BYU, do you know the way to San Jose?

[Donuthole]

Well there is the fun general explanation and there is the boring intuitive explanation. The fun explanation requires at least an undergraduate background in econometrics or regression analysis. Here is a little document I would use with my undergrads as an example of where regression analysis is used. This is the fun general explanation (I also posted a similar explanation on cougarguard one time):






That's the technical explanation of how computer models work. Well, at least MOV based computer models. Win/Loss computer models tend to use algos like Elo-Chess. I suppose these days Elo-Chess is besst known for being the algo they use to rank women at the beginning of the Social Network. It is actually a pretty simple iterative algo, but no W/L model is any good for college football. There are not a enough games given the noise in the outcomes; you can't afford to throw out margin of victory; it simply throws away too much information.

Very interesting read. Thanks for posting that, p-lag.

[Indy Coug]

I suppose these days Elo-Chess is besst known for being the algo they use to rank women at the beginning of the Social Network. It is actually a pretty simple iterative algo, but no W/L model is any good for college football. There are not a enough games given the noise in the outcomes; you can't afford to throw out margin of victory; it simply throws away too much information.

Amen

[pelagius]

That's fine, I am not some computer expert with how this stuff works.

I don't understand how the computer likes them. There's nothing in their record that stands out besides playing Stanford close. They got killed by USU at home. They have beaten up on a bunch of WAC schools. They're in third there. They beat SDSU which rates around 60 in Sagarin. They beat Navy and CSU which at both 90+. They rate at around 47 in Sagarin right now, which I think is high.

What am I missing here?

My attempt at an intuitive explanation:

Well, first it isn't like any model thinks that San Jose St is a great team. Most models are going to rank them in the 35-50 range. So a team that is above average relative to the population of 1A college football but not a ton above average. The following is driving the ranking in most models.

1. They have won most of their games: 6-2. It is a pretty good record. Even against easy schedules, below average teams (70 or worse range) don't typical win most of their games. For average or below average teams, if you play 6 games against below average teams it is very likely that you lose 1-2 of them.

2. San Jose St has beaten average to slightly below average teams on the road. While those teams aren't great, beating them (particularly on the road) is informative. If San Jose St was in the 50-70 range in quality (average) or worse than the probability that they win those games on the road is relatively low.

3. The really bad teams they have beaten t by a lot. There is not a lot information in beating really awful teams by a lot, but their play here is consistent with an above average team. Of course, losing to a really awful team is informative.

4. The win at Texas St was only 11 points, but that doesn't hurt them a lot. Texas St is a bad team: rank ~ 100, but they are not awful like UC Davis or Colorado St. Above average teams do lose occasionally to the Texas St's of the world.

5. They only have losses to good teams: Utah St and Stanford. Yes, the Stanford game is helping some. The Utah St game hurts because of the 22 point MOV but most models have Utah St in the top 30 so it is not killing their ranking; plus it is only one observation.

[pelagius]

What am I missing here? Your other thread still has the probability at around 55% of a BYU win, right?

I don't see it I guess. I'd think it would be close to 80%.

Well, I quit updating that thread because it was meant to give people a sense of the expected distribution of losses, but starting with last week the distribution collapsed. Only the Georgia Tech and SJSU games had any uncertainty so you only had to multiple two numbers together to figure out the potential distribution of losses. But it is easy enough to do probabilities for SJSU

Massey: 64%
Sagarin: ~ 57%-63%

So they are below your 80% but make BYU the favorite in both cases. If I was modeling it seriously I would have SJSU lower ranked than both Sagarin and Massey. Primarily because I would still shrink back SJSU to its historical mean. SJSU is way better than last year and their long run historical average. There is a reasonable chance that they have been fortunate/lucky this year given how big the deviation is from the historical performance (so it is more likely they are overrated relative to their true quality). So maybe around 45-50 rating: so that would put the the probability of a win around 70%. Not a gimmie but I think BYU will be a solid favorite.

[The_Douger]

Well, I quit updating that thread because it was meant to give people a sense of the expected distribution of losses, but starting with last week the distribution collapsed. Only the Georgia Tech and SJSU games had any uncertainty so you only had to multiple two numbers together to figure out the potential distribution of losses. But it is easy enough to do probabilities for SJSU

Massey: 64%
Sagarin: ~ 57%-63%

So they are below your 80% but make BYU the favorite in both cases. If I was modeling it seriously I would have SJSU lower ranked than both Sagarin and Massey. Primarily because I would still shrink back SJSU to its historical mean. SJSU is way better than last year and their long run historical average. There is a reasonable chance that they have been fortunate/lucky this year given how big the deviation is from the historical performance (so it is more likely they are overrated relative to their true quality). So maybe around 45-50 rating: so that would put the the probability of a win around 70%. Not a gimmie but I think BYU will be a solid favorite.

