(01-28-2023 04:05 AM)Milwaukee Wrote:  

(01-25-2023 05:32 PM)quo vadis Wrote:  I put "Final" in quotes because who knows if some more computers will check in, but it has been over two weeks since the CFP title game and there are 73 that have, so I'm calling it:

1) Georgia
2) Michigan
3) Ohio State
4) Alabama
5) Tennessee
6) TCU
7) Penn State
8) Kansas State
9) LSU
10) Clemson
11) Utah
12) Oregon
13) USC
14) Washington
15) FSU

Despite beating (#8) Kansas State and (#13) USC, Tulane is not listed in the top 15 of the Massey Composite rankings, which have consistently under-ranked the non-P5 teams, suggesting a possible pro-P5 bias, when compared with the AP and Coaches rankings.

In contrast, both the final AP (sportswriters) and Coaches polls ranked Tulane #9th, and that may be considered more authoritative than the Massey composite, in some respects.

Tulane was ranked #16 in the final CFP poll - - but that was before their NY6 bowl victory over (#10) USC, so it seems like the Green Wave would have been ranked at least #15th if there had been a final, post-season CFP poll in January.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Further evidence of potential pro-P5 bias in the Massey Composite:

Number of non-P5 teams in the final top 30 of the Massey Composite: two.

Number of non-P5 teams in the final AP top 30: six.

Number of non-P5 teams in the final Coaches Poll top 30: six.

Number of non-P5 teams in the final CFP top 25: three.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Interestingly, the apparent pro-P5 bias in the Massey Football Composite is not apparent in the Massey Basketball Composite, suggesting that the trends aren't systematic in Massey's system, but in the ranking systems that comprise the two composites.

It seems to me less likely that the MC is biased in football than that the human polls (including the CFP committee) are. The human polls take into account the fact that there is a guaranteed spot in the CFP for the highest ranked G5 champion, so there is public interest in how the top G5 teams stack up relative to each other. Since there is no reward for P5 teams who are ranked outside the top 12, voters can use the bottom half of the Top 25 to recognize the better G5 teams. With a 68 team field and 32 autobids, the MC component rankings don't need to take the equivalent of an NY6 into account.

(01-28-2023 06:05 AM)ken d Wrote:  

(01-28-2023 04:05 AM)Milwaukee Wrote:  

(01-25-2023 05:32 PM)quo vadis Wrote:  I put "Final" in quotes because who knows if some more computers will check in, but it has been over two weeks since the CFP title game and there are 73 that have, so I'm calling it:

1) Georgia
2) Michigan
3) Ohio State
4) Alabama
5) Tennessee
6) TCU
7) Penn State
8) Kansas State
9) LSU
10) Clemson
11) Utah
12) Oregon
13) USC
14) Washington
15) FSU

Despite beating (#8) Kansas State and (#13) USC, Tulane is not listed in the top 15 of the Massey Composite rankings, which have consistently under-ranked the non-P5 teams, suggesting a possible pro-P5 bias, when compared with the AP and Coaches rankings.

In contrast, both the final AP (sportswriters) and Coaches polls ranked Tulane #9th, and that may be considered more authoritative than the Massey composite, in some respects.

Tulane was ranked #16 in the final CFP poll - - but that was before their NY6 bowl victory over (#10) USC, so it seems like the Green Wave would have been ranked at least #15th if there had been a final, post-season CFP poll in January.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Further evidence of potential pro-P5 bias in the Massey Composite:

Number of non-P5 teams in the final top 30 of the Massey Composite: two.

Number of non-P5 teams in the final AP top 30: six.

Number of non-P5 teams in the final Coaches Poll top 30: six.

Number of non-P5 teams in the final CFP top 25: three.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Interestingly, the apparent pro-P5 bias in the Massey Football Composite is not apparent in the Massey Basketball Composite, suggesting that the trends aren't systematic in Massey's system, but in the ranking systems that comprise the two composites.

