Now's NCAA Doing With RPI
4 posters
Page 1 of 1
Now's NCAA Doing With RPI
According to my numbers, the NCAA's RPI made the correct prediction in all 16 of the seeded teams' games -- they all won. That's no surprise, since in each four-team group, the seeded team played the team with the weakest RPI.
What's interesting is that of the games between non-seeded teams, the RPI was right in 10 games and wrong in 6. One higher ranked team won a shootout game and two lower ranked teams won shootout games. I counted that as one correct prediction and two incorrect predictions. I could have counted it as three incorrect predictions, but didn't. Of, if we use the RPI formula's treatment of a tie (which is how the RPI treats shootouts), then the RPI was right in 10 1/2 games and wrong in 5 1/2. Is that a good percentage of correct predictions for the RPI?
What's interesting is that of the games between non-seeded teams, the RPI was right in 10 games and wrong in 6. One higher ranked team won a shootout game and two lower ranked teams won shootout games. I counted that as one correct prediction and two incorrect predictions. I could have counted it as three incorrect predictions, but didn't. Of, if we use the RPI formula's treatment of a tie (which is how the RPI treats shootouts), then the RPI was right in 10 1/2 games and wrong in 5 1/2. Is that a good percentage of correct predictions for the RPI?
Re: Now's NCAA Doing With RPI
Interesting... I guess it's not a terrible percentage, but it's clearly not a perfect system.
However, I would say that a soccer can be more prone to an "upset" between two somewhat similarly skilled teams, just due to how hard it is to create goals, while at the same time how "easy" it is to score on a freak play/corner kick/free kick etc.
Wait... does that even make sense? I think it's getting late.
However, I would say that a soccer can be more prone to an "upset" between two somewhat similarly skilled teams, just due to how hard it is to create goals, while at the same time how "easy" it is to score on a freak play/corner kick/free kick etc.
Wait... does that even make sense? I think it's getting late.
Stonehouse- Draft Pick
- Number of posts : 3242
Age : 42
Location : Portland, OR
Registration date : 2007-06-07
Re: Now's NCAA Doing With RPI
On another thread, I showed how Albyn Jones did. 2 misses in predicting losers, 5 in predicting winners. The tie-gray area accounted for the difference.
Geezaldinho- Pilot Nation Legend
- Number of posts : 11794
Location : Hopefully, having a Malbec on the square in Cafayate, AR
Registration date : 2007-04-28
Re: Now's NCAA Doing With RPI
RPI was right in 10 1/2 games and wrong in 5 1/2. Is that a good percentage of correct predictions for the RPI?
Isn't that 2 1/2 games away from flipping coins?
Geezaldinho- Pilot Nation Legend
- Number of posts : 11794
Location : Hopefully, having a Malbec on the square in Cafayate, AR
Registration date : 2007-04-28
Re: Now's NCAA Doing With RPI
I'd be interested to see what the prediction rate is for men's basketball brackets.... don't 15 seeds seem to get upset every year.... and wasn't something like last year was the first time all #1 seeds play... if we just flipped coins, or advanced the seeded teams why bother playing the game?
MSPDX- Recruit
- Number of posts : 3
Registration date : 2007-11-13
Re: Now's NCAA Doing With RPI
After thinking about it, I think the correct way to evaluate the RPI as a predictive tool is to treat a tie as a missed prediction, since the team with the higher RPI is supposed to be better. (After the tournament, among other RPI-related things, I'll try to do an analysis of what the RPI's "standard error" is, although one tournament probably doesn't provide enough data to really do that reliably.)
As I said previously, I don't think the seeded teams' first round performances count for much in an analysis, since all those games were "fixed" for the seeds to win by assigning them the weakest teams in their four-team groups.
Of the non-seeded teams' first round games, the RPI was correct 10 times and incorrect 6 times (again, counting tie games as the RPI being incorrect, no matter how close the teams' RPI numbers were). Of all of the second round games, the RPI so far is correct 9 times and incorrect 6 times. Assuming Penn State defeats Hofstra, it will be 10-6 for the second round. So, for the "non-fixed" first and second round games, the RPI is correct 62.5% of the time and incorrect 37.5 of the time.