I find this stuff interesting, thank you for the explanation. I took a few stats classes as requirements for my major at BYU so I have a basic understanding, but that helps me understand a lot better. Overall, when I look at something like Sagarin, I think it's pretty objective and I agree with it, but there's always a few outliers that don't seem to fit in and SJSU is one of those to me.

Sagarin has BYU ranked at 22, despite our four losses. I think that is high as well, but maybe we're just not giving ourselves credit for being a pretty good team.

[LiveCoug]

LOL.

[LiveCoug]

Does the SJSU game deserve it's own thread? They are, afterall, 8-2 with a very good QB. Nah, nevermind.

I do worry about this quarterback. Our pass D has been suspect lately, especially the longer stuff.

Also, good news for Dish subscribers. The game is going to be on ESPN2 and not ESPNU, so HD for us!

[Art Vandelay]

BYU, do you know the way to San Jose?

BYU is favored by 3. Haven't pelagenius' computers been saying for weeks, that BYU is likely to lose this game?

[Guest]

I am thinking that Brandon Doman will be fighting for his job with this game. If BYU's offense stinks it up. Doman should simply stay in San Jose.

[filsdepac]

Weather isn't looking good for this game, so perhaps that'll help BYU. Luckily i'll be dry in PAC's luxury box suite.

[old_gregg]

Weather isn't looking good for this game, so perhaps that'll help BYU. Luckily i'll be dry in PAC's luxury box suite.

[pelagius]

BYU is favored by 3. Haven't pelagenius' computers been saying for weeks, that BYU is likely to lose this game?

No, they've been saying that BYU would be about a 3-4 point favorite: see, http://cougaruteforum.com/showpost.p...0&postcount=19

Well, I quit updating that thread because it was meant to give people a sense of the expected distribution of losses, but starting with last week the distribution collapsed. Only the Georgia Tech and SJSU games had any uncertainty so you only had to multiple two numbers together to figure out the potential distribution of losses. But it is easy enough to do probabilities for SJSU

Massey: 64%
Sagarin: ~ 57%-63%

So they are below your 80% but make BYU the favorite in both cases. If I was modeling it seriously I would have SJSU lower ranked than both Sagarin and Massey. Primarily because I would still shrink back SJSU to its historical mean. SJSU is way better than last year and their long run historical average. There is a reasonable chance that they have been fortunate/lucky this year given how big the deviation is from the historical performance (so it is more likely they are overrated relative to their true quality). So maybe around 45-50 rating: so that would put the the probability of a win around 70%. Not a gimmie but I think BYU will be a solid favorite.

So Sagarin and Massey were both in the 3-4 point range, and I put it at a 7 point expected MOV.

[byu71]

No, they've been saying that BYU would be about a 3-4 point favorite: see, http://cougaruteforum.com/showpost.p...0&postcount=19



So Sagarin and Massey were both in the 3-4 point range, and I put it at a 7 point expected MOV.

This is the first game in a long, long time I have been tempted to bet. So you're saying this would be an emotional decision and not the fact the odds makers have screwed up?

[jay santos]

I am thinking that Brandon Doman will be fighting for his job with this game. If BYU's offense stinks it up. Doman should simply stay in San Jose.

What would his ward do?

[pelagius]

This is the first game in a long, long time I have been tempted to bet. So you're saying this would be an emotional decision and not the fact the odds makers have screwed up?

The line opened at BYU -7. It quickly moved to BYU -4, and over the next couple of days settled in at BYU -3 to -3.5. Using reasonable modeling, I would put the expected MOV at between 3 to 7. So I don't think either the opening or current line are likely to be seriously out of whack. Is it possible that the true E(MOV) is something closer to BYU -10*? Sure, it is possible.

*Note, I am talking about the expected MOV here and not realized MOV; this is about center mass of the distribution. Even if the true expected outcome of is BYU -3, 70-75% of the realized outcomes would fall in the range of San Jose St by 11 to BYU by 17. Realized outcomes in football have large standard deviation.