Don't Understand Congrove's Algorithmns

[Spence]

Something I have always wondered about with rankings, why doesn't the computer play off every team against every other team to determine preseason position. Now, what I know about setting something like that up is nothing. Still it seems to me that if the computer played all the games out and then used the real games as a learning curve to adjust as the season wears on that we would get a more accurate top to bottom ranking. Not that how a team does against their schedule isn't important, but how they would do against every other team seems to me to be what the rankings should be about. Not how they do with their schedule compared to how other teams do against theirs. I guess that is what power ranking are there for, but it seems to me that playing all the teams off every week and adjusting for how the season turns out would be a more accurate way of determining the rankings. It also sounds like a lot of work.

[CFP Admin]

Just what argument is that other than saying Ball State has no wins over non-MAC teams is pretty far off. They, in fact, have 249 such wins. I think saying they could beat a big team is fair at this point.

Isn't this all getting just a little silly?

Fine - billybud wasn't more specific in what he was obviously trying to state - Ball State has beaten no one of consequence outside of their conference, at least not in the modern era of 1-A/1-AA or FBS/FCS subdivisions which began in 1978.

I was wrong. I went through Howell's data too quickly.
Here are the 1-A teams Ball State defeated since 1975. Some were affilaited with conferences and those are highlighted in red:
1975 - Indiana State (5-5) (Independent), Richmond (5-6) (Southern), Illinois State (2-7-1) (Independent).
1976 - Indiana State (3-7) (Independent), Dayton (4-7) (Independent), Louisiana Tech (6-5) (Southland).
1977 - Indiana State (3-7) (Independent), Appalachian State (2-9) (Southern), Illinois State (3-7-1) (Independent), Villanova (4-7) (Independent)
1978 - Indiana State (3-8) (Missouri Valley), Illinois State (2-9) (Independent)
1979 - Illinois State (3-8) (Independent)
1980 - Indiana State (6-5) (Missouri Valley)
1981 - McNeese State (7-3-1) (Southland), Illinois State (3-7) (Missouri Valley)
1982 - Illinois State (2-9) (Missouri Valley)
1983 - Wichita State (3-8) (Missouri Valley)
1985 - Indiana State (4-6) (Missouri Valley)
1988 - Northern Illinois (7-4) (Independent)
1991 - Navy (1-10) (Independent)
1996 - UCF (5-6) (Independent)
2001 - Connecticut (2-9) (Independent)
2002 - Connecticut (6-6) (Independent)
2007 - Navy (8-5) (Independent), Western Kentucky (5-7) (unofficial FBS Transition team)

Something tells me this won't impress billybud. Seriously - why should it?

Should Ball State be ranked #40 if they finish with a 9-3 record? Reality is they will be in that ballpark, give or take a few notches, with most rankings if that becomes reality.

Should Ball State never be projected to possibly score an upset until they've actually scored one? It wouldn't be any fun to miss the opportunity to pick the upset. The computer has done it successfully numerous times.

Did the computer say Ball State would score any major upset this year? No - the schedule doesn't even present such an opportunity. It says they will beat Navy but Ball State is 2-0 against Navy. Their toughest game is probably the one at Indiana. The Cardinals are 0-4 all-time versus the Hoosiers but a Ball State win wouldn't set the football world on fire.

Bigger issue to billybud is probably the computer's forecast of a 6-6 season for Florida State. Well, it did nail their 7-5 season last year, but it missed terribly on an 11-1 pick the year before that. Four of the projected losses this year are at home by 2.94 points or less to Wake, VT, BC, and Clemson. Win 2 or 3 of those and the season starts to look totally different.

As for Ball State, the computer nailed their 2005 and 2006 records but underestimated them by 3 wins last year.

[Eric]

Don't stop Ktffan, this is getting good

Seriously, I don't think we have to get into semantics (that word's being used a lot around here ). Big deal if Ball State beat teams from the Southland pre-1978. It's not of any real relevance to the team today.

[CFP Admin]

Something I have always wondered about with rankings, why doesn't the computer play off every team against every other team to determine preseason position. Now, what I know about setting something like that up is nothing. Still it seems to me that if the computer played all the games out and then used the real games as a learning curve to adjust as the season wears on that we would get a more accurate top to bottom ranking. Not that how a team does against their schedule isn't important, but how they would do against every other team seems to me to be what the rankings should be about. Not how they do with their schedule compared to how other teams do against theirs. I guess that is what power ranking are there for, but it seems to me that playing all the teams off every week and adjusting for how the season turns out would be a more accurate way of determining the rankings. It also sounds like a lot of work.

