collegefbfan8898's Objective Poll

[..fanatic]

The point I'm trying to make, and apprently failing miserably at, is this....

If I am creating a computer ranking, I'm doing so because I want to know how teams would be ranked if no bias were involved. Therefore, my primary goal is to be certain I don't introduce anything that is biased. Thus, the assumption that computer rankings are biased just because humans wrote the formula, is misguided. The assumption should be that they are, indeed, unbiased or they would still be trying to finish what they set out to do.

IF it is NOT your goal to produce an unbiased ranking, you're just trying to figure out a way to have a computer HELP you rank teams. That is not the way people like Massey and Sagarin and Congrove and others attacked the problem. Most of these people began these endeavors as a college thesis and the specific problem they set out to solve was "How would teams be ranked if no bias was applied?" When you see their rankings in the papers and on websites, it is because they succeeded in proving their theory was, inded, unbiased, and produced reliable results. Yes, they did so in different fashions, but that doesn't mean any of them aren't unbiased.

[donovan]

Give me an example of an unbiased formula......

[Dossenator]

I am with Donovan. I think it is impossible. Someone still has to decide what is important and what is not. The data itelf is not biased....either you win or you don't...etc. But what percentage each gets, and what gets put in the equation and what gets left out, etc. Maybe I am not smart enough to see it any other way......my head is hurting.

[..fanatic]

Give me an example of an unbiased formula......

I already did - halfway down page 2 of this thread.

[Spence]

Maybe bias is a bad word. Different people use different criteria for producing a ranking. That, of course, is why you get different results in computer rankings. I believe that most people how do rankings (and vote in polls for that matter) do so to get a correct result. To make the system better. I don't believe that people go to all the time and trouble to set something like this up if they only wanted to tip the balance to a certain team or conference. It wouldn't make sense to do that and the results wouldn't hold up over time.

My opinion of a good solid computer ranking would be one that plays out the games (multiple times) to get the best guess on who wins when they play the game. Most good computer rankings do just that. Then they start picking individual games and match them to the actual scores. If they get a high percentage of right predictions, over time, then they can reasonably rank teams. I think you would have to have a system that plays out the games hundreds, if not thousands, of times to get enough data to be reasonably sure about how the games will turn out. No system is going to ever pick over 80% over time because I don't believe you can ever program "intangibles" into the system. Certain individual performances by a player or by a team cannot be predicted by man or machine. That much is random and it probably amounts to somewhere between 20 to 30% of the probability. That isn't necessarily a bad thing. Few people would pay attention to a game if the winner was predetermined.

Games like USC - Stanford, Michigan - AP State, Notre Dame-Navy. USC, Michigan, and even Notre Dame have more talent then their counter parts in these games. Especially USC -Stanford. If they play that game one hundred more times, SC probably wins 99 of them.

My point is that the perfect system isn't out there. It will never be out there. That doesn't mean people shouldn't keep trying to come up with a system that is more perfect then the current system. People who do computer rankings, I'm sure, try to find a way to make their systems better. They keep trying to find a way to boost their percentage of correct games picked up to the next percentage point. I see nothing wrong with trying to make a system better. Nothing wrong with trying to create a better system. That is how progress is achieved.

I could list a poll of my top 25 and people could go over it and find things in it they deem "wrong". People do it every day. Whole websites are dedicated to tear down other peoples polls or rankings. We all think we know who the best teams are and who would beat who. The fact is that we don't really know who is right until the games are played. Our prediction game proves that it is hard to predict games much over 70%. Only one person that has ever picked for a season has picked over 75% for the season. Most hang out between 55 and 65%. So the fact that we look at a poll or a ranking and decide if it is valid or not is more a reflection on our confidence in our abilities to understand the game then it is on how good the poll or ranking may be.

The good thing about most computer polls is they actually publish their results. There actually is a criteria in place that treats all teams in the same manner. The BCS should use the actual results of the computer polls instead of trying to make them adhere to being more like the human polls.

