"...(but) 3 and 4 get complicated"

[FAIRTIDES]

That's exactly what I'm trying to explain to Carl; the fact that the computers are a circular argument by design! Let's say that I start off saying that Team A is good (you need to start somewhere, so that you can evaluate strength of schedule). Therefore, Team B must be good because they beat Team A. Team C had a victory over Team D, but I didn't say that Team D was good, only Team A, so therefore Team C is not as good as Team A.

Now fill in the following names as real examples and maybe Carl will see what I'm saying, with record and Sagarin rating next to it (Baylor = Team A, 10-3, 13th / Texas A&M = Team B, 7-6, 14th / Team C = Houston, 13-1, 15th / Team D = Penn State, 9-4, 28th). So now you see how a team with FIVE MORE LOSSES can still be ahead of a team with 1 loss, and more importantly, considered a top 14 team (if the playoffs included 16 teams like 1-AA, the computers would have A&M in with 6 losses). This is the ridiculous nature of the computer system; they are only as good as the inputs....

It will not work unless everything is "zeroed-out" every year. Schedule and conference strength should be fluid variables, starting at zero, and recalculated every week. If factors like "SOS" become "fixed," based on pre-season notions or perceptions, instead of fluid "variables," the whole exercise will be a waste, little different from what we have now.

 

[UConnDan97]

It will not work unless everything is "zeroed-out" every year. Schedule and conference strength should be fluid variables, starting at zero, and recalculated every week. If factors like "SOS" become "fixed," based on pre-season notions or perceptions, instead of fluid "variables," the whole exercise will be a waste, little different from what we have now.

The problem truly is that there aren't enough games to perform a good computational model. In the example of Sagarin, he uses an Elo-Chess calculation (from Professor Arpad Elo, a formula used to figure out chess ratings). In the Elo calculations, you assume that victories over lower ranked opponents lead to lower increases in your tally and victories over higher ranked opponents lead to higher increases. There is an arbitrary assignment of rankings, and through large multiples of games, you can standardize rankings across a country...or a planet!

But in college football, you get 12 chances to do so (13 with a conference championship). So your arbitrary initial assignment becomes overwhelming to the data analysis! If the teams played 100 games, it might come out in the wash (especially with enough "crossover games" to calibrate assumptions), but clearly that isn't practical for any sport, much less football.

No, I'm afraid computer modeling will NEVER do a better job at removing prejudice from the system than humans, because the inputs are too heavily weighted. Make conferences, and have every conference champion meet in a 16 team playoff with the highest qualified "at-large" teams, and play it out like the FCS does!

 

[FAIRTIDES]

The problem truly is that there aren't enough games to perform a good computational model. In the example of Sagarin, he uses an Elo-Chess calculation (from Professor Arpad Elo, a formula used to figure out chess ratings). In the Elo calculations, you assume that victories over lower ranked opponents lead to lower increases in your tally and victories over higher ranked opponents lead to higher increases. There is an arbitrary assignment of rankings, and through large multiples of games, you can standardize rankings across a country...or a planet!

But in college football, you get 12 chances to do so (13 with a conference championship). So your arbitrary initial assignment becomes overwhelming to the data analysis! If the teams played 100 games, it might come out in the wash (especially with enough "crossover games" to calibrate assumptions), but clearly that isn't practical for any sport, much less football.

No, I'm afraid computer modeling will NEVER do a better job at removing prejudice from the system than humans, because the inputs are too heavily weighted. Make conferences, and have every conference champion meet in a 16 team playoff with the highest qualified "at-large" teams, and play it out like the FCS does!

I am not competent to to engage in any discussion related to statistically significant sample sizes. But:

. The only "input" is provided by a digital feed w/schools, scores, home or away. Every thing else is"internal" to the system.
. Not "modeling." Calculating an actual rank based on who beats who, victory margin, who "who" beats/margins in a string of "who's" equaling 144 (I think) games and results per team.
. Those much smarter than I can decide if the universe (not a "sample") size ='s irrelevance. But, one result of the argument should be a method the zero's out individual or group perceptions related to schedule/conference strength, history and tradition .

