CFB51 College Football Fan Community
The Power Five => Big Ten => Topic started by: Cincydawg on October 16, 2017, 07:50:00 AM
-
https://www.masseyratings.com/cf/compare.htm
Still around, barely noticed at all by anyone. Kinda funny.
-
https://www.masseyratings.com/cf/compare.htm
Still around, barely noticed at all by anyone. Kinda funny.
I post that each week in the power rankings thread.
I think each on its own has major flaws, but taken as a whole are pretty useful.
I also think they were a lot more impressive in the 80s and early 90s when there were just a couple and they were noteworthy. Now anyone with some time can do one pretty easily, and as that composite shows, there are a glut of them.
Not to mention we don't rely on them to determine anything anymore, but personally, I'd be cool with scrapping the committee and going with the top 4 from that under the current model.
-
I listen to Cade Massey on his Wharton 'moneyball' show on XM each week. He and his cohort Wharton Profs do a nice job, and Massey's ratings always come up.
-
I feel the computers have less bias than the humans
computers don't have eyeballs for the eyeball test
-
The bloggers and whatnot have largely moved to advanced stat metrics like S+P. These tend to use statistics from the games to derive how good the teams are and not simply the final score. S+P aims at beating Vegas about 54%. A lot of computer models just use the final scores of games, and when there are only 12-14 games in a season, that ain't too much in the way of inputs.
Advanced stats are also way more interesting than computer rankings, because they look at many facets of the game and give lots of fodder for long winded posts and podcasts.
-
I feel the computers have less bias than the humans
computers don't have eyeballs for the eyeball test
The have "bias" from their algorithms, or "Al Gore Rhythms".
-
I know they do, just believe it's less than humans and the bias doesn't change from week to week like the humans.
When Osborne was on the committee he used advanced stat metrics like S+P.
-
I feel the computers have less bias than the humans
computers don't have eyeballs for the eyeball test
I agree wholeheartedly.
-
It's important to distinguish results based rankings from predictive metrics, because they serve different purposes, and therefore can have discrepancies between them.
A team like Air Force is ranked higher in predictive metrics than results-based ones because it has played a fairly tough schedule so far. Conversely, Army has done well against a rather weak schedule to date, so it's results based ranking is higher than predictive metrics. Consequently, you shouldn't be surprised if Air Force ends up dominating them when they play each other this year.
In basketball, Kenpom is the best predictive metric, but it shouldn't be considered in the NCAA tournament selection or seeding process or teams like Clemson and Indiana would've been in the field last year instead of USC and VCU. There's still a strong correlation, to be sure, though. The RPI has its own flaws as far as results-based algorithms go (and it's a rather simple calculation), but so do other algorithms and formulas.
-
The best models incorporated multiple years to combat the small single-season sample size. Also the ones that used margin of victory were the most valid. The worst part about when the computers were included was that none of them were as-is.....they were all compromised, lesser versions of themselves.
I miss them.
-
A team like Air Force is ranked higher in predictive metrics than results-based ones because it has played a fairly tough schedule so far. Conversely, Army has done well against a rather weak schedule to date, so it's results based ranking is higher than predictive metrics. Consequently, you shouldn't be surprised if Air Force ends up dominating them when they play each other this year.
By S+P Army is 71 and Air Force is 86
-
It's important to distinguish results based rankings from predictive metrics, because they serve different purposes, and therefore can have discrepancies between them.
A team like Air Force is ranked higher in predictive metrics than results-based ones because it has played a fairly tough schedule so far. Conversely, Army has done well against a rather weak schedule to date, so it's results based ranking is higher than predictive metrics. Consequently, you shouldn't be surprised if Air Force ends up dominating them when they play each other this year.
In basketball, Kenpom is the best predictive metric, but it shouldn't be considered in the NCAA tournament selection or seeding process or teams like Clemson and Indiana would've been in the field last year instead of USC and VCU. There's still a strong correlation, to be sure, though. The RPI has its own flaws as far as results-based algorithms go (and it's a rather simple calculation), but so do other algorithms and formulas.
As this is the distinction too many people fail to recognize, and both have their pros and cons.
