bluepenquin
Hall of Fame
4-Time VolleyTalk Poster of the Year (2019, 2018, 2017, 2016), All-VolleyTalk 1st Team (2021, 2020, 2019, 2018, 2017, 2016)
Posts: 12,447
|
Post by bluepenquin on Nov 2, 2014 19:37:46 GMT -5
The projected record is now the rounded projected wins per Pablo rating for all the top conferences. Not really as hard to do as I thought.
1. Texas (24-1, 15-1) 2. Stanford (28-2, 18-2) 3. Florida State (29-1, 17-1) 4. Washington (30-1, 19-1) 5. Wisconsin (27-3, 18-2) 6. North Carolina (25-3, 16-2) 7. Florida (23-5, 17-1) 8. Oregon (22-8, 12-8) 9. Penn State (30-3, 18-2) 10. Colorado State (29-2, 17-1) 11. Illinois (24-7, 16-4) 12. Kentucky (26-5, 15-3) 13. Kansas State (24-6, 11-5) 14. UCLA (21-10, 12-8) 15. Nebraska (19-10, 13-7) 16. Oklahoma (21-9, 11-5) 17. Kansas (20-10, 8-8) 18. USC (17-13, 10-10) 19. Arizona (22-10, 11-9) 20. BYU (24-5, 15-3) 21. UCF (26-6, 19-1) 22. Marquette (24-7, 14-4) 23. Texas A&M (20-9, 12-6) 24. Alabama (25-7, 13-5) 25. Hawaii (21-6, 13-3) 26. Duke (21-8, 13-5) 27. Iowa State (15-12, 7-9) 28. Colorado (19-13, 11-9) 29. Ohio State (21-11, 12-8) 30. Long Beach State (25-5, 15-1) 31. Lipscomb (20-7, 13-1) 32. Miami-FL (19-10, 12-6) 33. Ohio (23-5, 15-1) 34. Western Kentucky (27-5, 15-1) 35. SMU (25-7, 14-6) 36. Arkansas-Little Rock (25-5, 19-1) 37. Loyola Marymount (23-7, 11-7) 38. Illinois State (23-6, 17-1) 39. LSU (18-9, 13-5) 40. Arizona State (19-13, 9-11) 41. Oregon State (18-13, 8-12) 42. Creighton (22-9, 15-3) 43. Purdue (24-8, 14-6) 44. Pittsburgh (25-6, 13-5) 45. San Diego (19-11, 12-6) 46. Cal State Northridge (19-9, 10-6) 47. Santa Clara (22-9, 12-6) 48. Michigan State (18-13, 11-9) 49. Pacific (23-8, 12-6) 50. Louisville (18-12, 10-8) 51. Ole Miss (23-9, 9-9) 52. Xavier (19-11, 12-6) 53. Virginia (18-13, 11-7) 54. Michigan (13-17, 8-12) 55. Baylor (15-16, 5-11) 56. Minnesota (18-13, 8-12) 57. Wyoming (23-8, 12-6) 58. UNLV (25-8, 13-5) 59. Utah (15-16, 5-15) 60. Tulsa (20-11, 12-8) 61. Missouri State (21-9, 14-4) 62. Northwestern (17-14, 7-13) 63. Appalachian State (23-7, 13-7) 64. Temple (23-9, 14-6) 65. Seton Hall (25-8, 13-5) 66. Memphis (24-10, 14-6) 67. Butler (21-9, 13-5) 68. George Washington (20-8, 10-4) 69. Virginia Tech (15-16, 8-10) 70. Rice (23-7, 13-3) 71. American (24-6, 15-1) 72. Texas-Arlington (25-7, 15-5) 73. New Mexico (19-12, 11-7) 74. Wichita State (19-9, 13-5) 75. Dayton (26-6, 13-1) 76. Coastal Carolina (22-6, 11-3) 77. UMKC (24-5, 12-2) 78. Denver (25-6, 13-3) 79. Wisconsin-Milwaukee (17-11, 11-3) 80. VCU (16-12, 9-5) 81. Arkansas State (19-10, 14-6) 82. Harvard (18-5, 11-3) 83. Florida Gulf Coast (19-10, 11-3) 84. West Virginia (15-15, 5-11) 85. Towson (26-5, 12-4) 86. Northern Iowa (20-10, 14-4) 87. LIU Brooklyn (22-7, 13-1) 88. Clemson (20-12, 8-10) 89. UNC Wilmington (21-9, 11-5) 90. NC State (16-14, 5-13) 91. Gonzaga (17-12, 9-9) 92. Texas-San Antonio (18-9, 14-2) 93. Yale (16-8, 11-3) 94. Northern Colorado (18-11, 11-5) 95. Northern Kentucky (19-9, 8-6) 96. Furman (20-7, 13-3) 97. South Carolina (18-13, 8-10) 98. Texas Tech (17-11, 5-11) 99. Northern Illinois (21-9, 14-2) 100. TCU (17-15, 5-11)