 

[UConnDan97]

I am not competent to to engage in any discussion related to statistically significant sample sizes. But:

. The only "input" is provided by a digital feed w/schools, scores, home or away. Every thing else is"internal" to the system.
. Not "modeling." Calculating an actual rank based on who beats who, victory margin, who "who" beats/margins in a string of "who's" equaling 144 (I think) games and results per team.
. Those much smarter than I can decide if the universe (not a "sample") size ='s irrelevance. But, one result of the argument should be a method the zero's out individual or group perceptions related to schedule/conference strength, history and tradition .

The problem is this; how does a computer (or a person, for that matter) parse through 10 teams that are all 11-1? If it is just based on "who beat who", you will end up without enough cross-conference information to make the determination. If you also include margin of victory, you are more likely to end up with the 11-1 team that happens to have the WORST schedule strength and not the BEST. All of these factors need to be "played out."

In the world of chess, a player will play hundreds if not thousands of games to see what their ranked score is. What I am proposing is to do away with the computers altogether, and let winning a conference send you into the playoffs. This way, you have played it out on the field. Let's say there are 9 conferences, leading to 9 conference winners. Invite another 7 "at-large" schools based on the rankings for a 16 team playoff. Seed them, so that 1 plays 16 and so on. At the end of the season, the champion is crowned, plain and simple. Nobody can argue that they didn't get a fair shake, because everybody has a chance to win their conference...

 

[Carl Spackler]

The problem is this; how does a computer (or a person, for that matter) parse through 10 teams that are all 11-1? If it is just based on "who beat who", you will end up without enough cross-conference information to make the determination. If you also include margin of victory, you are more likely to end up with the 11-1 team that happens to have the WORST schedule strength and not the BEST. All of these factors need to be "played out."

In the world of chess, a player will play hundreds if not thousands of games to see what their ranked score is. What I am proposing is to do away with the computers altogether, and let winning a conference send you into the playoffs. This way, you have played it out on the field. Let's say there are 9 conferences, leading to 9 conference winners. Invite another 7 "at-large" schools based on the rankings for a 16 team playoff. Seed them, so that 1 plays 16 and so on. At the end of the season, the champion is crowned, plain and simple. Nobody can argue that they didn't get a fair shake, because everybody has a chance to win their conference...

Dan. I'm in 100% accord w/ you. The thing is, there' only four teams to be considered right now. You have to take a step back and reaize that fact. For 130 years, the college football post season has been a popularity contest that revolves around making lots and lots of money.

The only people that can possibly ever make that change, just did, a few weeks ago.

I have no doubt, that a true playoff will sooner than later be in place, most likely very similar to what you describe, the biggest and most unbelievable hurdle has been cleared, the dust has to settle a bit for a while.

I'm quite confident that the new leadership of the largest BCS AQ conference in the United States, is going to serve our interests well.

 

[FAIRTIDES]

The problem is this; how does a computer (or a person, for that matter) parse through 10 teams that are all 11-1? If it is just based on "who beat who", you will end up without enough cross-conference information to make the determination. If you also include margin of victory, you are more likely to end up with the 11-1 team that happens to have the WORST schedule strength and not the BEST. All of these factors need to be "played out."

In the world of chess, a player will play hundreds if not thousands of games to see what their ranked score is. What I am proposing is to do away with the computers altogether, and let winning a conference send you into the playoffs. This way, you have played it out on the field. Let's say there are 9 conferences, leading to 9 conference winners. Invite another 7 "at-large" schools based on the rankings for a 16 team playoff. Seed them, so that 1 plays 16 and so on. At the end of the season, the champion is crowned, plain and simple. Nobody can argue that they didn't get a fair shake, because everybody has a chance to win their conference...

