NYT: BCS Formulas Secret, Systems Utilize Fuzzy Math
http://www.nytimes.com/2010/10/17/sports/ncaafootball/17score.html?_r=2&src=mv&ref=sports
They're pretty much just making this crap up as they go along.
October 18th, 2010 at 3:34 PM ^
I'll take crappy computer rankings over horrible human polls any day of the week. The BCS seems intent to just make the computer polls look like a joke so the typical idiot public can agree with eventually just getting rid of them, which is really sad.
I'd much prefer them switch it to be 50/50 human/computer and let folks like Sagarin/Massey use their actual rankings and boot out the computer polls which aren't done to a certain standard.
Of course, it'd also be nice if they could actually regulate the human polls. Get people that actually watch most/all of the games and have a fundamental understanding of football to do the rankings. The fact that the coaches' poll is still used blows my mind. How many coaches actually have the time to sit down and review all the teams games for the week and make a real decision as to who should be ranked where? Maybe Tim Brewster now, but that'd be it.
October 18th, 2010 at 3:46 PM ^
It would be kind of cool to allow the human polls to take place and let the computers use those numbers in their measurements. Ranking Miami high or low in the human polls might allow the computers to rank OSU higher or lower in a corresponding fashion.
October 18th, 2010 at 4:05 PM ^
The crappy human polls are better than Billingsley. But don't take my word for it, Billingsley's ratings have always been a joke:
http://mgoblog.blogspot.com/2006/11/putting-b-and-s-in-bcs.html
October 18th, 2010 at 4:33 PM ^
Reading that, it definitely sound stupid and ridiculous. However it doesn't sound that much worse than having some random coaches' aide fill out their rankings for them or whatever idiots are involved in the Harris poll.
October 18th, 2010 at 3:43 PM ^
They criticize the computer polls for ranking OSU low, but what does OSU do that same week? They lose and they end up where they were predicted to be by the computer polls.
If I'm not mistaken, margin of victory was part of the formula in the beginning and was taken out after people got upset that a Miami team beating up a bunch of nobodies got in. I think the main reason to take it out is that it encourages running up the score, kind of how Oklahoma ran up the score on TTU in 2008 to pass Texas in the voting polls and, in turn, the BCS.
This whole thing is just one giant case of the observer effect. If you use score as a measurement, the scores are going to change.
October 18th, 2010 at 3:50 PM ^
The problem with leaving data out is that the output is only as good as the data that goes in. Margin of victory could be a major predictor of outcome, but they have to leave it out to avoid poor UAB getting thrased too severely?
In that case, cut off the MOV at 21 points or something like that. At least then, you can see that OSU beat Illinois by 7 and MSU by 20. To ignore every input seems to be willfully ignorant at the true power of computer predictions.
October 18th, 2010 at 3:54 PM ^
I was thinking either this (the limit on MOV) or some sort of multiplier. When Wisconson beats OSU they get the MOV multiplied times x, when they beat up west north dakota community college at bismark it gets multiplied by .1x. That way the MOV is still used but it doesn't help as much to beat 1AA schools by 50 but only beat better teams by 1 or 2. Balance it out as well in case you beat a better team by a bunch (means more than just a game winning field goal).
October 18th, 2010 at 4:40 PM ^
use strength of schedule?
October 18th, 2010 at 5:01 PM ^
Strength of schedule but take MOV into account. If you play the toughest schedule in the nation and win every game by one should you really be ranked ahead of the team playing the second hardest schedule that won every game by 30?
October 18th, 2010 at 4:01 PM ^
You can't expect great computer rankings when you limit/selectively choose which data gets entered.
Honestly, the way it's set up right now, I'd rather just have the human polls. Sometimes you learn more from watching the games than we choose to admit -- is there a single person here who thinks LSU is better than Oregon? That's what the computers say, mostly because the computers don't know that Oregon is winning by 60 while LSU is requiring crazy luck to beat mediocre teams.
I'm not saying this to criticize LSU -- I'm just pointing out that it's stupid to only give the computers the wins, losses and records and not allow them to take any other information into account.
October 18th, 2010 at 4:21 PM ^
I don't disagree with you, I'm actually playing more of a devil's advocate role (which is something I do a lot more than I actually mean to on here).
I wrote a ranking program a few years ago and couldn't get it right because of mediocre teams demolishing opponents when they could. I figured out a couple months ago that the best way to mitigate this is to use something like this
http://en.wikipedia.org/wiki/Logistic_function
with yards, points, or margin of victory. Essentially, winning by 1 point might get you 2 points in the polls whereas winning by 7 might only get you 5 and winning by 21 or more might get you 10. So, if you are up by 14 with 6 minutes to go, is it really worth the risk to run up the score to get 1-2 more points in the polls when a turnover and TD could cost you 3?