|
|
|
Post by trollhunter on Nov 2, 2014 20:09:25 GMT -5
Thanks for posting this, Blue.
Note that some of the bubble teams have conference tournaments, which is not factored in to this (it appears).
In general, I don't think conference tournament will affect RPI significantly, but it could matter in a few cases.
If the team wins their conference tournament, they will get automatic bid anyways, and perhaps a slightly better seed due SOS playing 2-3 more +.500 teams and a couple more wins on overall record.
If the team loses their conference tournament, the SOS gains will probably be offset by the extra loss on their record. May even hurt their RPI if they are upset in the first round. Right?
|
|
bluepenquin
Hall of Fame
4-Time VolleyTalk Poster of the Year (2019, 2018, 2017, 2016), All-VolleyTalk 1st Team (2021, 2020, 2019, 2018, 2017, 2016)
Posts: 12,447
|
Post by bluepenquin on Nov 2, 2014 20:27:38 GMT -5
Thanks for posting this, Blue. Note that some of the bubble teams have conference tournaments, which is not factored in to this (it appears). In general, I don't think conference tournament will affect RPI significantly, but it could matter in a few cases. If the team wins their conference tournament, they will get automatic bid anyways, and perhaps a slightly better seed due SOS playing 2-3 more +.500 teams and a couple more wins on overall record. If the team loses their conference tournament, the SOS gains will probably be offset by the extra loss on their record. May even hurt their RPI if they are upset in the first round. Right? My sense is that the difference will be small - and if it relates to a team on the bubble of getting a seed, it would be within the general margin of error from what I am doing and the actual RPI with their adjustments.
The example where I think has the most likely real world impact is the Big East. I ran the future RPI under the scenario where Creighton beats Seton Hall, then loses to Marquette (after they beat Butler) in the tournament finals. The extra two wins for Marquette lowered their RPI ranking from #22 to #19. Creighton also dropped after going 1-1 from #42 to #37. Even a loss by Creighton in the 1st round has no impact on their RPI.
To answer the question - it depends on the specific case. With the Big East - the extra games is a net plus due to the increased RPI SOS and only taking the top 4 teams. There is likely to be no RPI risk for any of the 4 teams and a net improvement for at least a couple teams.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Nov 2, 2014 20:52:43 GMT -5
The projected record is now the rounded projected wins per Pablo rating for all the top conferences. Not really as hard to do as I thought.
