Ideal college football (and basketball) ranking system

[ISUinOR]

I was thinking about how bad the "human element" in the football and basketball ranking systems are and wondered if there was an ideal solution. This is a bit long, but provides a good pro/con for using betting services and prediction systems to rank college FB and BB. What is the IDEAL solution? I think a hybrid system that includes these options would be better than what we have. What do you think?

ChatGPT offered some interesting feedback. Essentially, market-average prediction systems (Vegas sportsbooks, betting analytics companies, market-based prediction systems) would essentially create a market-driven, probability-based ranking system.

Pros:
1. Betting markets are highly efficient
This usually makes their implied team strength ratings more accurate than human polls
2. No bias issues
Markets don’t care about brands—only probabilities that make money'
3. Continuously updated
This produces rankings that reflect true current strength, not reputation
4. Well-correlated with scoring margin
Better than human voters

Cons:
1. Betting markets predict performance, not résumé
Markets predict who would win on a neutral field today
2. Public money influence
While odds are mostly driven by sharp bettors and algorithms, large public markets inflate big brands and deflate low-visibility teams
3. Not transparent
Books do not publicly disclose internal power ratings, how injury adjustments are weighted, how much of the line is public vs. professional influence. A ranking with hidden methodology gets criticized immediately.
4. Financial incentives distort the model
Books set odds to balance risk and maximize profit—not to create the truest ranking
5. Predictive ≠ deserves
College sports selection committees explicitly value résumé: Wins over ranked teams, championships, road/neutral wins
A market-based rating ignores all that.

A market-based rating would provide human-friendly rankings, analytics-supported rankings, and market-validated rankings all at once.

Final verdict, according to ChatGPT:
A betting-market consensus Top 25 would be accurate for predicting who is best right now, but would not be ideal as the sole ranking system for postseason selection.
It’s valuable—but only as one piece of a larger ranking ecosystem.

Don't you think a market-average prediction ranking system, utilizing an analytics-based rankings from sportsbooks, betting analytics companies, etc. would be able to provide a better system than we have today with the idiocy of the humans in the room?

 

[ISUinOR]

Media companies, fans, etc. demand rankings for clicks, etc., be it preseason or during post-season selection time.

Let them ALL put out their own weekly (or daily) rankings of the top teams in FB or BB to garner their clicks and eyeballs. For football, maybe half-way through the season (week 6/7 ), they release the "analytics-driven" rankings. That gives all of the media companies, etc. a HUGE opportunity to give their own breakdowns of how their rankings compare to the "analytics".

Release a Week 9 update to the analytics ranking. Again, creates a bit more excitement and stress about where teams sit and how they can possibly improve their ranking.

Then, release a final analytics ranking of the top 25 after championship week that determines who makes the playoffs of not.

Then, use those same rankings to fill the non-playoff match-ups for bowl games. Top ranked teams after those in the playoff face each other in whatever bowl game. If it's 13 vs 14, 15 vs 17, etc. may make it more competitive.

 

[Cyclonepride]

I totally agree that the human element is a major problem (especially when the human element generally consists of small numbers of people who are personally and professionally connected to certain big players).

I'd prefer any system that takes that out of it as much as possible

 

[BryceC]

Another con would be the incentive to really run up the score to beat spreads.

 

[Cyhig]

Let's have various computers using different parameters come up with a ranking system and them compile all of them together to form the playoff teams

The BCS is alive again!

 

[FriendlySpartan]

I totally agree that the human element is a major problem (especially when the human element generally consists of small numbers of people who are personally and professionally connected to certain big players).

I'd prefer any system that takes that out of it as much as possible

Yep let’s go with the computers who moved Gonzaga up the rankings after getting blown out by 40.

The human element is there for a reason. Most of the computer systems are staggeringly flawed especially for football.

 

[FriendlySpartan]

I was thinking about how bad the "human element" in the football and basketball ranking systems are and wondered if there was an ideal solution. This is a bit long, but provides a good pro/con for using betting services and prediction systems to rank college FB and BB. What is the IDEAL solution? I think a hybrid system that includes these options would be better than what we have. What do you think?

ChatGPT offered some interesting feedback. Essentially, market-average prediction systems (Vegas sportsbooks, betting analytics companies, market-based prediction systems) would essentially create a market-driven, probability-based ranking system.

Pros:
1. Betting markets are highly efficient
This usually makes their implied team strength ratings more accurate than human polls
2. No bias issues
Markets don’t care about brands—only probabilities that make money'
3. Continuously updated
This produces rankings that reflect true current strength, not reputation
4. Well-correlated with scoring margin
Better than human voters

Cons:
1. Betting markets predict performance, not résumé
Markets predict who would win on a neutral field today
2. Public money influence
While odds are mostly driven by sharp bettors and algorithms, large public markets inflate big brands and deflate low-visibility teams
3. Not transparent
Books do not publicly disclose internal power ratings, how injury adjustments are weighted, how much of the line is public vs. professional influence. A ranking with hidden methodology gets criticized immediately.
4. Financial incentives distort the model
Books set odds to balance risk and maximize profit—not to create the truest ranking
5. Predictive ≠ deserves
College sports selection committees explicitly value résumé: Wins over ranked teams, championships, road/neutral wins
A market-based rating ignores all that.

A market-based rating would provide human-friendly rankings, analytics-supported rankings, and market-validated rankings all at once.

Final verdict, according to ChatGPT:
A betting-market consensus Top 25 would be accurate for predicting who is best right now, but would not be ideal as the sole ranking system for postseason selection.
It’s valuable—but only as one piece of a larger ranking ecosystem.

