Massey predicts BYU will go 10-2

[pelagius]

So says this cougarboard post (incorrectly): http://www.cougarboard.com/board/mes...tml?id=9120606

Massey predicts BYU to be 10-2
http://masseyratings.com/team.php?t=891&s=181623

Close wins against Utah and GTech, bad loss to Boise State and a loss to Notre Dame.

Pretty cool ...

I suspect a few more wins along the way the odds at ND will increas

This is not pick on the cougarboard poster. Most people (including very smart people are bad with probabilities). But, the problem is that Massey doesn't predict this at all. I could tell that from just casually looking at the numbers but it also got me kind of interested to compute what the Massey model actually predicts. Massey actually predicts that BYU will most likely go 9-3,8-4, or 7-5 with 9-3 and 8-4 being a bit more likely than 7-5. Yeah, I know, Earth shattering given Massey's current ranking of BYU.

Massey currently has BYU ranked 31st (which is mostly based on last years ranking ... see here for more info about these issues in general: http://cougaruteforum.com/showpost.p...&postcount=160 ). In my view (and not surprisingly for a computer model at this point) some of the rankings are really goofy:

Opponent        Rank    Prob
Washington St    94     100%
Weber St        146      99% 
Utah             45      51%
Boise St.        12      16%
Hawaii           84      88%
Utah St.         50      74%
Oregon St.       70      83%
Notre Dame       18      27%
Georgia Tech     52      53%
Idaho St.       128      98%
San Jose St.     54      57%
New Mexico St    89      76%

You will notice that BYU is favored (according to this model) in all but two games. Hence the poster concluded that Massey predicts BYU will go 10-2. We can see this is wrong by just looking at the probabilities of winning the Utah game and the Georgia Tech game: 51% and 53%. The probability BYU wins both according to the model is slightly over 25% (0.51*0.53). So clearly Massey doesn't expect BYU to go at least 10-2 since it expects BYU will most likely lose either the Utah or Georgia Tech game (or both). One needs to work out all these probability scenarios to figure out predicted distribution for the season (there's at lots of possible scenarios). I programmed this really fast but I think the following table is right (don't guarentee it though). Here is what those probabilities actually predict:

Losses  Prob
0       0.27
1       3.08
2      12.86
3      26.07
4      29.20
5      19.14
6       7.46
7       1.70
8       0.21

People, be careful out there when it comes to probabilities.

[Jeff Lebowski]

So says this cougarboard post (incorrectly): http://www.cougarboard.com/board/mes...tml?id=9120606



This is not pick on the cougarboard poster. Most people (including very smart people are bad with probabilities). But, the problem is that Massey doesn't predict this at all. I could tell that just from just casually looking at the numbers but it also got me kind of interested to compute what the Massey model actually predicts. Massey actually predicts that BYU will most likely go 9-3,8-4, or 7-5 with 9-3 and 8-4 being a bit more likely than 7-5. Yeah, I know, Earth shattering given Massey's current ranking of BYU.

Massey currently has BYU ranked 31st (which is mostly based on last years ranking ... see here for more info about these issues in general: http://cougaruteforum.com/showpost.p...&postcount=160 ). In my view (and not surprisingly for a computer model at this point) some of the rankings are really goofy:

Opponent        Rank    Prob
Washington St    94     100%
Weber St        146      99% 
Utah             45      51%
Boise St.        12      16%
Hawaii           84      88%
Utah St.         50      74%
Oregon St.       70      83%
Notre Dame       18      27%
Georgia Tech     52      53%
Idaho St.       128      98%
San Jose St.     54      57%
New Mexico St    89      76%

You will notice that BYU is favored (according to this model) in all but two games. Hence the poster concluded that Massey predicts BYU will go 10-2. We can see this is wrong by just looking at the probabilities of winning the Utah game and the Georgia Tech game: 51% and 53%. The probability BYU wins both according to the model is slightly over 25% (0.51*0.53). So clearly Massey doesn't expect BYU to go at least 10-2 since it expects BYU will most likely lose either the Utah or Georgia Tech game (or both). One needs to work out all these probability scenarios to figure out predict distribution for the season (there's at lots of possible scenarios). I programmed this really fast but I think the following table is right (don't guarentee it though). Here is what those probabilities actually predict:

Losses  Prob
0       0.27
1       3.08
2      12.86
3      26.07
4      29.20
5      19.14
6       7.46
7       1.70
8       0.21

People, be careful out there when it comes to probabilities.