1. Texas (24-1, 15-1) 2. Stanford (28-2, 18-2) 3. Florida State (29-1, 17-1) 4. Washington (30-1, 19-1) 5. Wisconsin (27-3, 18-2) 6. North Carolina (25-3, 16-2) 7. Florida (23-5, 17-1) 8. Oregon (22-8, 12-8) 9. Penn State (30-3, 18-2)
I can't work out the RPI permutations in my head, but based on the last RPI's and Pablo's, I do not understand how you came up with your "projected wins per Pablo" that led to your projected final RPIs: 1. Washington having one loss. (According to Pablo, Wash is #1 favored in all remaining games.) See #2 2. Stanford having two losses. (According to Pablo, SU is favored in all but the Wash game. If SU loses to Wash, who else does it lose to-- and who does Wash lose to, since it would not be SU, according to your analysis?)
|
|
|
Post by mikegarrison on Nov 2, 2014 20:57:26 GMT -5
The projected record is now the rounded projected wins per Pablo rating for all the top conferences. Not really as hard to do as I thought.
1. Texas (24-1, 15-1) 2. Stanford (28-2, 18-2) 3. Florida State (29-1, 17-1) 4. Washington (30-1, 19-1) 5. Wisconsin (27-3, 18-2) 6. North Carolina (25-3, 16-2) 7. Florida (23-5, 17-1) 8. Oregon (22-8, 12-8) 9. Penn State (30-3, 18-2)
I can't work out the RPI permutations in my head, but based on the last RPI's and Pablo's, I do not understand how you came up with your "projected wins per Pablo" that led to your projected final RPIs: 1. Washington having one loss. (According to Pablo, Wash is #1 favored in all remaining games.) See #2 2. Stanford having two losses. (According to Pablo, SU is favored in all but the Wash game. If SU loses to Wash, who else does it lose to-- and who does Wash lose to, since it would not be SU, according to your analysis?) I expect it's something like this: UW is favored in all their remaining matches, but the odds favor that they will lose at least one match, even if there is no individual match they are favored to lose. 70% * 70% = 49%, so if you are favored to win two matches by 70% each, the odds are slightly against you that you will win both matches.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Nov 2, 2014 21:15:26 GMT -5
I can't work out the RPI permutations in my head, but based on the last RPI's and Pablo's, I do not understand how you came up with your "projected wins per Pablo" that led to your projected final RPIs: 1. Washington having one loss. (According to Pablo, Wash is #1 favored in all remaining games.) See #2 2. Stanford having two losses. (According to Pablo, SU is favored in all but the Wash game. If SU loses to Wash, who else does it lose to-- and who does Wash lose to, since it would not be SU, according to your analysis?) I expect it's something like this: UW is favored in all their remaining matches, but the odds favor that they will lose at least one match, even if there is no individual match they are favored to lose. 70% * 70% = 49%, so if you are favored to win two matches by 70% each, the odds are slightly against you that you will win both matches. I haven't taken or utilized statistics in many a moon, but by that analysis, a team that is favored to win each game by 90% would lose its 7th match. It seems to me that the odds need to be freshly looked at after each game instead of being adjusted based upon the initial projection. If someone picks the accurate 50-50 probability that after 6 heads coin flips the next flip will be heads, why would it be any different analytically where the probability is changed? Why would not each ensuing game, assuming no change in Pablo, have the same 90% (or in your example, 70%) probability?