Dan, From your mouth to God's ears. Unfortunately, God ain't listening, so we are forced to talk about a chosen "4." Humans have foibles and biases and fears. Brainwashed humans think in terms of what they've been told should be. Computers don't think. They deal with what is.

Perfect? No, but it's better than having a cabal of beat writers and talking heads voting teams from the SEC one thru four.

 

[Carl Spackler]

Dan, From your mouth to God's ears. Unfortunately, God ain't listening, so we are forced to talk about a chosen "4." Humans have foibles and biases and fears. Brainwashed humans think in terms of what they've been told should be. Computers don't think. They deal with what is.

Perfect? No, but it's better than having a cabal of beat writers and talking heads voting teams from the SEC one thru four.

I agree. I'm no computer guy, and I understand completely, that computers only do what programmers tell them to do, so that's why it's so important that what the computers are told to do, is discussed openly and makes sense to everybody, and is reproducible by anyone, anywhere, entering data.

RIght now, rankings don't start in the computers until I think week 6 regular season. WTF? Run the rankings starting in week 1,and run them weekly till the end of the season.

A team that wins a conference championship game, clearly, should have significant weight in the final rankings.

Every season, everybody starts from scratch in week 1.

The human polls can continue to do their job of selling papers, but they can't be involved in this, until a system that includes conference champions either only (if less than 11 teams involved), or a system of all 11 conference champs - is put in place.

 

[UConnDan97]

Humans have foibles and biases and fears. Brainwashed humans think in terms of what they've been told should be. Computers don't think. They deal with what is.

Computers don't think. True.
Computers deal with "what is." False. For all the reasons I've already discussed, plus more, I'm sure. Computers deal with what they are told to deal with by the brainwashed humans, which have their foibles and biases and fears.

I should add that I am completely aware of the fact that we are not going to have a 16 team playoff anytime soon (or maybe at all), as well as the fact that the selection of the 4 team playoff will be just as biased as it has always been. I am just stating what would be the most fair and just system.

 

[FAIRTIDES]

Computers don't think. True.
Computers deal with "what is." False. For all the reasons I've already discussed, plus more, I'm sure. Computers deal with what they are told to deal with by the brainwashed humans, which have their foibles and biases and fears.

I should add that I am completely aware of the fact that we are not going to have a 16 team playoff anytime soon (or maybe at all), as well as the fact that the selection of the 4 team playoff will be just as biased as it has always been. I am just stating what would be the most fair and just system.

Those people (above) would not be allowed any where near the process. I would hire six, non-football fan, software engineers, impose "confidentiality," explain the issues and goals, define the input and output (essentially both "User" and "Engineering" specs) and send them off to develop the processing and return with a "ranking" from provided test data.. They would be told to deal with the raw data, only, and to not discuss the project with anyone. Finally, they would be told to develop documentation that explained the "logic" (all encompassing, including formulas/algorithms) that could be put in an understandable presentation.

The powers that be could then chose any, all or just one solution. See, once coded and accepted (the hard part), there would be no human intervention.

 

[whaler11]

If you ran computers week one you'd have 100 teams tied for number 1 as there would be no evidence past they are all 1-0.

 

[whaler11]

Those people (above) would not be allowed any where near the process. I would hire six, non-football fan, software engineers, impose "confidentiality," explain the issues and goals, define the input and output (essentially both "User" and "Engineering" specs) and send them off to develop the processing and return with a "ranking" from provided test data.. They would be told to deal with the raw data, only, and to not discuss the project with anyone. Finally, they would be told to develop documentation that explained the "logic" (all encompassing, including formulas/algorithms) that could be put in an understandable presentation.

The powers that be could then chose any, all or just one solution. See, once coded and accepted (the hard part), there would be no human intervention.

Until that formula put teams in the playoff that people don't think are in the top four. Then you start over.

