Post by mikegarrison on Sept 20, 2015 3:51:19 GMT -5
I'm going to do something different this year. I'm going to keep a weekly updated pablo predicted finish. Here's how it will work:
1) Every Monday, Pablo will come out with new ratings and a new HCA.
2) I have built a matrix of who plays who, and where. I will put the new ratings in and it will update the probabilities for future matches.
3) For past matches, I will put in who actually won and lost.
4) I will calculate the expected W-L records for every team.
Note! These are not, strictly speaking, predictions of which team will beat which team. They are predictions of the most likely final records. Think of these like the "expected wins and losses" you see in many baseball standings. However, since I will be updating them with actual wins and losses as well as updated pablo ratings, they should converge to the actual records by the end of the year.
Based on the Sept 14 pablo, here are the predicted records and predicted average winning probabilities:
Washington (15-5) .766
USC (15-5) .731
UCLA (15-5) .729
Stanford (14-6) .699
ASU (12-8) .584
Arizona (11-9) .553
Oregon (10-10) .486
WSU (7-13) .346
Colorado (7-13) .345
Utah (7-13) .342
OSU (5-15) .238
Cal (4-16) .182
[updated: The current numbers are now just the sums of the expected wins and losses. My earlier attempt turned out to predict a team would go 20-0 if their actual record was 18-2. That's not the desired behavior! The current model now does converge to the actual record when the predictions are replaced by results.]
1) Every Monday, Pablo will come out with new ratings and a new HCA.
2) I have built a matrix of who plays who, and where. I will put the new ratings in and it will update the probabilities for future matches.
3) For past matches, I will put in who actually won and lost.
4) I will calculate the expected W-L records for every team.
Note! These are not, strictly speaking, predictions of which team will beat which team. They are predictions of the most likely final records. Think of these like the "expected wins and losses" you see in many baseball standings. However, since I will be updating them with actual wins and losses as well as updated pablo ratings, they should converge to the actual records by the end of the year.
Based on the Sept 14 pablo, here are the predicted records and predicted average winning probabilities:
Washington (15-5) .766
USC (15-5) .731
UCLA (15-5) .729
Stanford (14-6) .699
ASU (12-8) .584
Arizona (11-9) .553
Oregon (10-10) .486
WSU (7-13) .346
Colorado (7-13) .345
Utah (7-13) .342
OSU (5-15) .238
Cal (4-16) .182
[updated: The current numbers are now just the sums of the expected wins and losses. My earlier attempt turned out to predict a team would go 20-0 if their actual record was 18-2. That's not the desired behavior! The current model now does converge to the actual record when the predictions are replaced by results.]