-
Honestly I don't see much point for "results based" computer rankings in college football. What's the point? Why use them over predictive based models? The things that are easy to see - overall record, head to head results, conference championships are pretty easy to see. For the things that are not easy to see, why not use the system that more reliably predicts a result?
-
Honestly I don't see much point for "results based" computer rankings in college football. What's the point? Why use them over predictive based models? The things that are easy to see - overall record, head to head results, conference championships are pretty easy to see. For the things that are not easy to see, why not use the system that more reliably predicts a result?
For message board debate and gambling, sure.
For determining entry into the CFP, or the NCAA tourney in hoops, I have no use for the predictive models.
-
For message board debate and gambling, sure.
For determining entry into the CFP, or the NCAA tourney in hoops, I have no use for the predictive models.
Yeah, but why?
-
Because I don't care that Team B should have won, but for some fluky bounces in favor of Team A, or Team B would win 9 out of 10 times.
I care that that the one time they played, Team B lost. That's what makes college football the best sport out there. That the season is so short, and that each loss is so detrimental. You remove fluky bounces and anamolies, you've castrated what makes the sport great.
As a fan of OSU, I'd get why you don't like that. That it generally takes something like that for OSU to lose, and that the schools that accumulate massive talent are going to be able to more consistently blow out the cupcakes on their schedule. But that's not college football to me. I don't give a rat's ass if the "best" team doesn't win the national title. If a team gets there through fluky bounces, narrow wins over mismatched opponents, and a key big upset, more power to them.
We all know who the "best" teams are. Progressing them in spite of the results, even if they are fluky, is boring as hell.
-
Do you personally glean any insight from computer polls that you find useful or helpful?
About all I see is spotting on occasion an "outlier" that is over or under ranked and I ponder why. If all they do is line up with human polls they are redundant.
We put together a computer poll here a few years back. The process was fascinating to me. In effect, we kept trying to get it to "make sense" by adjusting the algorithm. I realized we were devising an algorithm to predict HUMAN POLLS in effect.
Derp.
-
Because I don't care that Team B should have won, but for some fluky bounces in favor of Team A, or Team B would win 9 out of 10 times.
I care that that the one time they played, Team B lost. That's what makes college football the best sport out there. That the season is so short, and that each loss is so detrimental. You remove fluky bounces and anamolies, you've castrated what makes the sport great.
As a fan of OSU, I'd get why you don't like that. That it generally takes something like that for OSU to lose, and that the schools that accumulate massive talent are going to be able to more consistently blow out the cupcakes on their schedule. But that's not college football to me. I don't give a rat's ass if the "best" team doesn't win the national title. If a team gets there through fluky bounces, narrow wins over mismatched opponents, and a key big upset, more power to them.
We all know who the "best" teams are. Progressing them in spite of the results, even if they are fluky, is boring as hell.
I don't disagree, but I don't see what value computer polls give you in reaching that. I guess my point is - if I'm going to use a computer algorithm to compare teams, I'm going to use the best and most transparent one.
-
I don't disagree, but I don't see what value computer polls give you in reaching that. I guess my point is - if I'm going to use a computer algorithm to compare teams, I'm going to use the best and most transparent one.
I agree with you, if we are simply having a debate about the best teams. I think it's far more fun to dig deeper. Sometimes I find stuff that supports what I suspect, and it can be even more fun to realize how that data tells a far different story.
As for the value of results based computer rankings, I agree, it's limited. I don't have a problem with the polls, except as to 3 things, that the results based computer rankings do away with:
1.) Helmet bias
2.) Preseason expectations
3.) Gross recency bias
#1 is obvious. #2 is just the nature of things if you are going to have preseason rankings, and even if you don't have official ones, we can't simply clear our heads from our expectations going into the year. #3 is ok to an extent, but more as to trends than results. If a team is trending up or down, I don't mind rewarding/punishing that, but when you have two teams that are largely playing well, with no discernible trends, and each suffers one head scratching upset (of equivalent nature and margin) I don't see punishing the team more for having that game in November than September.
I kind of like the simplicity of SOR for selection purposes.
-
1.) Helmet bias
2.) Preseason expectations
3.) Gross recency bias
& 4.) Region bias
-
The best thing a computer poll COULD do is highlight that Team X is considerably over/under ranked by humans.
By considerably, I mean a team ranked 22 by humans and 8 by computers.