 

[CTMike]

If you ran computers week one you'd have 100 teams tied for number 1 as there would be no evidence past they are all 1-0.

I was thinking about that... and you know, I'd personally be okay with it. Yep, you'd have a lot of ties week one... but the rankings really only matter at the end. So what's lost if you have a bunch of 1-0 teams ranked week 1?

 

[whaler11]

I was thinking about that... and you know, I'd personally be okay with it. Yep, you'd have a lot of ties week one... but the rankings really only matter at the end. So what's lost if you have a bunch of 1-0 teams ranked week 1?

Nothing it's just a bit pointless.

 

[Dann]

just don't release the comps until week 4 or w/e. have a coaches poll for talking/media/fans bc but just dont let it count for jackpoop

 

[CTMike]

Nothing it's just a bit pointless.

Yeah, it is... I just don't think it's any less pointless than it is now!

 

[UConnDan97]

If you ran computers week one you'd have 100 teams tied for number 1 as there would be no evidence past they are all 1-0.

Thank you! This conversation was getting tough. What people have to understand is that you can program to correct out biases, but you need large sample sizes. That's not computers, people; that's statistics! Computers are simply glorified calculators. The ONLY way to remove biases completely is a head-to-head match. In order to insure that everyone has a fair shake, we need to get to a point in time where every conference champion is invited to the tourney. Outside of that, there will always be a bias. And yes, yes, again I am aware of the fact that we aren't there yet. 4 team playoff will likely contain biases. I'm just saying that for those of you who think a computer is a "magic bullet", it isn't. Period.

 

[FAIRTIDES]

If you ran computers week one you'd have 100 teams tied for number 1 as there would be no evidence past they are all 1-0.

Not exactly. But, the ranking would not be relevant until week 9 or 10.

 

[FfldCntyFan]

The reality is that there really is not enough "connectivity" (using the common term among the computer rankings) in college football to ever run a model that will not have a heavy bias influencing the results throughout the entire season. There really is not enough said connectivity in college basketball (where non-conference games normally nearly equal conference games) to remove all bias. In a sport where conference games outnumber non-conference games by either 2-1 or 3-1, it is a joke believing that it would be possible to ever reach any point of true connectivity among all participants.

The fact is that it would be an impossible undertaking to formulate a model that can run with no opening values for the participating teams. One true test of these models would be to run three parallel models, one with the initial stated pre-game one values, one with these values reversed across the board and one with purely randomly generated values. In a model that does reach the connectivity that is claimed (Jeff Sagarin claims that by week six in football full connectivity is achieved and all predetermined values are dropped), the final results will be identical (as it would only utilize actual results in its computations). Unfortunately, nobody who publishes any computer rankings (whether used by the BCs or not) has every claimed to have run parallel models to determine the accuracy of their results.

Whether it is by human vote or computer model those outside of the power conferences will always be disadvantaged as the teams within these conferences will (due to conference scheduling) always have more games played against better teams (per human perception and/or computer models, whether this is or is not reality), giving them higher overall rankings.

 

[Chin Diesel]

Without going through the whole thread, this isn't complicated.

The reason why the system is good at figuring out 1 & 2 and not as good as figuring out 3&4 is because the system was supposed to identify 1&2.

The "correct" solution was an answer of #1 and #2 and everyone else is the remainder of the solution. The BCS system wasn't designed to differentiate #3 from #9.

If a ranking system is needed to identify #1-#4, it can be done. There's plenty of historical data to make the system work for those that is supposed to work.

 

[pj]

They should design the algorithm to find the teams with the 4 best chances of winning the national championship, rather than the 4 best teams. A team might be very good, but proven inferior to another team; while another team may be undefeated but unproven. The latter team may well be worse than the first, but they have a better chance of being the best.

 

[Palatine]

You don't think the Rose Bowl has any relevance or special meaning in the world of college football today? Wow.

Explain it to me.