FYI, in a system that provides a "standard error" also called a "standard deviation," the interval of "rating +/- SE" should contain the correct rating 68% of the time; and the interval of "rating +/- 2xSE" should provide the correct rating 96% of the time. This is based on the assumption that the rating has a normal (bell-shaped) distribution. It also is less accurate for teams with extreme records (all wins or all losses, for example). This explanation is thanks to Albyn Jones of SoccerRatings.
Of course, when I attempt after the tournament to define a "standard error" for the RPI, I'll need to use all first round games. Still, the experience so far with the games suggests to me that the "standard error" for the RPI will be quite large.
I have the impression there's another statistic-based rating system out there for women's college soccer, in addition to the SoccerRatings system. Is that correct and, if so, is there a website where that system resides?
As I said previously, I don't think the seeded teams' first round performances count for much in an analysis, since all those games were "fixed" for the seeds to win by assigning them the weakest teams in their four-team groups.
Of the non-seeded teams' first round games, the RPI was correct 10 times and incorrect 6 times (again, counting tie games as the RPI being incorrect, no matter how close the teams' RPI numbers were). Of all of the second round games, the RPI so far is correct 9 times and incorrect 6 times. Assuming Penn State defeats Hofstra, it will be 10-6 for the second round. So, for the "non-fixed" first and second round games, the RPI is correct 62.5% of the time and incorrect 37.5 of the time.
FYI, in a system that provides a "standard error" also called a "standard deviation," the interval of "rating +/- SE" should contain the correct rating 68% of the time; and the interval of "rating +/- 2xSE" should provide the correct rating 96% of the time. This is based on the assumption that the rating has a normal (bell-shaped) distribution. It also is less accurate for teams with extreme records (all wins or all losses, for example). This explanation is thanks to Albyn Jones of SoccerRatings.
Of course, when I attempt after the tournament to define a "standard error" for the RPI, I'll need to use all first round games. Still, the experience so far with the games suggests to me that the "standard error" for the RPI will be quite large.
I have the impression there's another statistic-based rating system out there for women's college soccer, in addition to the SoccerRatings system. Is that correct and, if so, is there a website where that system resides?
Re: Now's NCAA Doing With RPI
another predictive rating system is Massey: http://www.mratings.com/rate.php?lg=csocw
His system includes goals as part of the rating.
You pair team "A's" offensive rating against team "B's" defensive rating, Plus the reverse, add the home field advantage, and it will tell you who wins and by how much.
At least that's how I think it works. He doesn't really have a full explanation for soccer.
But if you assume I'm right, last weekend we had a rating number against Denver of +3.17
We rated a +1.97 against UC yesterday, with home field (.36 goals) factored in.
He has his standard deviation as 1.84 goals, so both games worked in his system, within one standard dev.
I'll probably play with the brackets later this week a bit, to see how well he did overall.
Just subtracting power ratings seems to do about the same thing.
His system includes goals as part of the rating.
You pair team "A's" offensive rating against team "B's" defensive rating, Plus the reverse, add the home field advantage, and it will tell you who wins and by how much.
At least that's how I think it works. He doesn't really have a full explanation for soccer.
But if you assume I'm right, last weekend we had a rating number against Denver of +3.17
We rated a +1.97 against UC yesterday, with home field (.36 goals) factored in.
He has his standard deviation as 1.84 goals, so both games worked in his system, within one standard dev.
I'll probably play with the brackets later this week a bit, to see how well he did overall.
Just subtracting power ratings seems to do about the same thing.
Geezaldinho- Pilot Nation Legend
- Number of posts : 11794
Location : Hopefully, having a Malbec on the square in Cafayate, AR
Registration date : 2007-04-28
Page 1 of 1
Permissions in this forum:
You cannot reply to topics in this forum