It seems to me less likely that the MC is biased in football than that the human polls (including the CFP committee) are. The human polls take into account the fact that there is a guaranteed spot in the CFP for the highest ranked G5 champion, so there is public interest in how the top G5 teams stack up relative to each other. Since there is no reward for P5 teams who are ranked outside the top 12, voters can use the bottom half of the Top 25 to recognize the better G5 teams. With a 68 team field and 32 autobids, the MC component rankings don't need to take the equivalent of an NY6 into account.

Yes.

And also, humans get caught up in story lines and trends and chatter and all of the emotions of college football. To me, computers presumably do not. They have the bias of the person who codes them, but not the ongoing bias of social chatter.

And on that point, IMO, most computers are likely run by geeks - mathematicians and statisticians who care more about the numbers than the names on the jerseys. So that IMO makes it not likely that they have coded their computers to be biased in favor of A5 over G5, SEC over AAC, etc.

I also think that humans are just naturally more impressed with W-L record, whereas computers dig deeper in to SOS and other factors. This IMO tends to lead to the over-rating of G5 schools by humans that have posted gaudy records vs bad schedules.

All that leads me to believe that when there is a discrepancy between the human voters and computers, it is the human voters that are likely the biased party, the party whose results are further from the God-Only-Knows "real truth", not the computers.

That's why IIRC there is a tendency for the AP poll to have more G5 in the final top 25 than we see in the MC. IIRC, over the past five seasons, 30 G5 schools (counting BYU as G) have finished in the AP top 25, whereas 21 have in the MC. To me, that says the humans are overrating G5.

(01-25-2023 05:32 PM)quo vadis Wrote:  On the other hand, I feel vindicated by LSU at #9 ... They beat Purdue by literally a record-setting margin in their bowl

More like "Purdue" with that skeleton roster and coaching staff the Boilers took to Orlando (after winning a lousy B1G West by default).

LSU basically routed a MAC also-ran that plays in a mid-December bowl in front of 600 people.

The computers aren’t biased. But most heavily weight sos. That will favor the P5. Computers have trouble with intangibles. Humans have trouble filtering out buzz.

(01-28-2023 12:22 PM)bullet Wrote:  The computers aren’t biased. But most heavily weight sos. That will favor the P5. Computers have trouble with intangibles. Humans have trouble filtering out buzz.

Some computer algorithms do and some don't. I really don't know much about the vast majority of the ones that submit their rankings to the MC.

Recognizing that human voters may be biased, either consciously or otherwise, the AP and Coaches' polls rely on having a large sample of voters. It's likely that voters for one school or conference or group of conferences will be offset by voters who have an opposite bias. At least that's the theory. The selection committee has a much smaller sample size of voters, but rely on having most, if not all, conferences represented and having individual members sit out discussions and voting for schools in which they have a perceived affinity.

Computers, on the other hand, have the bias of a single programmer. That bias isn't based on teams or conferences, but rather on the relative weight of various criteria used in their algorithms. The makeup of the MC varies from year to year (sometimes from week to week) as the number of rankings that make up the total change.

Highly subjective scoring sytems, like figure skating, throw out each contestant's best and worst score. What I would like the MC to do is follow suit and throw out the rankings that are the outliers, either for or against individual schools. IMO, if a ranking places a school among the Top 10 when no other ranking has them in the Top 30 they shouldn't be counted.

At the end of the day, whatever ranking system is used, you can be certain that some fans will think it is flawed. It's the nature of the beast.

(01-28-2023 02:02 PM)ken d Wrote:  

(01-28-2023 12:22 PM)bullet Wrote:  The computers aren’t biased. But most heavily weight sos. That will favor the P5. Computers have trouble with intangibles. Humans have trouble filtering out buzz.

Some computer algorithms do and some don't. I really don't know much about the vast majority of the ones that submit their rankings to the MC.