Since the BCS, all the top bowls pay exactly the same to each team. Why is the Rose Bowl special? And why does the B10 so desperately want to hang onto it? It doesn't seem to make any sense.

 

[Palatine]

Without going through the whole thread, this isn't complicated.

The reason why the system is good at figuring out 1 & 2 and not as good as figuring out 3&4 is because the system was supposed to identify 1&2.

The "correct" solution was an answer of #1 and #2 and everyone else is the remainder of the solution. The BCS system wasn't designed to differentiate #3 from #9.

If a ranking system is needed to identify #1-#4, it can be done. There's plenty of historical data to make the system work for those that is supposed to work.

Then get the first two. Eliminate them, and do it a second time to get the next top two. Easy.

 

[Marty Jackson]

They should design the algorithm to find the teams with the 4 best chances of winning the national championship, rather than the 4 best teams. A team might be very good, but proven inferior to another team; while another team may be undefeated but unproven. The latter team may well be worse than the first, but they have a better chance of being the best.

Here's an idea... have teams play in order to decide who's better.
as long as there is some method used to determine who si the best team, there will be bias, prejudice, and unethical behavior.
The reason conferences are against the 'conference champion' rule is because they know it will level the playing field a bit. They'll say every game should matter, and we need to protect the season.
Well, what better way to make every game count than to only take the conference champion.

 

[UConnDan97]

Without going through the whole thread, this isn't complicated.

The reason why the system is good at figuring out 1 & 2 and not as good as figuring out 3&4 is because the system was supposed to identify 1&2.

The "correct" solution was an answer of #1 and #2 and everyone else is the remainder of the solution. The BCS system wasn't designed to differentiate #3 from #9.

If a ranking system is needed to identify #1-#4, it can be done. There's plenty of historical data to make the system work for those that is supposed to work.

Actually, that's not a true statement. For more on that, look at FfldCntyFan's comprehensive post on this page. I could argue (and I am sure that many in Stillwater and the Bay area did too) that Stanford and Oklahoma State had EVERY RIGHT to be considered #2 last year! And the fact that Alabama won the game does not mean that it deserved to be #2. There have been multiple battles for who #3 is over recent years. The reason Alabama was #2 was due to percentage points generated from either biased computer systems or biased voters. Did they pass the "eyeball test"? Sure. They are one heck of a team. But they are 1-1 against LSU. Unfortunately, LSU being 1-1 against Alabama doesn't make them champions. What would the game have been like if Okla St played? Who knows, but to say that the 1vs2 game is without controversy due to good computational design is not a true statement.

 

[nelsonmuntz]

Actually, that's not a true statement. For more on that, look at FfldCntyFan's comprehensive post on this page. I could argue (and I am sure that many in Stillwater and the Bay area did too) that Stanford and Oklahoma State had EVERY RIGHT to be considered #2 last year! And the fact that Alabama won the game does not mean that it deserved to be #2. There have been multiple battles for who #3 is over recent years. The reason Alabama was #2 was due to percentage points generated from either biased computer systems or biased voters. Did they pass the "eyeball test"? Sure. They are one heck of a team. But they are 1-1 against LSU. Unfortunately, LSU being 1-1 against Alabama doesn't make them champions. What would the game have been like if Okla St played? Who knows, but to say that the 1vs2 game is without controversy due to good computational design is not a true statement.

Agree completely. The current system is excellent at picking the two most popular programs from among the top 5-6 teams.

http://espn.go.com/blog/bigten/post/_/id/51014/b1g-open-to-compromise-favors-committee

The Big 10 seems open to a 3+1 model, which is fair. I prefer a Top 6 format personally. The Big 10 wants a Selection Committee. Inexplicably, the Big East does not. That position is idiotic, since any system based on polls is going to favor the major conferences. And unless the BCS computers are adjusted to accurately reflect home field advantage (which I believe only 1 incorporates home/away at all right now), the Big East will be further disadvantaged.