The big difference I see between the computers and the human polls, is that the computer tries to remove all prejudice by treating every team the same and the human polls try to put in enough prejudiced opinion from all regions of the country that the bias is removed and you get a common result from the group. If you think about it, they both basically put the same teams in the same groupings. Maybe the order is a little different, but the top 25 teams are basically the same in all polls and rankings. It is the order that is a little different. Meaning that all of these systems work pretty well, the argument is about finding the best one. That is what the argument always should be about.

[donovan]

Give me an example of an unbiased formula......

I already did - halfway down page 2 of this thread.

Ok...

One crude form of an unbiased formula.
1) Wait until every team has played 3 games. And what human dedcided that the first three games would not figure in ..why not two..or four....we know that early rankings affect the later ones.

2) Every team starts at "0". Ok...as long as somewhere down the line you do not seed them..as in SOS
3) Enter how much each team won or lost every game by. and who decides a 20 point win is better than a 5 point one...can think of all kinds of reasons why scores are kept in check...who decided that
4) Divide by the number of games played ....all games..regardless of the opponent...Division 1 an 11 are equal..who made that decision...certainly not the intent of the NCAA Division people.
5) Multiply by the number of wins....and you come up with a biased ranking......we do it all the time....gravity formulas are based on bias.....live with it..just go to a 12 step program and deal with it....and who decided 12 steps..Oh..Bill W...I digress
You have a crude ranking.

[Spence]

At some point you have to have a way to measure the strength of a team. If you only go by who wins the most games, because teams play different opponents, you would not ever have a conclusion. That would render rankings and bowl matchups impossible.

[Dossenator]

Exactly Spence...I said the only way for an accurate poll is to have every single team play every other team...that can't be done. So no matter what a computer says it is still a guess. You don't know if the teams don't play. We say last year how well the computers did....Florida vs Ohio St....wasn't even a game. I bet the computers did not play the game out like that.

If every computer poll comes up with different rankings how do we then determine which computer poll is correct and which is not? So, we have to average all of the computer polls together with the human polls because we know none of them are correct on their own. They all typed in unbiased data....why do that not all say the same thing. Because a human decided what was important and what was not important. The people who create the computer poll formulas are determing way to much. We have basically put who plays in the BCS games in the hands a few people.

As for the formula you gave...well I see many wholes in it. How does it factor in when you put all the bench players in for an entire half and they only score 14 more points...the starters could have scored 40 more. I am again with Donovan...why is a 1 point win better then a 20 point win.

I don't think I am ever going to think about college football polls again. I will now they are there and be excited when Ark is included in them. I will leave how they come up with the rankings to others to debatel

[Spence]

Exactly Spence...I said the only way for an accurate poll is to have every single team play every other team...that can't be done. So no matter what a computer says it is still a guess. You don't know if the teams don't play. We say last year how well the computers did....Florida vs Ohio St....wasn't even a game. I bet the computers did not play the game out like that.

They probably did have some games like that for Florida. It just wasn't the higher probability. If they played out 1000 games they probably had a small percentage of blowouts for both teams. The computers use the highest probability of results to produce their prediction so that result probably wasn't used in most models.

No poll or ranking will give you a locked in stone result and I don't think we would really want that. I wouldn't want to watch a game that a computer gave me 100% certainty of who would win. We would have no reason to play the game. It is that chance that our team pulls it out, that is the reason we watch. It is that one time in forty four times that Navy beats Notre Dame. If we had a way of predicting with 100% certainty who will win a game that game wouldn't be fun. It wouldn't be worth playing and it wouldn't be worth watching.

We try to put the best system in place to rank teams. It won't always give us the desired result and it won't always give us the right result. I don't think we need that. We just need a system that acts fairly to all teams. One that measures the strength of a team and creates the best match ups. The current system probably doesn't do that, but they are trying to do it and they keep trying to make it better. That is all you can really ask.