It doesn't much matter if a team is ranked 8 and should be 6.
-
but, it's a heck of a matter if at the end of the season a team is ranked 5th instead of 4th!
-
Yeah, I was musing about the critical 5-4 differential. However, that sort of very slim judgment that makes a huge difference is not something I'd trust to computers over humans.
And then of course you ponder if that #4 team is going to get pounded by #1 when the #5 team plays in a NY6 bowl game and perhaps gets a nice looking win over #9.
That's why I said a useful computer poll could tell us a team is better than #18 and should be in the 6-8 range.
-
hindsight cannot determine that if #5 beats the heck outa #6, that #5 would have given #1 a better game than #4
but, many folks do not share my opinion on that matter
-
I wasn't going to go here, but since this debate has continued, in football, due to the limited number of games, I evaluate teams largely based on proxy wins and losses, as much as anything.
For instance, Michigan's only proxy losses are Notre Dame (2nd degree) and Georgia (3rd degree), as a result of losing to Michigan State.
Conversely, Michigan has some okay but not great 2nd degree proxy wins over the likes of Tennessee, Vanderbilt, Kentucky, Minnesota, and Virginia.
In either direction, the nth degree W/L is less meaningful than the earlier ones (Michigan even has proxy wins over Clemson and USC, but that's relatively meaningless since they're so distant). If a team has a proxy win and loss to the same team (as often happens), I look at which degree was smaller. By no means is it a perfect system, but it usually works well enough.
-
Yeah, we all look at proxy wins, but often we see A beats B and B beats C and then C beats A. Syllogism doesn't work then.
I also look at close win/s losses. Imagine ND had not fumbled late and beat UGA 22-20. It could have happened obviously. It would not mean either team was better or worse. The UGA LB made a good play obviously, but it was timing that led to that late fumble, a coin toss kind of thing.
Similarly, ND beat MSU by a substantive margin in the score, but the internal stats and turnovers show a fairly even game. Does that mean UGA is substantially better than MSU? I don't think so.
The difference between 7-0 and 5-2 can be a few "lucky breaks".
-
For message board debate and gambling, sure.
For determining entry into the CFP, or the NCAA tourney in hoops, I have no use for the predictive models.
No use? None? ELA, I'm feeling a lack of creativity from you and MichiFan (and Michifan not having quite as strong a grip on the NCAA selection process as I'd hope)
Using predictive models to say who's in and who's out? Bad. It incentivizes bad things, makes models less useful. But using them as a primary tool to determine what's been accomplished, that's good if not vital. Don't say, "This teams rating is good, so X," say, "This is a lens through which to view a resume"
The alternative this silly second-order win/transitive property, where you say "team X beat team A and B, who in turn beat teams E,F,G." And it's a mess because is CFB teams just don't play all that many "good teams." So a hypothetical OSU might be a good example. Lets say the lose close to PSU, grind one out against MSU and take a rivalry upset to UM. They're 10-3 with wins against Iowa and MSU, which is good, but probably understates how good that win should be on resumes. (You probably already know this, and I'm sorry for hammering it home).
I'm torn on the question of team strength in seeding. Like if you treat Wichita State like it's as strong as it is, you give it a easier path than it earned. But if you don't you deny a team that earned an easier path what it earned. That's always tricky.
-
The best models incorporated multiple years to combat the small single-season sample size. Also the ones that used margin of victory were the most valid. The worst part about when the computers were included was that none of them were as-is.....they were all compromised, lesser versions of themselves.
I miss them.
Disagree.
You're just looking for a good indicator with a lot of repetition. Points are pretty useful for that, but in CFB, they can be a bit, lumpy for lack of a better word (there are few games and points can go big and small in various ways, points aren't scored in small, close increments). In baseball, runs are pretty good because there's a million games. In basketball, you essentially use MOV, but adjust for pace. In Hockey, shots are the quick and easy indicator. Football can use drives or even plays to get good stuff.
That said, scoring is old and reliable and works everywhere, like a Volvo that runs forever. It's why Sagarin, which can get a little weird for CFB can apply to sports with larger sample sizes and less data (HS with the right database or getting a reasonable sense of FCS).
-
It's tough to really trust any of them without knowing the formulas they use.