(12-05-2016 04:55 PM)MplsBison Wrote:  

(12-05-2016 04:43 PM)ken d Wrote:  Unless the programmer applies those weights differently depending on which team is playing, there is no bias.

That the algorithm is applied consistently does not mean it is void of bias.

Yes, in fact it does mean that. As I said, that word does not mean what you think it does.

(12-05-2016 08:00 PM)quo vadis Wrote:  

(12-05-2016 04:55 PM)MplsBison Wrote:  

(12-05-2016 04:43 PM)ken d Wrote:  Unless the programmer applies those weights differently depending on which team is playing, there is no bias.

That the algorithm is applied consistently does not mean it is void of bias.

You're misusing the term, because when we think of "bias" in terms of ranking teams, that refers to favoring or disfavoring teams or conferences merely because they are certain teams and certain conferences. E.g., I might vote for Notre Dame over ECU because I just assume they are better because they are "big name" Notre Dame, regardless of how they have actually done on the field so far. That's bias as it is commonly understood.

But let's say I'm writing a computer program to rank teams, and I believe that good offense is more indicative of how good a team is than good defense. i just fundamentally believe football is more about scoring points than not giving up points, so I think teams that score more are better than teams that give up very few points. So in my formula, scoring a lot of points is given more weight than giving up very few points.

In that case, my system will tend to produce a higher ranking for teams that have "high flying" offenses, that spread it out and throw it around all the time and win games 52 - 42, than teams that win by grinding it out 17-7 in low scoring games. My ranking will tend to reward Bob Stoop's Oklahoma teams moreso than Nick Saban's Alabama teams, at least until Kiffin souped up the offense a bit.

But, in this case, my bias isn't against Alabama or in favor of Oklahoma, it's a bias for a particular style of play or really, type of result. Should Oklahoma and Alabama switch offensive styles, then Alabama would be rewarded more. It would be wrong to say my system is biased against Alabama or the SEC and in favor of Oklahoma and the Big 12. It isn't. It's biased in favor/against a certain kind of result that, in this case, those teams just happen to produce.

So really, unless you believe that a programmer has a line of code that says "if team = Oklahoma, then add 10 points to ranking" or somesuch, there is no "bias" to speak of in the formula.

Well if you think low scoring defenses are more valuable and put that in your program, your program is generally biased towards the Big 10 and against the Big 12 and Pac 12.

(12-05-2016 08:18 PM)ken d Wrote:  

(12-05-2016 04:55 PM)MplsBison Wrote:  

(12-05-2016 04:43 PM)ken d Wrote:  Unless the programmer applies those weights differently depending on which team is playing, there is no bias.

That the algorithm is applied consistently does not mean it is void of bias.

Yes, in fact it does mean that. As I said, that word does not mean what you think it does.

You should look up bias in the dictionary. The word means exactly what he is saying. Bias is not defined as "bias against a particular team." In fact, every program is biased because it has a bias that particular things are important.

I think all of you agree. Its just that some of your are trying to argue against the dictionary.

this whole discussion reminds of how performance ratings and promotions are done at my work, which is a large corporation.

many employees demand an exact formula for what it takes to get an A rating.

but those of us that have been through other eras, know that as soon as you define what it means to get an A, people will start rigging and manipulating things to get that A rating.

sometimes its better to leave things ambiguous.

(12-06-2016 10:16 AM)bullet Wrote:  

(12-05-2016 08:00 PM)quo vadis Wrote:  

(12-05-2016 04:55 PM)MplsBison Wrote:  

(12-05-2016 04:43 PM)ken d Wrote:  Unless the programmer applies those weights differently depending on which team is playing, there is no bias.

That the algorithm is applied consistently does not mean it is void of bias.

