2007 Preseason and Regular season Polls

[Mike80]

Let's be honest; the Buckeyes were NEVER in jeopardy of losing a game other teams however have 1.LOST THE GAME ON THE ROAD 2.LOST THE GAME AT HOME 3.PLAYED WAY TOO CLOSE TO LESSER OPPONENTS 4.HAD MORE THAN ONE CLOSE CALL AGAINST AN AVERAGE OPPONENT and if you think that's the end it's not, you're a moron if: 5.THINKING THE GAME WAS ACTUALLY CLOSE BECAUSE THE FINAL SCORE WAS 24-17 MAKES YOU A COMPLETE FUCKING MORON WHO DID NOT WATCH THE GAME AND HAS NO FUCKING CLUE WHAT THEY'RE TALKING ABOUT.

:crazy:

 

[Steve19]

Every model contains bias based on the criteria selected. Maybe the bias is more accurate for what you are measuring, maybe it isn't. The fact is that the goal of the all the models is to create an algorithm by which to determine a very difficult subject: how to rank 117 teams across a long time span with no direct comparisons between each of the teams (ie they can't play each other on a neutral field every week). Do the BCS models include a variable for the number of points allowed by the defense? If not, I think they are missing what I think to be an important measurement for a good team, thus they are biased.

No wisdom here. Had enough stats classes to know their application to phenomenon as complex as college football is fraught with misapplications and abuses. Question for the esteemed professors is how many of them have had any philosophy of science classes? Way too many people out there using stats without the requisite understanding of the foundation and assumptions that underlie it.

But if the computer polls don't have biases, why is there variation between them? Look at Virginia. Some have them 6th, one has them 17th! The fluctuation is because each chooses a different model, with different assumptions, and thus different biases. Cooley for example assumes that point spreads shouldn't be taken into account (good for Oklahoma after last week, huh). But wouldn't some consideration of point spreads give a more robust finding? He assumes it away (not without good reason btw). But his assumption has introduced bias.

Finally, SOS is taken into account I believe in some of the polls, though through an indirect means.

Contrary to what is said in an earlier post, I have no desire to best anyone on BP. After all, Clarity's vision was about a place where Buckeye fans could come freely and feel at home, speak their minds, and relate to one another.

You've probably noticed a number of posts by the BP staff about the constant whinging that has virtually taken over some conversations on BP. Although I wasn't the first to speak about it, I have come to agree with this point of view. It seems to me that the best way to stop the practice is for us to interogate our own points of view when a thread starts down that road.

If there is objective evidence of bias against us, then by all means then I think we should talk about it (i.e., the ESPiN attack on Ohio State a few years ago). But otherwise, it just becomes an irritation and keeps conversations here from moving forward. So, in that vein, I will respond to the points you raise.

1. Your point in the earlier post was not that the computer rankings are biased in terms of their weighting of different criteria, but that they are weighted against Ohio State and the Big Ten. If I understand you correctly, I think we can now agree that we can set that argument aside now?

2. The statistical models used in the BCS computer rankings appear to typically employ Bayesian Markov chain Monte Carlo models these days and, to a lesser extent, logistic regression.

The models are indeed complex but the advent of faster chips and more advanced generation languages, such as the implementation of Bell Lab's S language (e.g., in the R statistical package), makes it much easier computationally to implement even complex models. Things have speeded up dramatically and what took more than a day as recently as 2000 now will run in less than 15 minutes.

The BCS actually would be a rather simple model to implement and the dataset would be rather modest, compared to what marketing modelers often do in database marketing or what financial modelers do in technical analysis of securities. Consider, for example, a bank that is modeling the transactions of 100,000 randomly chosen credit card users over the span of a year. One credit bureau that I managed in South Africa frequently produced credit scores using Bayesian statistics for a data set including eight years of credit data for 42 million consumers. So, BCS really is a very small and simple model to implement in comparison.

3. I don't think anyone knows exactly what is in the models, as the statistical model comprises the family jewels and the modelers don't disclose much. As I think you know, people build models by including lots of data and then begin to drop variables that don't contribute to predicting a game outcome. So, it is not surprising that models would include different variables, because as you note, they start with different variables and drop different variables, depending on their analysis.

Not sure about defensive scoring and models. If a model includes the scores of a game, then points scored against a team would be captured in the model. I am not sure how many models would be able to differentiate between turnover points scored against the offense, though, and I am not sure that it is really relevant. Would it really matter how points are scored when the scores are added to determine a winner, in terms of the objectives of someone trying to model which teams will win in future games?

Thanks for the discussion guys.