|
|
|
Post by BeachbytheBay on Nov 2, 2014 21:40:26 GMT -5
Interesting comparison of WCC vs. SEC - so which is really the better conference?? using RPI cut-off @ 45, SEC gets 5 teams in, and the WCC 3. WCC has whopping differences between their RPI & Massey
PRI (projected), then followed by Massey
7. Florida (23-5, 17-1) 12. Kentucky (26-5, 15-3) 20. BYU (24-5, 15-3) 23. Texas A&M (20-9, 12-6) 24. Alabama (25-7, 13-5) 37. Loyola Marymount (23-7, 11-7) 39. LSU (18-9, 13-5) 45. San Diego (19-11, 12-6) 47. Santa Clara (22-9, 12-6) 49. Pacific (23-8, 12-6) 51. Ole Miss (23-9, 9-9) 91. Gonzaga (17-12, 9-9)
vs. Massey (SEC with 7 top 100 teams, WCC with 8 top 100 teams)
11. Florida 12. BYU 19. Kentucky 26. LMU 28. Pacific 29. Santa Clara 33. San Diego 36. Texas A&M 44. LSU 50. Alabama 52. Gonzaga 63. San Francisco 78. Ole Miss 98. St. Marys 102. South Carolina
|
|
|
Post by mikegarrison on Nov 2, 2014 22:05:43 GMT -5
I expect it's something like this: UW is favored in all their remaining matches, but the odds favor that they will lose at least one match, even if there is no individual match they are favored to lose. 70% * 70% = 49%, so if you are favored to win two matches by 70% each, the odds are slightly against you that you will win both matches. I haven't taken or utilized statistics in many a moon, but by that analysis, a team that is favored to win each game by 90% would lose its 7th match. It seems to me that the odds need to be freshly looked at after each game instead of being adjusted based upon the initial projection. If someone picks the accurate 50-50 probability that after 6 heads coin flips the next flip will be heads, why would it be any different analytically where the probability is changed? Why would not each ensuing game, assuming no change in Pablo, have the same 90% (or in your example, 70%) probability? No, it doesn't work like that at all. A team that is favored by 90% in each of seven matches would have a 48% chance of winning them all. It gives no information at all about WHICH match you might lose, just that your chances of winning all of them together are slightly less than even odds. But if they win the first one, now there are only six matches, so the chance they will win all six remaining is 53%. If they won the first six, the chance of winning the seventh would be a simple 90%. The point is that even a small chance of losing starts to add up over a large number of chances to lose. But matches you have already won are no longer chances to lose. Your chances of getting "heads" seven times in a row starting from the top are less than 1%, but if you already have six "heads", then the seventh one is 50%. Thinking that since you already got 6 heads in a row, the chance of getting the seventh one must be 1% is so wrong (but so common) that it has a specific name: "the gambler's fallacy." Wikipedia has an amusing story about a real-life case of this happening: 26 blacks in a row. According to the article, people lost a lot of money by betting red, thinking red was "due". In real life, if you see the roulette wheel land on black 26 times in a row you probably should bet black, because there may be something wrong with the wheel!
|
|
|
Post by mikegarrison on Nov 2, 2014 22:19:52 GMT -5
ps. It's not different than what we all have been saying here all along -- that it doesn't seem likely either team is going to make it through the conference schedule unbeaten. But the longer they do go unbeaten, the fewer matches remain, and the more likely it becomes that one of them (can't be both!) will do it.
|
|
bluepenquin
Hall of Fame
4-Time VolleyTalk Poster of the Year (2019, 2018, 2017, 2016), All-VolleyTalk 1st Team (2021, 2020, 2019, 2018, 2017, 2016)
Posts: 12,447
|
Post by bluepenquin on Nov 2, 2014 22:21:58 GMT -5
Here is the WP of each of Washington's final 8 matches producing a win expectancy of 6.76 more wins.
Oregon 78.9% Oregon State 92.7% at Colorado 83.0% at Utah 89.8% Arizona 84.9% Arizona State 90.1% Stanford 64.2% at Washington State 92.1%
and here is Stanford for a WE of 6.16 more wins:
Arizona State 86.4% Arizona 80.1% at USC 73.6% at UCLA 67.7% Utah 92.6% Colorado 87.0% at Washington 35.8% at Cal 92.3%
|
|
|
Post by mikegarrison on Nov 2, 2014 22:24:58 GMT -5
The simple way to do this is to just work out the percentages for each team, ignoring that they are not independent of the results of each other team. I would guess that is what Blue is doing. The more correct way is to run a simulation -- randomly (according to the probabilities) assign victors and losers to each upcoming match. Record the outcomes. Then do it again. And again. Lather, rinse, repeat until you have a respectable number of randomized "seasons" (a few thousand, maybe).
Then average out the results of each simulated season. That will give you something similar to the simple method, but will include the effects of the outcomes of the matches not actually being independent.
|
|
|
Post by alpacaone on Nov 2, 2014 22:52:41 GMT -5
Interesting I've never fully understood how to look at your "gambler's fallacy," a 1/128 odds of flipping heads 7 times comes down to a 50/50 chance, and still a 50/50 chance for anyone wanting in on the last flip? It's still hard for me to realize, I do see it, but it just doesn't seem right. It's my nine year olds probability question about a family with three boys wanting to know the odds of getting another boy if they have two more kids, and if the odds were even it's 3/4 regardless of the earlier boys. I get it, it just doesn't seem right.
|
|
|
Post by mikegarrison on Nov 2, 2014 23:43:46 GMT -5
Interesting I've never fully understood how to look at your "gambler's fallacy," a 1/128 odds of flipping heads 7 times comes down to a 50/50 chance, and still a 50/50 chance for anyone wanting in on the last flip? It's still hard for me to realize, I do see it, but it just doesn't seem right. It's my nine year olds probability question about a family with three boys wanting to know the odds of getting another boy if they have two more kids, and if the odds were even it's 3/4 regardless of the earlier boys. I get it, it just doesn't seem right. Ah, but you see coin tosses and roulette wheels are intentionally randomized. If a family has three boys already, that could be random. Or it could be something about the sperm of the father that favors boys. You have to be careful about applying the gambler's fallacy idea to real world situations. But yeah, assuming the odds of a boy or a girl actually are 50/50 biologically for those parents, then the number of previous boys is irrelevant. Once you flip the coin it is flipped, and it doesn't affect the next toss. That's the assumption behind the pablo calculations too, that the outcome of each match is independent of the other matches. Athletes train a long time to make that happen -- to play one game at a time, one rally at a time, one hit at a time. But we know that's not always possible, and sometimes in the real world, with real people, what you did in the last match may have some influence on what you do in the next match. That's not math; that's psychology. And that's why it is hard for people to intuitively believe it. Our brains evolved to help us understand a world where effects have causes, and repeatable events repeat with patterns. It rained three days in a row, and storm cells are only so big, so we really are "due" for the rain to be over. But gambling devices like coins and dice and cards are carefully designed to be as random as possible. Nothing in our normal world is really so independent of all cause and effect, which is why our brains get a little confused about probabilities.
|
|
|
Post by badgerbreath on Nov 3, 2014 0:01:34 GMT -5
The simple way to do this is to just work out the percentages for each team, ignoring that they are not independent of the results of each other team. I would guess that is what Blue is doing. The more correct way is to run a simulation -- randomly (according to the probabilities) assign victors and losers to each upcoming match. Record the outcomes. Then do it again. And again. Lather, rinse, repeat until you have a respectable number of randomized "seasons" (a few thousand, maybe). Then average out the results of each simulated season. That will give you something similar to the simple method, but will include the effects of the outcomes of the matches not actually being independent. I imagine this is right. It would be interesting to see if the assumption that the results are not independent results in some impossible results with respect to records among teams. A simulation randomly sampling from a distribution based on Pablo differences would be more robust, but also more of a pain to do. It's trickier using the direct calculation method to assess how many losses a team will have if it is likely to have more than one. That means assigning a probability to every combination of 1 loss,2 losses, 3 losses, 4 losses etc., and adding them up to determine if 1, 2, 3 or 4 losses is more likely. Of course the probabilities may be pretty similar in the end...like 10% for 1,18% for 2,23% for 3 and 13% for 4...and none of them greater than 50%. This method of displaying the results only allows you to state the most likely outcome, even it is only likely to occur 23% of the time. So there can be a lot of nuance to the results that isn't obvious in these tables. Hard to really present that nuance though. I appreciate what we get.
|
|
|
Post by ay2013 on Nov 3, 2014 0:08:45 GMT -5
You are all nerds
|
|