PDA

View Full Version : ESPN's prediction


thunderkyss
10-09-2010, 08:46 PM
Anyone ever see these (http://espn.go.com/blog/afcsouth/post/_/id/16263/prediction-city-results-before-kickoff)? Pretty math intensive, and by simulation I wonder if they mean Madden.

HTown2ATX
10-09-2010, 08:51 PM
I doubt they mean Madden cause they had in letters at the bottom that the Texans were favored after 10,000 simulations.

Even if you set the timer to 1 minute per quarter, that's a hell of a lot of time that would take up.

Probably some algorythm they have.

SrslySirius
10-09-2010, 09:18 PM
It is fashionable these days to dismiss mathematics and the scientific method, especially in matters that are seemingly objective; like sports. I could tell you that the probability of rolling a 4 on a 6-sided die is 16.6%, and that the chance of rolling anything else is 73.4%. Then I could roll a 4 six times in a row. The skeptic would say "Aha! So much for your junk science."

The skeptic would be right (sort of), but for the wrong reasons. The issue at hand here is sample size. Rolling the die 6 times isn't enough to prove anything. But if we roll that die a few thousand times, then the percentages should adhere to their expected values.

I believe it is possible to make accurate predictions in football with complete information. I'm not sure if ESPN is doing so accurately, but let's assume for the moment that they are. The fact still remains that this game only gets played one time. Any statistical projections are meaningless, just as they would be for trying to predict a die roll. I don't care that the Houston Texans are a 59% favorite over the New York Giants unless we get to play at least a few hundred games against them.

b0ng
10-09-2010, 09:28 PM
They are using the What If Sports simulator to run these things. You absolutely cannot predict with any sort of certainty how a ball is going to bounce.

The Pencil Neck
10-09-2010, 09:31 PM
The whole Accuscore thing depends on the model that you're running it against. And the model you're running it against is based on what you expect to happen. So... your results will usually tell you what the person who set it up expected to happen.

So...

It's BS.

Yankee_In_TX
10-09-2010, 09:33 PM
Too many variables, period.

Jackie Chiles
10-10-2010, 01:08 AM
It would be interesting if there was a way to check their accuracy, ie if the teams they favor 60 to 40 actually win close to 60 percent of the time.

midway
10-10-2010, 01:49 AM
I could tell you that the probability of rolling a 4 on a 6-sided die is 16.6%, and that the chance of rolling anything else is 73.4%.

What about the 10% when you roll neither a 4, nor something that isn't a 4 ?

(sorry, couldn't help myself)

Txn_in_Oki
10-10-2010, 02:23 AM
They made a prediction?!?!!?

I saw Jenny Dell and lost my focus.

The Pencil Neck
10-10-2010, 02:39 AM
It would be interesting if there was a way to check their accuracy, ie if the teams they favor 60 to 40 actually win close to 60 percent of the time.

Prior to the season starting, they forecast us as 4th in the AFC South with 6 wins, iirc. I think they gave us a 0.0% chance to get to the playoffs. I remember the 6 wins because that was what I think Football Outsiders predicted as well.

So... let's get to the playoffs and prove these guys wrong.

(Accuscore had the Colts beating us 27-23 and gave the Colts a 63% chance to win the game.)

YoungTexanFan
10-10-2010, 06:28 AM
It is fashionable these days to dismiss mathematics and the scientific method, especially in matters that are seemingly objective; like sports. I could tell you that the probability of rolling a 4 on a 6-sided die is 16.6%, and that the chance of rolling anything else is 73.4%. Then I could roll a 4 six times in a row. The skeptic would say "Aha! So much for your junk science."

The skeptic would be right (sort of), but for the wrong reasons. The issue at hand here is sample size. Rolling the die 6 times isn't enough to prove anything. But if we roll that die a few thousand times, then the percentages should adhere to their expected values.

I believe it is possible to make accurate predictions in football with complete information. I'm not sure if ESPN is doing so accurately, but let's assume for the moment that they are. The fact still remains that this game only gets played one time. Any statistical projections are meaningless, just as they would be for trying to predict a die roll. I don't care that the Houston Texans are a 59% favorite over the New York Giants unless we get to play at least a few hundred games against them.

There has to be an acknowledged standard deviation, noted in some fashion. We can assume for our purposes that it is sigma. As it was mentioned in a previous post, we have to take into account expected value, which I believe was mentioned at 6 wins. Now, 6 is a constant number that can't be quantified into a percentage without context, but say we take the 6 out of 16 to get 37.5% wins. Now, that number is meaningless as you mention but for a different reason. For the sole purpose of this weeks game vs. the Giants, it doesn't provide a baseline. It is, itself, just a number. The total expected value of 6 is given, but no other factor is given. In math/statistical terms, we are trying to find the area under a normal slope between 6 and whatever the Giant's win total is projected as. That number, or percentage, should be OUR expected percentage before addition of the prior established "left of 6" when speaking in terms of the normal graph. You can divide said fraction by 1, giving you Mu. With Mu and Sigma figured out, you can use either the binomial method or the exponential (harder) method to predict accurately (as possible) the potential outcome based on the given info.

axman40
10-10-2010, 08:49 AM
Prior to the season starting, they forecast us as 4th in the AFC South with 6 wins, iirc. I think they gave us a 0.0% chance to get to the playoffs. I remember the 6 wins because that was what I think Football Outsiders predicted as well.

So... let's get to the playoffs and prove these guys wrong.

(Accuscore had the Colts beating us 27-23 and gave the Colts a 63% chance to win the game.)
Yes but that was before they knew Foster is Texan for victory!
:texflag:

SrslySirius
10-10-2010, 11:13 AM
What about the 10% when you roll neither a 4, nor something that isn't a 4 ?

(sorry, couldn't help myself)

No worries. I was completely hammered when I wrote that last night. Haha.

Big Lou
10-10-2010, 03:27 PM
Well there is a Software Engineer somewhere that sucks as bad as the Texans do.......