 

[MililaniBuckeye]

West Virgina (6-1) #6, best wins @ Maryland...Mississippi State (?)
South Flordia (6-1) #11 best wins @ Auburn, vs West Virginia
Makes sense to me

Yet another great example of voters going on the timing of the losses rather than to whom the losses were...

 

[bkochmc]

Yet another great example of voters going on the timing of the losses rather than to whom the losses were...

At the same time isn't there something to be said about timing of losses? I hate to use them as an example but look at Michigan... they are playing much better ball right now than they did the first two weeks of the season. If they played then like they are now they would have beaten App State (though I will get infinite joy out of that loss) and most likely would have given Oregon more of a game (though Oregon still would have won because they are a better team).

It's very tough to accurately gauge where teams should be ranked, both human and computer.

 

[Waller]

Contrary to what is said in an earlier post, I have no desire to best anyone on BP. After all, Clarity's vision was about a place where Buckeye fans could come freely and feel at home, speak their minds, and relate to one another.

You've probably noticed a number of posts by the BP staff about the constant whinging that has virtually taken over some conversations on BP. Although I wasn't the first to speak about it, I have come to agree with this point of view. It seems to me that the best way to stop the practice is for us to interogate our own points of view when a thread starts down that road.

I read that post as well. My understanding was that it was more directed towards people bashing the players (really tired of all the posts ragging Maurice), but I can't find it now, so maybe your take is more accurate. Either way, I don't see too much bickering, and WAY less than on other boards. If my reply to your post was inappropriate, my apologies.

1. Your point in the earlier post was not that the computer rankings are biased in terms of their weighting of different criteria, but that they are weighted against Ohio State and the Big Ten. If I understand you correctly, I think we can now agree that we can set that argument aside now?

Yes, as I have not stated they are biased against us now. My issue was with the blanket statement that computer [programs] are not biased. That is what I responded to. Rather, I think the models are biased more along schools of play, or coaching approaches. Sometimes that goes against OSU (eg 2002), and sometimes not (eg 2006). But the models contain bias because of the variables they choose to use, and the ones they don't use.

2. The statistical models used in the BCS computer rankings appear to typically employ Bayesian Markov chain Monte Carlo models these days and, to a lesser extent, logistic regression.

The models are indeed complex but the advent of faster chips and more advanced generation languages, such as the implementation of Bell Lab's S language (e.g., in the R statistical package), makes it much easier computationally to implement even complex models. Things have speeded up dramatically and what took more than a day as recently as 2000 now will run in less than 15 minutes.

The BCS actually would be a rather simple model to implement and the dataset would be rather modest, compared to what marketing modelers often do in database marketing or what financial modelers do in technical analysis of securities. Consider, for example, a bank that is modeling the transactions of 100,000 randomly chosen credit card users over the span of a year. One credit bureau that I managed in South Africa frequently produced credit scores using Bayesian statistics for a data set including eight years of credit data for 42 million consumers. So, BCS really is a very small and simple model to implement in comparison.

Yeah, what he said. Bayesian logic is beyond me. Only thing I remember about Bayesian logic is that if I am ever on Let's Make a Deal, I always want to make the switch to the other prize. Regardless, whether they are using Bayesian logic, maximum likelihood estimate, or anything else, there is bias.

3. I don't think anyone knows exactly what is in the models, as the statistical model comprises the family jewels and the modelers don't disclose much. As I think you know, people build models by including lots of data and then begin to drop variables that don't contribute to predicting a game outcome. So, it is not surprising that models would include different variables, because as you note, they start with different variables and drop different variables, depending on their analysis.

Actually, at least two of the models are quite tranparent (Massey, Colley). Interesting reads actually, and one of the computers does take into consideration conference strength (I believe the phrase he used was a rising tide (team) lifts all boats). I wonder if these guys actually like college football?

Not sure about defensive scoring and models. If a model includes the scores of a game, then points scored against a team would be captured in the model. I am not sure how many models would be able to differentiate between turnover points scored against the offense, though, and I am not sure that it is really relevant. Would it really matter how points are scored when the scores are added to determine a winner, in terms of the objectives of someone trying to model which teams will win in future games?

Thanks for the discussion guys.

Good point. What I have taken from this chat is that I would really like to know more about the models being used, but also, what exactly are we measuring? Who could beat who on any given Saturday? But football is all about the matchups, something no model is really able to capture in any direct way. Know of any good chat rooms discussing the philosophy of "best football teams"?:)

GO BUCKS!!!

 

[Carmen Ohio]

If we don't crush Penn State, does BC jump OSU in the BCS?

 

[MililaniBuckeye]

If we don't crush Penn State, does BC jump OSU in the BCS?

BC didn't exactly crush VT either. Now before you start telling me how VT was ranked #8 and PSU is currently #25, BC had to pull the game right out of their ass. Plus Crappy Valley is as an intimidating place to play at night as there is. The only way we drop below BC is if we lose...

 

[OregonBuckeye]

If we don't crush Penn State, does BC jump OSU in the BCS?

What do you think?

 

[Carmen Ohio]

What do you think?

I'm not sure if you're being sarcastic or not but I asked the question because I'm soliciting informed opinions like MililaniBuckeye's. I don't know what to think.

 

[Carmen Ohio]

BC didn't exactly crush VT either. Now before you start telling me how VT was ranked #8 and PSU is currently #25, BC had to pull the game right out of their ass. Plus Crappy Valley is as an intimidating place to play at night as there is. The only way we drop below BC is if we lose...

I'll buy it

 

[BrutusBobcat]

If we don't crush Penn State, does BC jump OSU in the BCS?

Without getting into too much math, here's the breakdown:

OSU leads BC by .0258 points in the BCS poll.

BC is getting .97 points in computer polls, and are already ranked #1 in three of them. Presuming that they were ranked #1 in all of them, they would get 1.00 (100%) from the computers, which adds .01 to their BCS score. OSU still leads by .0158, presuming that the Bucks lose no points in the computers (currently .93)

Now on to the human polls. Humans don't like to drop #1 ranked teams unless they lose. Currently, OSU is getting .999 points in each, which is near perfect. BC is getting .941 in one and .939 in the other. In Harris, 111 voters out of 114 total have OSU at #1. In Coaches', 58 voters out of 60 have OSU #1. Since no voter would reasonably drop OSU below #2, we can estimate how many voters would need to change their votes in order to make up the remaining .0158 that is left in the worst-case computer scenario.

Splitting evenly between the two polls, that drop would require 36 Harris voters and 18 Coaches' voters, or right around 1/3rd of the poll voters waking up on Monday thinking that BC is the best team in the country.

That's a very long-winded way of saying "no, not a chance". Even given a heart-stopping close win (like BC's win over VT) OSU remains #1 in the BCS.

 

[OregonBuckeye]

I'm not sure if you're being sarcastic or not but I asked the question because I'm soliciting informed opinions like MililaniBuckeye's. I don't know what to think.

I was being sarcastic. No, they won't jump us as long as we win. The only scenario I could think of where we won and got jumped is if we benefited from an Oklahoma-Oregon level mistake by the refs. Even then, I doubt it. BC isn't a favorite amongst voters.

 

[Nordberg]

If we don't crush Penn State, does BC jump OSU in the BCS?

I thought BC looked like poop. I was thinking last night that if we played BC, our defense would crush Ryan. He may get lucky 1-2 times, but they can't run the ball and our defensive line would eat up their O-line. I don't think we need to crush Penn State, we just need to win. Even if they do jump us. Their schedule isn't that great down the road. They have Florida State (4-3, 1-3), @ Maryland (4-3, 1-2), @ Clemson (5-2, 2-2), & Miami FL (5-3, 2-2). Our Schedule is much better. @ Penn St (6-2, 3-2), Wisc (6-2, 2-2), Illinois (5-3, 3-2) & @ scUM (6-2, 4-0).

I think if both teams win out. Our schedule would suggest to the voters that we are the better team. But as everyone knows, we just need to be #1 or 2 to go to the big game. So either way, I think our remaining schedule should keep us up in those two slots, regardless.

 

[Gatorubet]

I thought BC looked like poop. I was thinking last night that if we played BC, our defense would crush Ryan. He may get lucky 1-2 times, but they can't run the ball and our defensive line would eat up their O-line. I don't think we need to crush Penn State, we just need to win. Even if they do jump us. Their schedule isn't that great down the road. They have Florida State (4-3, 1-3), @ Maryland (4-3, 1-2), @ Clemson (5-2, 2-2), & Miami FL (5-3, 2-2). Our Schedule is much better. @ Penn St (6-2, 3-2), Wisc (6-2, 2-2), Illinois (5-3, 3-2) & @ scUM (6-2, 4-0).

I think if both teams win out. Our schedule would suggest to the voters that we are the better team. But as everyone knows, we just need to be #1 or 2 to go to the big game. So either way, I think our remaining schedule should keep us up in those two slots, regardless.

Win and you're in. No worries. (Hopefully you'll be in at #2, which is a better predictor of victory in the damn thing)

 

[Yertle]

I thought BC looked like poop.

By golly, me too! They looked like gosh darn heep of donkey poop.