collegefbfan8898's Objective Poll

Say it all here
Forum rules
NOTICE: Please be sure to check the CFP Message Board Rules and Regulations and the Read Me page before posting.
User avatar
donovan
Athletic Director
Athletic Director
Posts: 8634
Joined: Wed Oct 04, 2006 6:41 am
Location: Pacific Northwest

Re: collegefbfan8898's Objective Poll

Postby donovan » Tue Nov 06, 2007 5:52 pm

4. statistics distortion of results: the distortion of a set of statistical results by a variable not considered in the calculation, or the variable itself

it may have been used too liberally in the sense of too often...but it is the statistical term used to describe..or the attempt to on the subject that has passed my boredom stage...
Statistics are the Morphine of College Football

User avatar
Spence
Administrator
Administrator
Posts: 20970
Joined: Wed Dec 28, 2005 11:52 pm
Location: Chillicothe, Ohio (Ohio's First Capital)
Contact:

Re: collegefbfan8898's Objective Poll

Postby Spence » Tue Nov 06, 2007 6:05 pm

I thought we were on the same page other then my use of the word bias. :wink: My use of bias was probably to strong a word to desribe the favoring of one set of numbers of another set of numbers in the overall criteria used to set it up. I do fell that most computer rankers (is that a word :lol: ) do not set out to favor any certain team over another. The only "differences" are the question over what statistics and facts are used to create the rankings and that would be at the discretion of the ranking's creator.

That is the opposite of the human polls. The human polls try to eliminate bias by inserting lots of bias from all over the country. That system doesn't work that badly either. The bad thing about the human polls is that they tend to front load to much past history into the polls and it takes them a while to fix that, where the computers just spit out results from data without regard to the common opinion.

The human polls will get a major test in proving whether or not they can be fair this year if Kansas beats Missouri and Oklahoma to win the B-12. No matter what the common opinion is at this time, if Kansas can beat Missouri and Oklahoma and finish the year undefeated they should play in the championship game. They will have beaten three ranked teams and will have ran the table in a top 6 conference. By taking care of all of their business they will, in my eyes, prove they belong. And they would have done it the hard way, coming from being unranked.
"History doesn't always repeat itself but it often rhymes." - Mark Twain

User avatar
donovan
Athletic Director
Athletic Director
Posts: 8634
Joined: Wed Oct 04, 2006 6:41 am
Location: Pacific Northwest

Re: collegefbfan8898's Objective Poll

Postby donovan » Tue Nov 06, 2007 6:35 pm

..I don't think any of us are ever too far apart.....just need conversation. I do think a valid point is if you just do the human thing....you tire....who has time..full time job and if you miss something....so the computer does not tire and in some sense it will be more accurte over the long haul. Billybud always likes the Vegas odds...a ranking etc by another name...here humans do a lot of it and they do not tire because they are paid full time...Spence has often said just the Top 25 ranking is a lot of work..indeed it is...and I assume most at some point say...ok...here it is spent enough time and really do not know who should be 22 or 23....next subject
Statistics are the Morphine of College Football

User avatar
Spence
Administrator
Administrator
Posts: 20970
Joined: Wed Dec 28, 2005 11:52 pm
Location: Chillicothe, Ohio (Ohio's First Capital)
Contact:

Re: collegefbfan8898's Objective Poll

Postby Spence » Tue Nov 06, 2007 6:50 pm

The big advantage the computers have is that they play the games out virtually according to the statisics entered in and they spit out the most common scenerio among the hundred or more games that are played off. "The Vegas poll" operates much the same way only they use information that is not commonly known by the public and so in cannot be incorporated into the computer polls. They know things about players that their coaches don't know. So it is in fact a computer poll that is done with more intimate knowledge of the team. It can never be part of a BCS equation, though, because of some of the methods used to find out this knowledge.

Doing a T25 poll and trying to be fair is very hard. You have to pay attention to games that you would other wise have no interest in. It does give you more perspective on teams outside your interest. If you do it right it is a real learning experience. You find out that some of the contradictions in the polls that you used to pass off as bias have valid reasoning behind them. Even if you may disagree with the reason.
"History doesn't always repeat itself but it often rhymes." - Mark Twain

User avatar
Dossenator
Moderator
Moderator
Posts: 5293
Joined: Tue Jul 25, 2006 8:00 pm
Location: Back in NW Arkansas!!!!
Contact:

Re: collegefbfan8898's Objective Poll

Postby Dossenator » Tue Nov 06, 2007 9:12 pm

I did not mean that the polls were biased towards certain teams or conferences but with the numbers....giving more importance to one stat or the other. The person who sets up the computer formula has to decide what numbers are more important than others (that is what I was refering to when I was using the word bias).
"A team with something to play for is dangerous, but a team with someone to play for is unstoppable..." Arkansas OL Brey Cook quote following the death of teammate Garrett Uekman (Nov. 2011).

