Bayesian 101 - Printable Version
+- The CardBoard (https://thecardboard.org/board)
+-- Forum: C-House! (https://thecardboard.org/board/forum-4.html)
+--- Forum: The CardBoard (https://thecardboard.org/board/forum-5.html)
+--- Thread: Bayesian 101 (/thread-10101.html)
+- The CardBoard (https://thecardboard.org/board)
+-- Forum: C-House! (https://thecardboard.org/board/forum-4.html)
+--- Forum: The CardBoard (https://thecardboard.org/board/forum-5.html)
+--- Thread: Bayesian 101 (/thread-10101.html)
Bayesian 101 - BostonCard - 10-04-2015
https://twitter.com/TedMillerRK/status/650790870459879428
It's hard to know what the quoted probability means. It could be that they ran 10,000 simulations and looked at how many losses the Pac-12 champs had, whoever it was. Alternatively, they could have looked at the maximum likelihood number of losses for each team (Stanford is projected to end at 9.9 - 3, which implies a 90% chance that we will make the Pac-12 championship) and weighted according to the probability that each team wins the championship (it's 40% for Stanford, by the way). If it is the latter, they are doing it wrong. For example, the fewer games we lose, the greater the likelihood that we wind up as Pac-12 champions (indeed, our probability of winning out is 8%).
In the extreme case, consider that Stanford and USC run the table and meet in the Pac-12 championship. Though this scenario guarantees that the Pac-12 champion will have only 1 loss, if the teams are viewed as more or less even, each team will be given a 50% chance of winning the championship, and a 50% chance of having two losses. Thus, if you used a non-Bayesian approach, you would argue that the Pac-12 champion had a 50% chance of having two losses.
BC
Re: Bayesian 101 - Rally - 10-04-2015
(10-04-2015, 09:34 PM)Boston Card link Wrote:https://twitter.com/TedMillerRK/status/650790870459879428
It's hard to know what the quoted probability means. It could be that they ran 10,000 simulations and looked at how many losses the Pac-12 champs had, whoever it was. Alternatively, they could have looked at the maximum likelihood number of losses for each team (Stanford is projected to end at 9.9 - 3, which implies a 90% chance that we will make the Pac-12 championship) and weighted according to the probability that each team wins the championship (it's 40% for Stanford, by the way). If it is the latter, they are doing it wrong. For example, the fewer games we lose, the greater the likelihood that we wind up as Pac-12 champions (indeed, our probability of winning out is 8%).
In the extreme case, consider that Stanford and U$¢ run the table and meet in the Pac-12 championship. Though this scenario guarantees that the Pac-12 champion will have only 1 loss, if the teams are viewed as more or less even, each team will be given a 50% chance of winning the championship, and a 50% chance of having two losses. Thus, if you used a non-Bayesian approach, you would argue that the Pac-12 champion had a 50% chance of having two losses.
BC
I'm missing something. Not for the first time. In your scenario, where Stanford and U.Sc enter the championship game with one loss each, how can the Pac12 Champion have two losses?
Re: Bayesian 101 - BostonCard - 10-04-2015
(10-04-2015, 09:42 PM)ferrari link Wrote:[quote author=Boston Card link=topic=13047.msg132754#msg132754 date=1444019651]
https://twitter.com/TedMillerRK/status/650790870459879428
It's hard to know what the quoted probability means. It could be that they ran 10,000 simulations and looked at how many losses the Pac-12 champs had, whoever it was. Alternatively, they could have looked at the maximum likelihood number of losses for each team (Stanford is projected to end at 9.9 - 3, which implies a 90% chance that we will make the Pac-12 championship) and weighted according to the probability that each team wins the championship (it's 40% for Stanford, by the way). If it is the latter, they are doing it wrong. For example, the fewer games we lose, the greater the likelihood that we wind up as Pac-12 champions (indeed, our probability of winning out is 8%).
In the extreme case, consider that Stanford and U$¢ run the table and meet in the Pac-12 championship. Though this scenario guarantees that the Pac-12 champion will have only 1 loss, if the teams are viewed as more or less even, each team will be given a 50% chance of winning the championship, and a 50% chance of having two losses. Thus, if you used a non-Bayesian approach, you would argue that the Pac-12 champion had a 50% chance of having two losses.
BC
I'm missing something. Not for the first time. In your scenario, where Stanford and U.Sc enter the championship game with one loss each, how can the Pac12 Champion have two losses?
[/quote]
That is exactly the point; in that scenario, both situations (Pac-12 champions and having only 1 loss is caused by the same thing, winning the game). But if you are computing probabilities, and mistakenly assume that the two events are unrelated, you would say that both USC and Stanford have a 50% chance of having a second loss and a 50% chance of winning the Pac-12 championship. Thus, by this mistaken logic, the Pac-12 champion has a 50% chance of having two losses.
BC
Re: Bayesian 101 - winflop - 10-05-2015
It's a bit more sophisticated than that BC. It's a Monte Carlo simulation based on their FPI.
I really think there are only four teams truly capable of winning the conference title at this point: Utah, Stanford, U$C, and UCLA. Stanford could slip once more and still win the North, only because the division is that weak (I doubt any other team will be better than 6-3 and only Oregon & Cal will finish over .500). But I could see a 7-2 team winning the South and then beating Stanford in the CCG. I'm not sure I'd go as high as 70%, but that's purely on gut/eye test.
Re: Bayesian 101 - washingtonismoney - 10-05-2015
ESPN's FPI would be shocked if the conference champ is a team other than SC or Stanford -- gives those teams a combined approximate ~84% chance to win the conference.