Don't you think a market-average prediction ranking system, utilizing an analytics-based rankings from sportsbooks, betting analytics companies, etc. would be able to provide a better system than we have today with the idiocy of the humans in the room?

Someone is too reliant on AI because you don’t know how gambling lines work…

 

[Cyclonepride]

Yep let’s go with the computers who moved Gonzaga up the rankings after getting blown out by 40.

The human element is there for a reason. Most of the computer systems are staggeringly flawed especially for football.

Most humans are staggeringly easy to corrupt, especially in small numbers. If it has to have a human element, make it as wide as possible

 

[FriendlySpartan]

Most humans are staggeringly easy to corrupt, especially in small numbers. If it has to have a human element, make it as wide as possible

And guess who the computer systems are programmed by…an even smaller group of humans

 

[Dale]

The human element is there for a reason. Most of the computer systems are staggeringly flawed especially for football.

College football is the most challenging sport to have computer rankings for because the sample size is so small. There's a dearth of data with only 12 games, particularly if you hamper yourself by not including MoV.

 

[MeowingCows]

The real problem is early and mid-season rankings with very small samples. Metrics-based rankings work fine later on. They are volatile early when minimal data exists.

Don't rank CFB until the ~9 games played mark or so, and don't rank CBB until about 20 games played. The main issue with rankings is the necessity for constant rankings, instead of a few rank intervals looking at the big picture.

 

[Dale]

Another con would be the incentive to really run up the score to beat spreads.

An idea I've had in my head for a year or two now is a ranking system where a team's MoV is included, but ONLY for their opponents' strength of schedule benefit. For example, Iowa State would get credit for beating Purdue, of course, but the margin of victory would only come into play in the strength of schedule for TCU, Kansas, Iowa, UCF, etc. Iowa State would get no bump for the size of the victory. This would allow MoV to still have an impact on making the accuracy of the rankings as good as possible, but teams wouldn't have an incentive to run up the score. (Not that ISU did that versus Purdue!) I don't think any other system does this, so someday when I have some time, I might try to implement it.

 

[KennyPratt42]

There should just be a group of about 40 people selected (people like Josh Pate, Chris Fallica, Joel Klatt) that vote on a top 20 that gets released every Tuesday starting after week 3. Nobody is included that is employed by a University or Conference or covers a specific team on a weekly basis. If you wanted to include a few computer rankings systems in that 40 that would be fine, but no more than a handful.

 

[CascadeClone]

You could use Vegas betting models along with other computer models - its just another computer model, basically. The bigger issue in my mind is the predictive vs. achieved argument.

One team goes 11-1 and another goes 9-3. But the predictive metrics say the 9-3 is better. What to do?

The CFP has basically said "the hell with achieved, we want the BEST teams, and predictive is that". This infuriates people (rightly) who then say "why play the games?" ESPECIALLY people that aren't fans of SEC or blue bloods that think they are getting screwed.

For me the answer would be to weight both to some degree and blend it. In the example above, lets say the predictive measure is season wins to match the achieved season wins.
Achieved Predictive vs avg schedule
11-1 9.6-2.4
9-3 10.4-1.6
So if you blend it 50/50 then the 11-1 team is 10.3 and the 9-3 team is 9.7 so the 11-1 team gets in.
If you blend 80/20 in favor of predictive, then 11-1 is 9.88 and 9-3 is 10.12, so the 9-3 team gets in.

That's how I would do it, and probably 50/50. You should be a lot better in theory than someone that was actually better in reality, in order to jump them.

You will NEVER get a "right" answer imho, and you ABSOLUTELY NEVER will make everyone happy. But if you put in some of both I think it makes sense.

 

[CascadeClone]

College football is the most challenging sport to have computer rankings for because the sample size is so small. There's a dearth of data with only 12 games, particularly if you hamper yourself by not including MoV.

That's true, but the only thing worse than using computer rankings is NOT using them.

I think you have to average a handful of good computers, and then blend with human rankings too. Grab 10 best computer models, grab 10 honest and informed humans (if there are 10 left to be found lol) and then jam it all together.

 

[FriendlySpartan]

That's true, but the only thing worse than using computer rankings is NOT using them.

I think you have to average a handful of good computers, and then blend with human rankings too. Grab 10 best computer models, grab 10 honest and informed humans (if there are 10 left to be found lol) and then jam it all together.

You do realize that the playoff committee has basically picked exactly what the computers have most years. The seeding might be a touch different but really your looking for a solution to a problem that doesn’t exist

 

[Cyclonepride]

And guess who the computer systems are programmed by…an even smaller group of humans

The weighing of the various factors can be made by wide agreement and the code can be made public. I'm not saying you can or should eliminate human involvement completely, but I definitely do not think it should be made by a small group of people specifically selected by the powers that be.

 

[Cyched]

1-888-BETSOFF

The 1-888-BETSOFF helpline has been a resource to problem gamblers since 2008. Call the helpline today to receive free help.

888betsoff.org

 

[FriendlySpartan]

The weighing of the various factors can be made by wide agreement and the code can be made public. I'm not saying you can or should eliminate human involvement completely, but I definitely do not think it should be made by a small group of people specifically selected by the powers that be.

It always is, even with your computer models. It’s a small group of people designing them. For football this is impossible, for basketball it’s a touch more doable but still has major issues

 

[Cyclonepride]

It always is, even with your computer models. It’s a small group of people designing them. For football this is impossible, for basketball it’s a touch more doable but still has major issues

You can design a computer model based on predetermined and widely agreed upon factors and weighting on those factors. The fact that just a few people build the actual model is immaterial. That's like giving credit for a bridge design to the construction workers that built the bridge. It's not like those programmers can add their own agenda into the models if the code is transparent.