October 18th, 2010 at 4:28 PM ^
At least then, you can see that OSU beat Illinois by 7 and MSU by 20.
OSU actually beat Illinois by 11, in Champaign, when Pryor missed part of the game. MSU beat Illinois by 20 in East Lansing, with Cousins going the distance. I'd say those are pretty comparable results.
October 18th, 2010 at 5:35 PM ^
Certainly taking margin out did skew things in the computer polls. The problem was that the polls were putting too much into margin of victory and instead of tuning it they just got rid of it.
Reasonably, what you want to do is look at margin on a per quarter basis and saturate the benefit at about 1 TD per quarter advantage. Back in the day some people actually came up with some incredibly complex formulas for margin but none of them ever made it in since they eliminated margin. Most of these formulas were drive/quarter based and also accounted for a lot of other factors. One of the issues with most computer polls is that they only look at end game results and not the flow through the game which results in slightly skewed results just like human voters who only look at final scores can get a skewed view of the game.
October 18th, 2010 at 4:10 PM ^
“I’m not a mathematician,” Billingsley said. “I’m not even a highly educated man, to tell you the truth. I don’t even have a degree. I have a high school education. I never had calculus. I don’t even remember much about algebra.
October 18th, 2010 at 4:13 PM ^
If you like the article, check out their newly-released book, Death to the BCS. That article is essentially one chapter pulled from their book. I picked it up on Saturday and finished it pretty quickly. It's about 190 pages and reads like a very long investigative journalism piece (because that's basically what it is).
October 18th, 2010 at 4:30 PM ^
I think one simple change would be beneficial: don't make the human voters cast their ballots on Saturday night. Give them a few days to digest the week's results and release the polls on say, Tuesday.
October 18th, 2010 at 4:54 PM ^
This is one of many simple-to-institute good ideas that will probably never happen.
Also, your MGoPoint total is mind-boggling.
October 18th, 2010 at 5:08 PM ^
It's mostly a sign of having too much down time at my computer at work. Several people have more points than me.
October 18th, 2010 at 5:39 PM ^
Of course it won't happen. If they do that, Brian can go grab a lawyer and sue them for copying his methodology with the blog poll of actually waiting until people can review the actual results.
October 18th, 2010 at 5:41 PM ^
They need to put margin of victory back in, but with a ceiling of four touchdowns (28 points). That way, 48-20 would count the same as 70-20. Blowouts are part of college football, but there is a point where the points become meaningless. Besides, keeping starters around for the fourth quarter of a blowout and trying to score as many points as possible is a way to get players on both sides of the ball injured.
October 18th, 2010 at 5:54 PM ^
The purpose of the computer ratings in the BCS is not to wield power, but to draw attention away from it.
I guarantee you that the #1 and #2 teams in the polls will meet in the BCS championship game. The computer ratings might disagree, but they are outvoted 2-to-1. Ever since the formula was changed after the 2004 season to give the pollsters 2/3 of the vote, the computer ratings have been meaningless.
In 2005, USC and Texas were 1-2 in both polls, and 1-2 in the BCS ratings.
In 2006, Ohio State and Florida were 1-2 in both polls, and 1-2 in the BCS ratings.
In 2007, Ohio State and LSU were 1-2 in both polls, and 1-2 in the BCS ratings.
In 2008, Oklahoma and Florida were 1-2 in both polls, and 1-2 in the BCS ratings.
In 2009, Alabama and Texas were 1-2 in both polls, and 1-2 in the BCS ratings.
The pollsters have all of the power in determining the championship game participants. The computer ratings are included in the ratings for the sole purpose of allowing sportswriters to write angry columns ("blah blah COMPUTERS pick the NATIONAL CHAMPION blah blah UNACCEPTABLE!!!").
October 18th, 2010 at 6:37 PM ^
Don't be so sure. This season could be the perfect opportunity for the pollsters to vote Boise State at No. 2 and then have the computers put them at No. 6 or whatever (due to the ever-dropping strength of schedule). Then -- when there's a one-loss team at No. 3 in the polls that finishes No. 2 in the BCS -- they can say, "Well, it's not our fault Boise didn't play in the national championship game -- we voted them No. 2."
With the so-called "mid-majors," there could be enough difference in the computers to actually cause some controversy this year.
October 18th, 2010 at 7:29 PM ^
I am pretty sure that most of the "computer" polls are ad hocked enough to get any result the powers that be want to achieve.