Weekly game probabilities are available now at the nytimes.com Fifth Down. The lead-in is a re-purposed version of my discussion of intangibles. The only upset the model likes this week is PIT over NE.
Weekly Game Probabilities
published on 10/27/2011
Subscribe to:
Post Comments (Atom)
I think you are knocking on the wrong doors by listing "upsets the model likes".
if the bettings lines have Team A as a 53% favorite and your model has them at 47%, that's a difference, sure, and it would be an "upset".
but it's not nearly as big or significant a difference as Team A being a 53% favorite per betting lines and your model having them as 70% to win.
Did you remember that the WAS-BUF game is being played in Toronto?
Maybe I missed a prior post or maybe Brian or a reader can explain something. It is clear that the GWPs are substantially more regressed this year. No team this year has a GWP of over 67% whereas last year there were four teams with GWPs over 70% going into week 8. Similarly, this year the teams at the bottom all have GWPs between 36-40%; last year there were four teams between 23-26% going into week 8. Either there is rampant parity this year in the NFL (seems unlikely) or the formula has been tweaked, appartently substantially. Can anyone explain why this change was made this year? Jason
Aaron-Yes. