|
|
Post by Ramen on Oct 25, 2015 14:58:28 GMT
Just stumbled across this website by Nate Silver's 538. It has Atlanta with a very high probability of making the playoffs, but low for going deep. projects.fivethirtyeight.com/2015-nfl-predictions/Carolina is predicted to be the division winner and get the first round bye. Atlanta also has a 70% chance of winning the game today, according to the models. You can click on future weeks to see predictions for all of the NFL teams, and look back over past weeks to see how well the model performed. |
|
|
|
Post by Guvmintcheeze on Oct 25, 2015 16:22:21 GMT
It's batting around just over 50%.
Any kid flipping a coin....
|
|
|
|
Post by JDaveG on Oct 26, 2015 11:51:03 GMT
Yeah, I think metrics work better for some things than others. I don't value them as much for football as baseball, because football has a lot more to do with playing together than it does putting a bunch of pieces on the field.
If your 2nd baseman can play, he'll get his share of great plays. If he can hit, he'll get his share of hits. The only place in baseball this is an issue is if the guy hitting in front of him doesn't do his job he'll see fewer good pitches. But then, metrics can give you some information on that too. In football, it doesn't matter what a guy did last season, if he's missing blocks at LG, it affects the LT and C play, as well as the QB play, the WRs, the defense, etc. One guy can drag a whole team into the sewer in football. It is the ultimate team sport. And I just don't see how statisticians can measure a man's heart or desire, so I place a lot less value on it in this context.
Also, it obviously gets more accurate as it goes along. If you want to see how bad it really is, go look at the model from September 10. The one that has Indanapolis and Philadelphia in the top 10 and Detroit (DETROIT!) at 13.
|
|
|
|
Post by Ramen on Oct 28, 2015 2:46:57 GMT
It's batting around just over 50%. Any kid flipping a coin.... By my count, he correctly predicted 73 out of 98 games, or 74.5% of the games. I'm not counting the two 50/50 game predictions as accurate, and I'd like to see how well the games with close predictions (e.g., 52/48%) went. If you predict a 55% chance that a team will win for 10 games, you would expect your prediction to be correct 5-6 times. If you predict a 95% chance that a team will win for 10 games, you would expect your prediction to be correct 9-10 times. I want to look into that specific question about teams predicted to have close to a 50% chance versus those close to having a 100% chance of winning. But it looks like the model has a pretty good record overall. |
|