You're misusing the term, because when we think of "bias" in terms of ranking teams, that refers to favoring or disfavoring teams or conferences merely because they are certain teams and certain conferences. E.g., I might vote for Notre Dame over ECU because I just assume they are better because they are "big name" Notre Dame, regardless of how they have actually done on the field so far. That's bias as it is commonly understood.

But let's say I'm writing a computer program to rank teams, and I believe that good offense is more indicative of how good a team is than good defense. i just fundamentally believe football is more about scoring points than not giving up points, so I think teams that score more are better than teams that give up very few points. So in my formula, scoring a lot of points is given more weight than giving up very few points.

In that case, my system will tend to produce a higher ranking for teams that have "high flying" offenses, that spread it out and throw it around all the time and win games 52 - 42, than teams that win by grinding it out 17-7 in low scoring games. My ranking will tend to reward Bob Stoop's Oklahoma teams moreso than Nick Saban's Alabama teams, at least until Kiffin souped up the offense a bit.

But, in this case, my bias isn't against Alabama or in favor of Oklahoma, it's a bias for a particular style of play or really, type of result. Should Oklahoma and Alabama switch offensive styles, then Alabama would be rewarded more. It would be wrong to say my system is biased against Alabama or the SEC and in favor of Oklahoma and the Big 12. It isn't. It's biased in favor/against a certain kind of result that, in this case, those teams just happen to produce.

So really, unless you believe that a programmer has a line of code that says "if team = Oklahoma, then add 10 points to ranking" or somesuch, there is no "bias" to speak of in the formula.

Well if you think low scoring defenses are more valuable and put that in your program, your program is generally biased towards the Big 10 and against the Big 12 and Pac 12.

No it's not. It's biased in favor of teams that have low scoring defenses, not against "the Big 12" or any other conference. If a team has a low scoring defense, the formula would favor them, whether they happen to be from the Big 12, PAC, SEC, MAC, or whatever.

Like Bison, you are misusing the term "bias" by applying it to conferences and teams rather than what it truly is, which is "bias" in favor of certain performance outcomes, like in this example, having a low scoring defense.

(12-05-2016 11:23 AM)EvilVodka Wrote:  The committee as it is right now is a joke....The Playoff is good, how it's decided is dumb....keep trying college football

The human element needs to be removed from determining the participants for the playoffs. It really is that simple.

(12-06-2016 10:40 AM)quo vadis Wrote:  

(12-06-2016 10:16 AM)bullet Wrote:  

(12-05-2016 08:00 PM)quo vadis Wrote:  

(12-05-2016 04:55 PM)MplsBison Wrote:  

(12-05-2016 04:43 PM)ken d Wrote:  Unless the programmer applies those weights differently depending on which team is playing, there is no bias.

That the algorithm is applied consistently does not mean it is void of bias.

You're misusing the term, because when we think of "bias" in terms of ranking teams, that refers to favoring or disfavoring teams or conferences merely because they are certain teams and certain conferences. E.g., I might vote for Notre Dame over ECU because I just assume they are better because they are "big name" Notre Dame, regardless of how they have actually done on the field so far. That's bias as it is commonly understood.

But let's say I'm writing a computer program to rank teams, and I believe that good offense is more indicative of how good a team is than good defense. i just fundamentally believe football is more about scoring points than not giving up points, so I think teams that score more are better than teams that give up very few points. So in my formula, scoring a lot of points is given more weight than giving up very few points.

In that case, my system will tend to produce a higher ranking for teams that have "high flying" offenses, that spread it out and throw it around all the time and win games 52 - 42, than teams that win by grinding it out 17-7 in low scoring games. My ranking will tend to reward Bob Stoop's Oklahoma teams moreso than Nick Saban's Alabama teams, at least until Kiffin souped up the offense a bit.

But, in this case, my bias isn't against Alabama or in favor of Oklahoma, it's a bias for a particular style of play or really, type of result. Should Oklahoma and Alabama switch offensive styles, then Alabama would be rewarded more. It would be wrong to say my system is biased against Alabama or the SEC and in favor of Oklahoma and the Big 12. It isn't. It's biased in favor/against a certain kind of result that, in this case, those teams just happen to produce.

So really, unless you believe that a programmer has a line of code that says "if team = Oklahoma, then add 10 points to ranking" or somesuch, there is no "bias" to speak of in the formula.

Well if you think low scoring defenses are more valuable and put that in your program, your program is generally biased towards the Big 10 and against the Big 12 and Pac 12.

No it's not. It's biased in favor of teams that have low scoring defenses, not against "the Big 12" or any other conference. If a team has a low scoring defense, the formula would favor them, whether they happen to be from the Big 12, PAC, SEC, MAC, or whatever.

Like Bison, you are misusing the term "bias" by applying it to conferences and teams rather than what it truly is, which is "bias" in favor of certain performance outcomes, like in this example, having a low scoring defense.

Or in favor of another performance outcome like winning a lot of games.

Before I responded to Bison, I did consult the dictionary. It gave two definitions.

1. Prejudice in favor of or against one thing, person or group compared with another, usually in a way considered to be unfair.

2. A systematic distortion of a statistical result due to a factor not allowed for in its derivation.

Bison's example fits neither of these definitions. If 100 different algorithms are applied to the results of football games, and they all produce similar results with respect to the ranking of the teams, it would be hard to support a claim of bias, and harder still to support a claim that bias produced a distorted result. If an algorithm did consistently show such a bias, it would be hard to find because that algorithm would have already been tossed on the metaphorical junk heap by serious analysts.

Ultimately, all the major computer algorithms are heavily weighted toward winning football games, and some give more weight to the margin of victory than others do. If teams, or conferences, consistently win more often on the field, they should be ranked high, and they are. If P5 conference teams beat G5 conference teams with regularity (and they do), an unbiased analysis would rate those P5 conferences as being stronger than the G5 conferences. And they do.

If you want to accuse somebody of bias, you need more evidence than anybody has ever presented to back that up.

(12-05-2016 08:00 PM)quo vadis Wrote:  unless you believe that a programmer has a line of code that says "if team = Oklahoma, then add 10 points to ranking" or somesuch, there is no "bias" to speak of in the formula.

False.

If a programmer selects a higher weight for a parameter that he knows will favor one conference over the others, that's a bias baked into the algorithm.

All selections of weightings are arbitrary. They can only come from the programmer's preference. And to pretend to bypass that by just weighting all parameters equal is simply just another arbitrary setting of the weights, no less biased than anything else.

Thus, all algorithms are biased.



I already know you'll dismiss that and hand-waive it away.

That's fine. In the end, everyone only believes what they want to believe. It's all just justifying what you think should be correct. There is no truth. Even the results on the field are apologized for, rationalized, hand-waived.

(12-05-2016 08:18 PM)ken d Wrote:  Yes, in fact it does mean that.

Nope. See above.

(12-06-2016 10:21 AM)bullet Wrote:  In fact, every program is biased because it has a bias that particular things are important.

Correct.

If at least one person gets it, then I'm happy.

(12-06-2016 10:28 AM)goofus Wrote:  this whole discussion reminds of how performance ratings and promotions are done at my work, which is a large corporation.

many employees demand an exact formula for what it takes to get an A rating.

but those of us that have been through other eras, know that as soon as you define what it means to get an A, people will start rigging and manipulating things to get that A rating.

sometimes its better to leave things ambiguous.

"That's strange, we don't seem to have that problem"
--signed NFL, NBA, MLB, College Basketball

(12-06-2016 10:52 AM)BamaScorpio69 Wrote:  

(12-05-2016 11:23 AM)EvilVodka Wrote:  The committee as it is right now is a joke....The Playoff is good, how it's decided is dumb....keep trying college football

The human element needs to be removed from determining the participants for the playoffs. It really is that simple.