Dang it, Spence - Now I gotta go play with numbers.

[Spence]

Don't stop Ktffan, this is getting good

Seriously, I don't think we have to get into semantics (that word's being used a lot around here ). Big deal if Ball State beat teams from the Southland pre-1978. It's not of any real relevance to the team today.

That is what should be the point. It doesn't matter how good or bad you were yesterday. It is how good you are today that counts.

[Spence]

Dang it, Spence - Now I gotta go play with numbers.

It won't give you the outcome your predictions do, because rarely do the best teams ever meet in the same year, but it does seem to me it would give a more accurate accounting of how teams should actually be ranked. The problem I see is that a team with a couple losses could rate out stronger then a good team with no losses or one loss and all the people would look at that ranking and cry bias even though the computer would just be ranking according to how a team would fair against every other team. That ideally is what a ranking should be.

[CFP Admin]

Something I have always wondered about with rankings, why doesn't the computer play off every team against every other team to determine preseason position. Now, what I know about setting something like that up is nothing. Still it seems to me that if the computer played all the games out and then used the real games as a learning curve to adjust as the season wears on that we would get a more accurate top to bottom ranking. Not that how a team does against their schedule isn't important, but how they would do against every other team seems to me to be what the rankings should be about. Not how they do with their schedule compared to how other teams do against theirs. I guess that is what power ranking are there for, but it seems to me that playing all the teams off every week and adjusting for how the season turns out would be a more accurate way of determining the rankings. It also sounds like a lot of work.

Dang it, Spence - Now I gotta go play with numbers.

As I started messing with this, I was quickly reminded by myself that I kinda sorta began the process this way. First problem I ran into is it throws SOS out the window if everyone is playing everyone in theory. Well, the theory and the reality don't mesh. The reality is everyone doesn't play everyone so the purpose is to rank teams and predict outcomes based on that reality.

[Spence]

I know, it goes back to the not enough info argument. For purposes of ranking teams you have to consider how teams do against their schedule vs. how the other teams do against theirs. It just doesn't give you a true 1-119 ranking, but if you put out a real 1-119 ranking then people would cry bias because a team with more loses would at some point be ranked ahead of a similar team with less losses. People aren't ready for that kind of hard logic. No one would take it seriously.

By the way, nice job with the previews. Lots of good info. Anyone who hasn't checked it out should.

[CFP Admin]

Here's my theory, to some degree.

Upper echelon teams have all the help they need - higher profiles for better recruiting, better conference affiliations for better TV contracts, better money to create even higher profiles and recruiting and go around the circle again and again. So the question becomes how do you rank the disadvanatged among the advantaged.

Well, if a Ball State (with seemingly so little going for it) can crack through with a 9-3 record against a slate of opponents it compares to, and rich Alabama with years of history behind it can only manage 7-5 against a slate of opponents that it compares to, what's wrong with Ball State being rewarded for 9-3? I mean, really, the Ball State's of the world should not be expected to be able to compete against the Alabama's - there are a multitude of disadvanatges weighing them down. A 9-3 season for Ball State may not seem like much for Alabama fans becasue if 'Bama had the same schedule, they'd probably go 12-0. But Ball State got that 9-3 season against its own peers, and 'Bama went 7-5 against its own peers. The best you can do is take apples and oranges and try to make them apples and apples, or oranges and oranges.

[CFP Admin]

By the way, nice job with the previews. Lots of good info. Anyone who hasn't checked it out should.

Thank you very much!

[Eric]

Here's my theory, to some degree.

Upper echelon teams have all the help they need - higher profiles for better recruiting, better conference affiliations for better TV contracts, better money to create even higher profiles and recruiting and go around the circle again and again. So the question becomes how do you rank the disadvanatged among the advantaged.

Well, if a Ball State (with seemingly so little going for it) can crack through with a 9-3 record against a slate of opponents it compares to, and rich Alabama with years of history behind it can only manage 7-5 against a slate of opponents that it compares to, what's wrong with Ball State being rewarded for 9-3? I mean, really, the Ball State's of the world should not be expected to be able to compete against the Alabama's - there are a multitude of disadvanatges weighing them down. A 9-3 season for Ball State may not seem like much for Alabama fans becasue if 'Bama had the same schedule, they'd probably go 12-0. But Ball State got that 9-3 season against its own peers, and 'Bama went 7-5 against its own peers. The best you can do is take apples and oranges and try to make them apples and apples, or oranges and oranges.