Cool. Simple Monte Carlo simulation I presume?

[CardiacCoug]

Yeah, I look at those Massey predictors and guess it would predict 3.5 losses.

Utah, GaTech, and San Jose all about 50% so that's 1.5 losses (not sure about the 57% over San Jose State number though).

Boise and Notre Dame look like 1.5 losses (no way do I buy us at just 16% to beat Boise, but whatever).

Add all other games up and it's about half a loss.

Sure enough, the Vegas over/under was 8.5 regular season wins I believe.

[pelagius]

Cool. Simple Monte Carlo simulation I presume?

That would work just as well, and is a easier to program quickly but I decided I wanted to do it by mathematically solving (just for fun but the Monte Carlo is really way more sensible and less likely to have an error) so I did all the possible combinations and the probabilities.

[pelagius]

Yeah, I look at those Massey predictors and guess it would predict 3.5 losses.

Very close ... I think it works out around 8.2 wins

[Jeff Lebowski]

That would work just as well, and is a easier to program quickly but I decide I wanted to do it by mathematically solving (just for fun but the Monte Carlo is really way more sensible and less likely to have an error) so I did all the possible combinations and the probabilities.

I just threw together a quick Monte Carlo analysis with 5000 instances, just for fun. Not quite to statistical convergence, but close. This is what I got. Looks like your equations were correct.

I am not liking our chances for an undefeated season.

[New Mexican Disaster]

I just threw together a quick Monte Carlo analysis with 5000 instances, just for fun. Not quite to statistical convergence, but close. This is what I got. Looks like your equations were correct.

I am not liking our chances for an undefeated season.

I don't like the chances either but how good can the data be at this point if it is collected mostly from last year, i.e. Boise is probably not the 12th strongest team this year.

[pelagius]

I just threw together a quick Monte Carlo analysis with 5000 instances, just for fun. Not quite to statistical convergence, but close. This is what I got. Looks like your equations were correct.

I am not liking our chances for an undefeated season.

Yeah, if we want an undefeated season we better hope that our true quality is more like 10th rather than 30th. Even then 1-2 losses would be the most likely outcomes but there would probably be a decent chance at zero losses.

[Jeff Lebowski]

I don't like the chances either but how good can the data be at this point if it is collected mostly from last year, i.e. Boise is probably not the 12th strongest team this year.

Go ahead and give me your best guess of game-by-game probabilities and I will throw them in and see what happens.

Same offer for anyone else.

[pelagius]

I don't like the chances either but how good can the data be at this point if it is collected mostly from last year, i.e. Boise is probably not the 12th strongest team this year.

That's right. Some of those numbers are probably goofy (although if I had to guess the Utah one is too low of ranking and so is Georgia Tech so it may wash out a bit). The big issue is if 30th is the right rank for BYU. Let's hope not.

[The_Douger]

I think we're better than 30th.

[creekster]

At least we wont go winless

[SoCalCoug]

I just threw together a quick Monte Carlo analysis with 5000 instances, just for fun. Not quite to statistical convergence, but close. This is what I got. Looks like your equations were correct.

I am not liking our chances for an undefeated season.

I'd be interested to see what the probability for an undefeated season in 1984 as of the start of the season was.

[Indy Coug]

Of course, Massey et al have virtually no statistical credibility with their rankings for 2012 after week 1.

[All-American]

Go ahead and give me your best guess of game-by-game probabilities and I will throw them in and see what happens.

Same offer for anyone else.

Massey's newest rankings give some different numbers than those in the first post.

http://masseyratings.com/cf/compare.htm

In these rankings, BYU is 27, and:

Weber is unlisted
Utah is 30
Boise is 22
Hawaii is 97
USU is 68
Oregon State is 69
ND is 18
Georgia Tech is 41
Idaho State is unlisted
SJSU is 80
NMSU is 110.

I don't know how to figure the probabilities from that.