0 Members and 1 Guest are viewing this topic.

I'm fine with any model that tells me the Cavs have only an 11% chance of winning the title.

When Chris Paul has been healthy and playing full strength, Houston has been the best team in the league all season. It is not surprising that a model based on past performance would have them as the best team.It is also not surprising that the Vegas odds favor the Warriors. Remember that Vegas odds reflect how they think people will bet, not who they think will win.Should be a great series. I think Houston can beat them although I would not put it at 80% likely. Perhaps more like 55% likely.

This prediction business is fickle. Nate Silver went from a star to a scrub in one night (presidential election). I think he's somewhere in between- a good data analyst that will often get things wrong, like other people.

Hopefully, the model is constantly being tweaked to try to gain accuracy. Also, it's not an all or nothing; it is a helpful tool to ask "why" and look for reasons to agree or disagree with the model. Making a wrong prediction doesn't necessarily invalidate the entire model. If a statistical model told you that if you flipped a coin 10 times you should get 5 heads and 5 tails, but then you actually flipped the coin and got 7 heads and 3 tails...would you throw out the model?

Quote from: Fan from VT on May 09, 2018, 07:47:56 PMHopefully, the model is constantly being tweaked to try to gain accuracy. Also, it's not an all or nothing; it is a helpful tool to ask "why" and look for reasons to agree or disagree with the model. Making a wrong prediction doesn't necessarily invalidate the entire model. If a statistical model told you that if you flipped a coin 10 times you should get 5 heads and 5 tails, but then you actually flipped the coin and got 7 heads and 3 tails...would you throw out the model?100%.The idea that a model needs to "understand" something is like asking a hammer to understand something. It is just a sophisticated and useful tool. You as a person looking at the model need to use it appropriately.Given that its a relatively straightforward ELO system with adjustment for opponent, rest, etc just means its only a little different than point differental, SRS, etc.The Rockets are really good and while I do think the Warriors are the better team overall they are now looking like they might pull off the upset.

Quote from: Fafnir on May 23, 2018, 09:50:39 AMQuote from: Fan from VT on May 09, 2018, 07:47:56 PMHopefully, the model is constantly being tweaked to try to gain accuracy. Also, it's not an all or nothing; it is a helpful tool to ask "why" and look for reasons to agree or disagree with the model. Making a wrong prediction doesn't necessarily invalidate the entire model. If a statistical model told you that if you flipped a coin 10 times you should get 5 heads and 5 tails, but then you actually flipped the coin and got 7 heads and 3 tails...would you throw out the model?100%.The idea that a model needs to "understand" something is like asking a hammer to understand something. It is just a sophisticated and useful tool. You as a person looking at the model need to use it appropriately.Given that its a relatively straightforward ELO system with adjustment for opponent, rest, etc just means its only a little different than point differental, SRS, etc.The Rockets are really good and while I do think the Warriors are the better team overall they are now looking like they might pull off the upset.Right. As an interpreter of the stats, you get to take any prediction model in context. So, the "general consensus" is GSW is the best team, but 538 had Houston favored, 80-20. So, if that runs counter to the eye test, what is this predictive model seeing that we aren't, and/or what are we seeing that the predictive model isn't. Likely, the predictive model is seeing how good Houston's defense secretly is (despite the offense getting the hype), how surprising good Capella is (his impact I think outpaces his budding reputation), and assigns some historically determined value to home-court advantage. Now, it is also missing the fact that the Warriors may have "coasted," or, at the very least, had some very very good currently healthy players miss a lot of time due to rest and injuries. So you can take this as learning something we might have missed about Houston, but also us knowing something vital that the model might not know, and make our own predictions.But as I've mentioned before, just because a model says with 1 dice role, you have a 5/6 chance of not rolling a 1, then you go and roll a 1, does not mean the prediction model was "wrong" or "Broken."

If a statistical model told you that if you flipped a coin 10 times you should get 5 heads and 5 tails, but then you actually flipped the coin and got 7 heads and 3 tails...would you throw out the model?