Ding, ding, ding! Thank you!

That's something many don't necessarily understand. Yeah, when looking at Ball State from a national scale, they might be the 60th best team in the country or something like that (they should, in all honesty, be one of the best non-BCS teams this side of BYU and Utah). But it's all relative. A win against Indiana may not seem like a big deal to a school like Michigan, but it might be to a smaller school like Ball State.

[billybud]

Ding! Ding! NO THANK YOU

Is a 9-3 Ball State comparable to a 7-5 Alabama? That is the question. When we rank the totality of IA teams, we are ranking them against their "peers" and peers means all 120 or so teams.

How a team does against its peers (defined as the schedule) vs how another team does against its peers is not a measure that makes sense...not when you are purportedly providing some kind of way to measure that apple against that orange...and talk about "rewarding" for being a smaller program means affirmative action, that is, action not supported by performance (performance being a measure of strength vs total IA team strength).

Measuring how Ball State does against Akron, Buffalo, Kent State, et al.....has a much, much different meaning of strength than how Alabama does with Georgia, Clemson, Florida, Tennessee, LSU, and Auburn...that murderer's row of football.

A 7-5 Bama, playing maybe five or six end AP ranked teams, would be a stronger football team than a 9-3 Ball State whose season highlight may be playing Navy. There is no comparison that I can think of that makes sense based on how they do against their peers. Lincoln High School in Tallahassee will go 9-3 this year...how they do against their peers has no relevance to a 7-5 Bama team...and while Ball State isn't as far removed from Bama as Lincoln High School in degree of difference, the concept is the same.

We all know the better team when teams play each other. We bestow that title upon the winner, for that day, anyway. What ranking systems attempt, is to give us a way to compare teams that do not play each other. I look at it straight on...how would a team do in a game against a team. That is what I want a system to simulate.

[billybud]

If you are just ranking teams with the primary focus on how they do with their schedule, than SOS is a minor player. Sure, its fine to recognize teams that achieve in their particular sphere...but when the "sphere" being compared is the total of IA, than a system needs to rank all 120+- teams on the same parameters. Apples that play five or six end ranked teams vs Oranges that may play no team in the top 60.

Wins are not just wins. As I have said before, you should get a whole lot more points for degree of difficulty when doing a triple twisting full Gainer than you do with a Cannonball.

It appears that maybe CCR is talking about "norming" the records...that is exactly what we do with almost all employment testing. We separate test scores by racial group so that one is only graded against those who are in the same category. A ninety for one group may be the same as a 75 for another group. So, when you rank applicants, the applicants are ranked not on how they performed on their tests against the total group of applicants, but on how they performed against applicants of their same demographic. A 90, 75, 70, 85, 80 might be the actual order...with some ranked higher than individuals who actually had higher scores.

And..LOL...CCR..I have FSU at 7-5 this year...I would be thrilled with an 8-4. Getting Virginia Tech from the other Division this year makes for a little tougher schedule...VT, Colorado, Wake Forest, Florida, Clemson, BC, and Miami will all play the Noles tough.

[Spence]

That isn't how rankings work at all. The computers (and humans) are not graded against those who are in the same category. The grading is equal, it is the test that is different. The power rankings are put in to try and make the test equal.

The flaw in the system isn't how the rankings operate. It is that they don't have enough information. No system can work perfectly without the necessary information to make an equal comparison. What CCR and most other rankings are doing is taking the best info they have and trying to apply it in a non biased way to all 119 teams. By their own admission they are not 100% accurate. CCR is 74.7% accurate. The others are + or - that, but comparable. But they are trying to apply the same logic to all teams equally and get an accurate picture of who should be where. Not based on past history or what you or I may think is correct, but by comparing them in the now. When you think about all the unknowns in the set up these guys (all of them) do a pretty good job. The difference in CCR compared to some others is that CCR has refused to change their system based on what the BCS committee wants. The others involved did. If you believe in what you are doing to be the correct way then you shouldn't change for anyone. Even if it means that your formula is not part of the system. That is the big problem I have with Sagarin and company. He conformed not based on what he believed, but based on what the committee wanted.

[billybud]

I agree re changing to suit BCS, Spence...even Sagarin states that his BCS rankings are not as predictive as his "Predicter" rankings.

The question isn't about the tests per se, it is about Norming. When you look at teams in any way that doesn't reflect the totality of their position within the 119 teams of IA, norming sneaks in.