FCS Tournament Expansion.

[ASUG8]

Upsets happen, John - close games happen when you don't expect them. You of all people are extrapolating on a single data point.

Yes that is true. But it's still the case that Stony Brook comes from a conference that many people would say shouldn't get an automatic bid and they put up a very respectable performance.

Anyway, the most important thing to me is having a system that takes human belief out of it to the greatest extent possible. And one thing that contributes to moving in that direction is to make sure that every conference champion gets an automatic bid.

Another thing they'd do if I had my way is get rid of the selection committee concept and replace it with a system that uses a power rating system that everybody agrees on before the season to fill out the out large bids. Human opinion is involved there because people have to have opinions on what constitutes the best power rating system for doing that. But it's involved BEFORE the season starts. It's conceptual. Everybody agrees on what goes into picking the teams that most deserve to get at large bids ahead of ti
me then quantify it. So AFTER the fact there is no bias. The chips fall where they may.

I agree with you JSO, but let me know when you find a fully objective human or computer system. It's an ideal we likely won't achieve.

[JohnStOnge]

I agree with you JSO, but let me know when you find a fully objective human or computer system. It's an ideal we likely won't achieve.

At some level one can always say there's some subjectivity...some opinion...about some aspect of the system. But let's say we say that the best power rating system does the best job of predicting outcomes in terms of bias and mean squared error with respect to how the predicted point spreads compare to the actual point spreads. Ok. So we have had an opinion with respect to what criteria should be used to choose the best power rating system (model).

But once we do that then we have an objective system. They system is just trying to minimize the bias and means squared error. Totally objective.

In that sense, power rating systems are far more objective than human beings can ever be. They don't "care" about what the name of the school is.

And that's what I'm getting at. Let the "subjective" part transpire before the season. Let it be about how accomplishment is quantitatively measured before anyone knows which teams are going to be in the running. Then, once the season starts, it's "hands off" as far as human beings go. The criteria for what constitutes an accomplishment level has been established and the chips fall where they may.

To me that's way better than having a selection committee come in after the fact with no real objective set of criteria at all.

[putter]

you will default to what the human selection committee does. For the most part the SoCon and CAA's SOS should be higher than Patriot and/or NEC. The computer model should reflect this and the committee goes into the at-large selection knowing this too. Is it fair, well it's as close to the BCS as the FCS wants to admit...the power conferences are going to get the bulk of the at-large bids most years.

[89Hen]

In that sense, power rating systems are far more objective than human beings can ever be.

Yup, more objective, but not more accurate. Here we are, 6-7 years after I first challenged you to a human vs computer match-up and you're still ducking me.

[BlueHen86]

It won't be long before FCS gets to a 32 team field. Expanding from 16 teams to 20 teams was a big jump becasue ti added a week to the playoffs. Expanding to 24, then 32 doesn't add an extra week.

does that mean that since the hoops tourney expanded to 65 (and now 68), that it will inevitably expand to 128 (since that doesn't add any additional rounds?)

Maybe. There has been talk...

[BlueHen86]

It won't be long before FCS gets to a 32 team field. Expanding from 16 teams to 20 teams was a big jump becasue ti added a week to the playoffs. Expanding to 24, then 32 doesn't add an extra week.

Except that there aren't any more conferences that want auto bids, & there's still room to add one without expanding the field since there are only going to be 11 auto bids.

I still see 32 teams in our future. I hope not, but I think it will happen.

[Mvemjsunpx]

Except that there aren't any more conferences that want auto bids, & there's still room to add one without expanding the field since there are only going to be 11 auto bids.

I still see 32 teams in our future. I hope not, but I think it will happen.

I suppose it's possible, but DII has been stuck at 24 teams for a while.

Unless we end up with noticeably more conferences than we have now, it seems unlikely. There just wouldn't be a real incentive to expand otherwise.

[JohnStOnge]

In that sense, power rating systems are far more objective than human beings can ever be.

Yup, more objective, but not more accurate. Here we are, 6-7 years after I first challenged you to a human vs computer match-up and you're still ducking me.

I don't think that I ducked you. I remember having conversations in which I referred people to a page showing that the better power rating systems basically equivalent to the closing Vegas line in terms of picking winners and losers, minimizing bias (being very close to bias = 0 just as the Vegas line is), and minimizing the average departure by the predicted score from the actual score. You may be the one who said you don't think they're as "accurate" for I-AA games but then that web site did I-AA games for a few years and the results suggested that while there may be a difference it isn't large. During the four years when they did that, for instance, the average percent correct over a given season for Sagarin was 74.2% for I-A games vs. 71.8% for I-AA games.

I remember someone comparing those kinds of accuracy rates to an accuracy rate for predictions being made by fans on some message board or something that had to do with the CAA but then I saw something that I interpreted as meaning they only picked games involving top 25 teams so that the results couldn't really be compared.

I actually suggested having a human vs. computer matchup but never got around to doing it. But if you want to do it this year we can. What it would take to compare a human to a power rating system, though, would be either to pick ALL games involving FCS teams or a probability sample of such games. We can use Sagarin since it's easy to access and easy to derive the predictions. We'd wait until Sagarin's site says the ratings are unbiased...meaning there have been enough games to get rid of subjectivity in the system. And we'd look at percent correct, bias, and mean error.

As I said before what I predict that what will happen is that if we get a good human expert the percent correct comparison will probably be close. It will probably not be a "statistically significant" difference at 95% confidence. But Sagarin will absolutely kill the human expert when it comes to the bias and mean error estimates.

Remind me and we will do it. You or whoever you want the human expert to be can either post or e mail me their predictions for who will win each FCS game and by how much. Then I will keep track of and run the numbers.

Oh...one more thing: Just one human expert. The reason for that is that as you make more comparisons the higher the odds get that at least one human expert will do "significantly" better than the system just by random chance. So it has to be a single comparison. Also, the human expert has to promise not to look at any power rating systems to assist him or her. We'll have to go on the honor system for that but it'll be important if we really want to compare the human expert thought process to the power rating output.