I would RATHER have a formula where everyone knows what they have to do to get in.

The very fact that AFTER the conference championship games end, ALL the teams, coaches, fans wait for a group of people to huddle in a backroom and come up with the playoff is RETARDED

Having faith in this system moving forward is completely naive

(12-05-2016 05:34 PM)quo vadis Wrote:  Nobody should ever again make the dumb claim that 'losses don't matter'. The top 4 collectively have 3 losses.

"Say what???"
--signed Western Michigan

(12-06-2016 12:05 PM)MplsBison Wrote:  

(12-05-2016 08:00 PM)quo vadis Wrote:  unless you believe that a programmer has a line of code that says "if team = Oklahoma, then add 10 points to ranking" or somesuch, there is no "bias" to speak of in the formula.

False.

If a programmer selects a higher weight for a parameter that he knows will favor one conference over the others, that's a bias baked into the algorithm.

If he is motivated by a desire to favor a conference, yes, that's a bias. But nobody has claimed that possibility because there's no evidence for that.

Furthermore, even in that case, the bias may not work as he expects, because you can't predict how teams are going to perform. E.g., I might be a Big 12 partisan, so I write my formula to favor what i *think* will help the Big 12, like maybe high scoring offense. But then as the season turns out, the Big 12 offenses are low-scoring, so it backfires on me. So I would have to constantly re-jigger the formula as the season unfolds.

Which, of course, is what biased *humans* do in their own minds when they cast votes in polls, committees, and the like.

With computers, that kind of rejiggering would become pretty obvious to other computer stat geeks, they would quickly identify changing patterns in the numbers and call out whoever was doing that kind of thing.

I think it's pretty clear that these computer guys - almost all of whom have math degrees from schools that do NOT play big time college football - are far more dedicated to the integrity of the math than they are to teams and conferences. So in all liklihood, the computer formulas are as "objective" as we think they are.

Plus, that's why you compile a bunch of them, just in case of outliers and the like.

And again, if the programmer isn't trying to craft a formula that favors one conference over another, then his formula is unbiased among conferences, it is merely biased towards certain performance outcomes. BIG difference, because that makes the formula objective and neutral among conferences.

No amount of obfuscating and denial on your part can change that.

(12-05-2016 05:30 PM)bullet Wrote:  

(12-05-2016 12:34 PM)bronconick Wrote:  The reason they look like idiots is because they have to go out there midweek a half-dozen times on ESPN and contradict themselves from the previous time. They probably wish they were on the basketball committee who only gets interviewed once by CBS

And they make stupid decisions. I don't even think the Aggies thought their #4 ranking in the first poll made any sense.

The committee said they looked "krunk" that week

I thought they looked krunk that week...definitely deserved that #4 ranking

(12-06-2016 12:36 PM)EvilVodka Wrote:  

(12-05-2016 05:34 PM)quo vadis Wrote:  Nobody should ever again make the dumb claim that 'losses don't matter'. The top 4 collectively have 3 losses.

"Say what???"
--signed Western Michigan

Don't make the mistake of assuming that because losses aren't EVERYTHING that this means they "don't matter".

They matter, and more than anything else.

(12-05-2016 02:10 PM)MplsBison Wrote:  

(12-05-2016 01:58 PM)AubTiger16 Wrote:  Michigan State earned their spot in the CFP last season. Ohio State was exactly where they should have been. That is what they earned.

Ohio St no less earned its spot in the CFP this year.

...because they have shiny helmets?

They won THE GAME

(12-06-2016 12:39 PM)quo vadis Wrote:  

(12-06-2016 12:36 PM)EvilVodka Wrote:  

(12-05-2016 05:34 PM)quo vadis Wrote:  Nobody should ever again make the dumb claim that 'losses don't matter'. The top 4 collectively have 3 losses.

"Say what???"
--signed Western Michigan

Don't make the mistake of assuming that because losses aren't EVERYTHING that this means they "don't matter".