User avatar
openSkies
Administrator
Administrator
Posts: 1288
Joined: Wed Dec 28, 2005 4:10 pm
Location: Boston, MA, USA
Contact:

Re: collegefbfan8898's Objective Poll

Postby openSkies » Wed Nov 07, 2007 3:41 am

Bias in numbers is in everything. As someone pointed out (this thread is waaaaay too long to go back and check who)... what person decided that winning by 3 is not as good as winning by 42? Well, yeah, that's "bias" in the crude sense of it, because you're putting some amount of importance on the point spread. But is it really correct to call that bias? Nah. That's just logic.

And, you know, eliminating numbers "bias" doesn't really matter in my opinion. As long as the computer can produce effective, accurate results, then it's doing its' job. If one day some computer ranking comes out and gets every game 100% correct, how can you dispute that it is the best method for determining rankings and bowl match-ups? Yeah, it puts more importance on SOS than some other number, such as wins/losses, but... WHO CARES?!

I'd like to see some of these ridiculous computer rankings put together by morons, as well as the different voter polls, try and get close to the Congrove Computer Ranking's average.

Hell, it's blowing y'all away in the weekly picks! Haha. (Except you, Donovan, but you don't count in this argument :wink: ).
Image

User avatar
donovan
Athletic Director
Athletic Director
Posts: 8634
Joined: Wed Oct 04, 2006 6:41 am
Location: Pacific Northwest

Re: collegefbfan8898's Objective Poll

Postby donovan » Wed Nov 07, 2007 6:09 am

Well......I am far from an expert in statistics or verbiage, but words are important and it is my opinion that when you are talking statistics and computer modeling of those statistics that it is reasonable to use the correct terms and not to intertwine meanings from the denotation to the connotation.

"Bias" as used in the field of statistics refers to directional error in an estimator. Statistical bias is error you cannot correct by repeating the experiment many times and averaging together the results.

Bias-variance decomposition states that the expected squared error is equal to the squared directional error, or bias, plus the squared random error, or variance. The law of large numbers says that you can reduce variance, not bias, by repeating the experiment many times and averaging the results.

If you see a dog for the first time, and it is black. It does not follow logically that the next dog you see must be black, but black seems like a better guess than any other color. A rigid algorithm in a computer, if it sees a black, may thereafter predict that any dog seen will be black. But this, of course, does not follow logically - though algorithms of this sort are often misnamed "logical". For a purely logical reasoner to call the next dog black as a deductive conclusion, it would need an additional assumption: "All dogs are the same color." This is a wonderful assumption to make if all dogs are, in reality, the same color; otherwise, not so good. "Inductive bias of a machine learning algorithm are the assumptions that must be added to the observed data to transform the algorithm's outputs into logical deductions."

As we sit here and post prattle on what should be included in figuring out who is the best team in the galaxy; I listen to people who do statistics for a living have the same kind of discussions on whether there is "useful" bias or not. There is then always one who boldly states that in statistics bias has a definition and that is becoming distorted and with that another discussion erupts. There is a plethora or arguments on both sides, always bordering on a new definition of boring...until they start using real life examples.

In most fields, and certainly mine, we are forced to look at statistics before we plan a course of action. It did not use to be that way. When a course of action is taken that deviates from what statistics say gives the best results, your outcome had better be correct or you will have more explaining to do than you may want, and it could be in a court.

On the germane subject,look at the computer polling for football teams. There are no two that are even close in accuracy. Why? We know the Congrove model has an accuracy rating that is superior to most, yet, few on this site want to even consider it. Why is that. Because is does not support preconceived desires.

I would suspect that the person that knows more about this subject than all of us combined and certainly more than I care about is Mr. Congrove. (I have never been sure of the relationship between the two with the same last name on this site, but I refer to the one that I think does the computer modeling.) At least he can chuckle over neophytes talking about a subject they know nothing about.
Statistics are the Morphine of College Football


Return to “General Discussion”

Who is online

Users browsing this forum: Bing [Bot] and 83 guests