Recognizing that human voters may be biased, either consciously or otherwise, the AP and Coaches' polls rely on having a large sample of voters. It's likely that voters for one school or conference or group of conferences will be offset by voters who have an opposite bias. At least that's the theory. The selection committee has a much smaller sample size of voters, but rely on having most, if not all, conferences represented and having individual members sit out discussions and voting for schools in which they have a perceived affinity.

Computers, on the other hand, have the bias of a single programmer. That bias isn't based on teams or conferences, but rather on the relative weight of various criteria used in their algorithms. The makeup of the MC varies from year to year (sometimes from week to week) as the number of rankings that make up the total change.

Highly subjective scoring sytems, like figure skating, throw out each contestant's best and worst score. What I would like the MC to do is follow suit and throw out the rankings that are the outliers, either for or against individual schools. IMO, if a ranking places a school among the Top 10 when no other ranking has them in the Top 30 they shouldn't be counted.

At the end of the day, whatever ranking system is used, you can be certain that some fans will think it is flawed. It's the nature of the beast.

With 73 up to 99 rankings, the outliers in the MC are immaterial and irrelevant.
You are right that not all computer rankings use SOS, either directly or indirectly, but most do.
As for humans, they watch ESPN and get the ESPN propaganda for or against certain schools and conferences. So there is group bias.

(01-28-2023 11:08 AM)PeteTheChop Wrote:  

(01-25-2023 05:32 PM)quo vadis Wrote:  On the other hand, I feel vindicated by LSU at #9 ... They beat Purdue by literally a record-setting margin in their bowl

More like "Purdue" with that skeleton roster and coaching staff the Boilers took to Orlando (after winning a lousy B1G West by default).

LSU basically routed a MAC also-ran that plays in a mid-December bowl in front of 600 people.

Coaching, didn't Purdue have one Brohm in the place of the departed Brohm? I think Brian was Jeff's OC for about 5 years, so probably knew the overall system pretty well for the game. And I think Drew Brees came on board to coach the QBs for the game. That impressed me. I'm being somewhat facetious, but I'm not sure how much the coaching changes mattered. The kids are pretty much coached-up, or not, by December, IMO.

As for players. IIRC, LSU had about 23 players miss that game due to opt outs and transfer portals and the like, which I believe was more than any other bowl team. I imagine they were missing a lot more overall talent than was Purdue.

(01-28-2023 10:08 PM)quo vadis Wrote:  IIRC, LSU had about 23 players miss that game due to opt outs and transfer portals and the like, which I believe was more than any other bowl team. I imagine they were missing a lot more overall talent than was Purdue.

I think we are rapidly reaching the point — if we haven’t reached it already — that non-NY6 bowls (or non-Playoff games, starting in ’24) simply won’t be viewed as indicative of a team’s ability because so many of the team’s players and coaches won’t have participated.

As for the Massey ratings, they are useful but I always go by the median ranking and not the mean, which gives too much input to the outliers, IMHO.

I don’t feel the need to argue over polls and rankings. People are going to see what they want to see and not-see what they don’t want to see. With that said, I’m confident that Tulane’s #9 ranking in both major polls is a more accurate reflection of their accomplishment, and also will be more clearly remembered by college football fans long after the Massey composite ratings are forgotten. I would point out that traditionally, Massey’s final post-bowls rating had more than 73 polls. In 2019, for example, there were 103 ratings in Massey’s final composite. Curious to compare that.

(01-29-2023 06:36 AM)Native Georgian Wrote:  

(01-28-2023 10:08 PM)quo vadis Wrote:  IIRC, LSU had about 23 players miss that game due to opt outs and transfer portals and the like, which I believe was more than any other bowl team. I imagine they were missing a lot more overall talent than was Purdue.

I think we are rapidly reaching the point — if we haven’t reached it already — that non-NY6 bowls (or non-Playoff games, starting in ’24) simply won’t be viewed as indicative of a team’s ability because so many of the team’s players and coaches won’t have participated.