They matter, and more than anything else.

So they are something, just not everything

The committee is a lot like a Seinfeld episode

(12-06-2016 10:28 AM)goofus Wrote:  this whole discussion reminds of how performance ratings and promotions are done at my work, which is a large corporation.

many employees demand an exact formula for what it takes to get an A rating.

but those of us that have been through other eras, know that as soon as you define what it means to get an A, people will start rigging and manipulating things to get that A rating.

sometimes its better to leave things ambiguous.

Agreed. The most important thing should be playing your schedule and winning as many games as possible, so long as a team isn't intentionally manipulating things to try to improve its ranking. If you announce specific requirements, then teams and conferences *will* manipulate things intentionally.

If you announce that "game control" or margin of victory is an important factor, you'll have teams playing their starters the whole 4th quarter in routs just to try to win 49-3 instead of 21-3, trying to score meaningless points instead of just milking the clock when they have a comfortable lead.

If you announce that conference championships are not a deciding factor, then maybe conferences manipulate their tiebreakers to protect a highly ranked team (like Ohio State this year) from risking a loss in a CCG.

Or if you go the other way and conference championships are announced as the #1 factor, maybe conferences play as few division crossover games as possible (to avoid difficult games vs. the best teams in the other division) to try and get two 12-0 or 11-1 teams in their CCG every year.

If you announce that simply playing an FCS opponent is a disadvantage, then a would-be contender replaces its FCS opponent with a weak G5 team, but we get no further insight as to whether that team is more or less deserving.

Etc., etc.

(12-06-2016 10:28 AM)goofus Wrote:  sometimes its better to leave things ambiguous.

That's how large and prominent universities handle tenure. There is no schematic. It's people who meet and mind**** candidates for years as they work themselves stupid, all while saying one thing publicly, then privately telling them how it really is. There is no rule book. And people who sought it, achieved it, then became gatekeepers for the next class call it a game.

In one arena, it's getting grant funding. In some, it's publishing in journals nobody reads. To others, it's about teaching and involvement with campus life. But, the balance between each is different from candidate to candidate.

This system is very much like getting tenure. Fits college athletics like a glove. It's the only style they know.

Quote:They won THE GAME

You didn't get the memo after both UM and OSU beat Wiscy that it was the actual B1G CCG and not the formality that happened over the weekend?

(12-06-2016 12:37 PM)quo vadis Wrote:  but we get no further insight as to whether that team is more or less deserving.

But actually, "deserving" isn't correct. That's not what it is.

Easy counter example: a team that starts the season off 0-6, but then gets key star players back from injury, changes its schemes, and comes roaring back to go 6-0 in the final six games, destroying top 25 ranked opponents by 4 touchdowns on average. Clearly, that team deserves a spot in the playoff. But it won't get one.


That's because what the CFP actually is, is this: a resume competition.

It's impossible to "prove" among the top X teams, which team is really the "best" or which team really is the most "deserving". But they can all put a resume on paper, and have it judged.

(12-06-2016 01:35 PM)The Cutter of Bish Wrote:  This system is very much like getting tenure.

It's not that nebulous.

(12-06-2016 12:37 PM)quo vadis Wrote:  But nobody has claimed that possibility because there's no evidence for that.

It's self-evident. All algorithms must weight the input parameters. The weightings are arbitrary. There is no "correct" set of weightings.

You exactly exposed that this is correct in your post: when the programmer first starts building the classifier, it will spit out a nonsensical answer at the beginning. It might say the Sun Belt is the #1 conference and the SEC is the last conference.

So the weights have to be tuned, in order to get a sensible answer. Ah ... but when does the output stop being nonsensical and start being sensible?? Arbitrary.



Just as I'm certain you'll clip this part of my post out of any quoted response, I'm certain that you're fighting this because you yourself are the author of an algorithm to rank conferences. Your algorithm is biased, all the same.