[..fanatic]

Good heaven's people. Why don't you actually read the entire original post, and understand it in its proper context, before you bash it. Here's the post you are wrongly referring to. Note that it never said it is the way to actually create a ranking. See the underlined sentences. It was only directly answering to Donovan's comment that, somehow, all formulas have bias. I was simply showing basic ways in which that's not true.

I can not think of one computer program I have ever used....save balancing numbers.....that did not have programmers bias. None. This rages from Microsoft Word....tons of bias...to medical models...even more....

Give me an example of what you mean by "bias" because I'm not seeing the impossibility of eliminating all bias. "Bias" to me would be an assumption that you should give more credit to a conference, for example. - In that case, you are prejudging a factor and you are blatantly biased.

But how is a spreadsheet bias before you tell it to do anything? It's blank.

One crude form of an unbiased formula.
1) Wait until every team has played 3 games.
2) Every team starts at "0".
3) Enter how much each team won or lost every game by.
4) Divide by the number of games played
5) Multiply by the number of wins
You have a crude ranking.

Team "A" won 3-out-3 by an average of 24 points = rating of 72
Team "B" won 2-out-of-3 with a total positive spread of 17 points (17 points divided 3 games x 2 wins = 11.33 points.

SOS can be determined by addding the rating of all of the teams played and dividing by the number of games (3, in this case).

I'm not putting this in a spreadsheet right now, so I don't know what the heck you'll get for answers. It's probably total crap. But all I'm trying to demonstrate is that formulas and answers can be totally unbiased and no computer program can make it not so.

Therefore, if you use you the right facts (without assuming weight to different criteria) and avoid "prejudging" what should be more important, you can create an unbiased ranking formula that makes sense. The trick is to find the facts and the formula that work in combination. You can judge its accuracy by seeing how it performs in picking pointspreads. Is it 50% right or 75% right or 25% right. Can you make it 80% right? How? That would seem to be the process. You can't judge it by putting it together and then asking yourtself if the rankings "look" right. You have to see if it performs with any reliability. And, all the while, maintaining an unbiased approach. If, through application, you find that weighting a factor improves the reliability in picking winners or spreads, than you are not necessarily inputting "bias" because this wasn't prejudged, it was proven.

[donovan]

In parting....there is a reason ivory soap is 99 and 44/100% pure.

[Spence]

To see if any ranking works, you have to test is over time. When you play the games out and get your predictions and your ranking order, you have to keep the results and see how you do. That is the only way you will know if your ranking is solid. That is all I am trying to say. I'm not saying you are right or wrong, just that you need to track your results to test your ranking. If you formula works it will work in every situation, every year at a high percentage. If it doesn't hold up then you have to work on it. Any ranking can pass the looks test, but it doesn't mean it will stand up over time.

[..fanatic]

To see if any ranking works, you have to test is over time. When you play the games out and get your predictions and your ranking order, you have to keep the results and see how you do. That is the only way you will know if your ranking is solid. That is all I am trying to say. I'm not saying you are right or wrong, just that you need to track your results to test your ranking. If you formula works it will work in every situation, every year at a high percentage. If it doesn't hold up then you have to work on it. Any ranking can pass the looks test, but it doesn't mean it will stand up over time.

100% true.

I think the word "bias" has been used too liberally in this argument. It is one thing to be biased toward a set of numbers over another. It's a completely different thing for a poll to be purposely biased toward certain teams or conferences. This is where the computer folks can claim "unbiased" with some credibility because they are not intentionally stacking the deck in anyone's favor. If they devise a formula that passes the test of time in accuracy, but is biased toward a certain set of numbers, I wouldn't call that "bias" - I'd call it "applied logic".

As Spence so aptly pointed out, the "accuracy" isn't determined by comparing it to other rankings - it's looking at its reliability factor in predicting game results and adjusting properly to contrary results.

Geez, I feel so "CLF-ish" (remember Colorado Loves Football) in this thread, I may just have to go shoot myself.