As for the Massey ratings, they are useful but I always go by the median ranking and not the mean, which gives too much input to the outliers, IMHO.

I don’t feel the need to argue over polls and rankings. People are going to see what they want to see and not-see what they don’t want to see. With that said, I’m confident that Tulane’s #9 ranking in both major polls is a more accurate reflection of their accomplishment, and also will be more clearly remembered by college football fans long after the Massey composite ratings are forgotten. I would point out that traditionally, Massey’s final post-bowls rating had more than 73 polls. In 2019, for example, there were 103 ratings in Massey’s final composite. Curious to compare that.

Interesting idea about medians. FWIW, I just checked on that and LSU's median is 11, two below their overall ranking, while Tulane's is 17, one position better than their overall ranking of 18.

As for the number of computers in the MC, yes, that seems to fluctuate, and IIRC one trend is that the final post-bowl number of rankings tend to be lower than the number for the regular season. I guess some computers just do not do a final post-bowl ranking.

Still, 73 rankings is a lot of rankings IMO, more than a critical mass for me to rely on it.

As for Tulane, I agree that their #9 human poll ranking will be remembered, not the MC ranking. Heck, I don't think most CFB fans are even aware of the existence of the MC, whereas everyone knows the human polls.

I just like the MC because, for reasons given earlier, it is IMO a more accurate indicator of how good a team was compared to human polls. Not perfect by any means, and sometimes the results for some teams can be really baffling, but overall better, to me.

(01-29-2023 11:37 AM)quo vadis Wrote:  

(01-29-2023 06:36 AM)Native Georgian Wrote:  

(01-28-2023 10:08 PM)quo vadis Wrote:  IIRC, LSU had about 23 players miss that game due to opt outs and transfer portals and the like, which I believe was more than any other bowl team. I imagine they were missing a lot more overall talent than was Purdue.

I think we are rapidly reaching the point — if we haven’t reached it already — that non-NY6 bowls (or non-Playoff games, starting in ’24) simply won’t be viewed as indicative of a team’s ability because so many of the team’s players and coaches won’t have participated.

As for the Massey ratings, they are useful but I always go by the median ranking and not the mean, which gives too much input to the outliers, IMHO.

I don’t feel the need to argue over polls and rankings. People are going to see what they want to see and not-see what they don’t want to see. With that said, I’m confident that Tulane’s #9 ranking in both major polls is a more accurate reflection of their accomplishment, and also will be more clearly remembered by college football fans long after the Massey composite ratings are forgotten. I would point out that traditionally, Massey’s final post-bowls rating had more than 73 polls. In 2019, for example, there were 103 ratings in Massey’s final composite. Curious to compare that.

Interesting idea about medians. FWIW, I just checked on that and LSU's median is 11, two below their overall ranking, while Tulane's is 17, one position better than their overall ranking of 18.

As for the number of computers in the MC, yes, that seems to fluctuate, and IIRC one trend is that the final post-bowl number of rankings tend to be lower than the number for the regular season. I guess some computers just do not do a final post-bowl ranking.

Still, 73 rankings is a lot of rankings IMO, more than a critical mass for me to rely on it.

As for Tulane, I agree that their #9 human poll ranking will be remembered, not the MC ranking. Heck, I don't think most CFB fans are even aware of the existence of the MC, whereas everyone knows the human polls.

I just like the MC because, for reasons given earlier, it is IMO a more accurate indicator of how good a team was compared to human polls. Not perfect by any means, and sometimes the results for some teams can be really baffling, but overall better, to me.

The human polls reflect the subjective opinions of 62 voters. By saying it is your opinion that the MC is more accurate you are saying it is more accurate compared to the subjective opinion of one voter - you - and of course that your opinion is "more accurate" than the human polls.

(01-29-2023 01:45 PM)ken d Wrote:  The human polls reflect the subjective opinions of 62 voters.

In all the years I’ve followed college football, I’ve never quite understood how the AP decides which sportswriters — or what # of sportswriters — get a vote in their football/basketball polls. And the same goes for coaches in the UPI poll or whatever it’s called now. I freely admit (because a lot of coaches have already admitted it) that they fill out the Top 5 or Top 10 or Top whatever, and then let an assistant coach or SID fill out the rest.

Still, the combined weight of all those votes — this year it’s 63 coaches and 63 writers (coincidence?) — means that if you combine the two polls, your seeing the aggregate ranking of 126 different people who follow college football very closely.

(01-29-2023 11:37 AM)quo vadis Wrote:  

(01-29-2023 06:36 AM)Native Georgian Wrote:  

(01-28-2023 10:08 PM)quo vadis Wrote:  IIRC, LSU had about 23 players miss that game due to opt outs and transfer portals and the like, which I believe was more than any other bowl team. I imagine they were missing a lot more overall talent than was Purdue.

I think we are rapidly reaching the point — if we haven’t reached it already — that non-NY6 bowls (or non-Playoff games, starting in ’24) simply won’t be viewed as indicative of a team’s ability because so many of the team’s players and coaches won’t have participated.

As for the Massey ratings, they are useful but I always go by the median ranking and not the mean, which gives too much input to the outliers, IMHO.

I don’t feel the need to argue over polls and rankings. People are going to see what they want to see and not-see what they don’t want to see. With that said, I’m confident that Tulane’s #9 ranking in both major polls is a more accurate reflection of their accomplishment, and also will be more clearly remembered by college football fans long after the Massey composite ratings are forgotten. I would point out that traditionally, Massey’s final post-bowls rating had more than 73 polls. In 2019, for example, there were 103 ratings in Massey’s final composite. Curious to compare that.

Interesting idea about medians. FWIW, I just checked on that and LSU's median is 11, two below their overall ranking, while Tulane's is 17, one position better than their overall ranking of 18.

As for the number of computers in the MC, yes, that seems to fluctuate, and IIRC one trend is that the final post-bowl number of rankings tend to be lower than the number for the regular season. I guess some computers just do not do a final post-bowl ranking.

Still, 73 rankings is a lot of rankings IMO, more than a critical mass for me to rely on it.

As for Tulane, I agree that their #9 human poll ranking will be remembered, not the MC ranking. Heck, I don't think most CFB fans are even aware of the existence of the MC, whereas everyone knows the human polls.

I just like the MC because, for reasons given earlier, it is IMO a more accurate indicator of how good a team was compared to human polls. Not perfect by any means, and sometimes the results for some teams can be really baffling, but overall better, to me.

The Massey composite ranks by their relative average ranking. So the median isn't real relevant. It could be that all but 8 schools have median of 11 or higher.

(01-29-2023 02:49 PM)Native Georgian Wrote:  

(01-29-2023 01:45 PM)ken d Wrote:  The human polls reflect the subjective opinions of 62 voters.

In all the years I’ve followed college football, I’ve never quite understood how the AP decides which sportswriters — or what # of sportswriters — get a vote in their football/basketball polls. And the same goes for coaches in the UPI poll or whatever it’s called now. I freely admit (because a lot of coaches have already admitted it) that they fill out the Top 5 or Top 10 or Top whatever, and then let an assistant coach or SID fill out the rest.

Still, the combined weight of all those votes — this year it’s 63 coaches and 63 writers (coincidence?) — means that if you combine the two polls, your seeing the aggregate ranking of 126 different people who follow college football very closely.

Once you get past the top 15 or so teams there's very little separating the next group of 15. If you really want to get a good idea who the 25 best teams are, allow the voters to only vote for teams they believe are among the top 15. Then, it's a pretty good bet that the 25 teams with the most votes aren't there just to rank G5 teams to see who goes to the NY6.

The Massey Composite continues adding data.
Up to 76 now

Now up to 80. Georgia is unanimous #1. That is rare, but not (I believe) unprecedented.