wins and all the tossups start looking like losses.”
Some of the polls I used at the beginning of the year don’t continue to do a 1-120 ranking each week but I was able to find enough polls to get a good cross section of data and remove the effects of any outliers. Currently Michigan’s average ranking is 22 compared to the preseason average of 49. Future opponents are now ranked as follows (preseason ranking in parentheses):
EMU 112 (113), IU 83 (91), MSU 50 (29), Iowa 26 (23), Delaware State NA, PSU 10 (11), Illinois 60 (38), Purdue 57 (76), Wisconsin 40 (42), and OSU 8 (8).
I then used a simple model**. I start every game as a 55-45 proposition in favor of the home team. I then add/subtract a point based on the difference in ranking. So a home team ranked #20 would be given a 75% chance of beating a team ranked #40 (55 + (40-20)). Given upsets in recent years, I topped it out at a 95% chance of winning or losing.***
I also looked at spreads and money lines to determine what point spread was being implied to get a sanity check on the numbers. From looking at a combination of spreads and money lines for this weekends game I determined that a 3% increase in your chance of winning increases the spread by 1 point. So a 59% chance of winning a game implies a spread of 3 points.****
Here is what the numbers looked like then and now for Michigan’s probability of winning each game along with the spread implied by that percentage:
Game preseason week2 implied spread
WMU .79 1.00
ND .34 1.00
EMU .95 0.95 -33
IU .95 0.95 -23.5
MSU .25 0.72 -7
Iowa .19 0.49 pick
Del St .95 0.95 NA
PSU .17 0.42 +2.5
Ill .34 0.82 -10
Pur .82 0.90 -13
Wisc .38 0.63 -4
OSU .14 0.41 +3
Total expected wins preseason = 6.28
Current expected wins = 9.25
What a difference a few weeks makes. My main concern pre-season was that the Big Ten Schedule was the exact opposite of what I wanted. In all of the home games UM looked to be heavy favorites or heavy dogs. Meanwhile all of the away games looked like tossups. Now I like this schedule much better. I’d still trade having PU and IU at home to have MSU and Iowa at home, but having PSU and OSU at home now looks a lot more fun.
From a point spread perspective its looks decent. Vegas EMU spread is currently -24 whereas I am saying it should be -33 (for the EMU and IU games I used the actual delta as opposed to my 95% artificial limit). If the MSU game was happening tomorrow I would expect a point spread of 4-5 not 7. If Illinois was tomorrow I would probably expect 6-7 not 10. But the others actually look pretty reasonable if those games were being held tomorrow. *****
*Model used neither solid nor proven.
**for the sake of simplicity
***I realize there are more scientific and complicated ways to do this. There are inputs better than poll numbers. However I am a back of the envelope guy and at the end of the day you are trying to determine a win total in a pretty small band – no matter what you use you are probably going to come out with a number between 8 and 10
**** I’m not a regular gambler, I deduced this by looking at a dozen or so spreads and money lines, if anyone has better data let me know
September 18th, 2009 at 6:48 PM ^
September 18th, 2009 at 7:41 PM ^
September 18th, 2009 at 6:52 PM ^
September 18th, 2009 at 7:40 PM ^
September 18th, 2009 at 8:03 PM ^
September 18th, 2009 at 7:02 PM ^
September 18th, 2009 at 8:14 PM ^
September 18th, 2009 at 8:17 PM ^
September 18th, 2009 at 8:50 PM ^
September 18th, 2009 at 9:52 PM ^
Comments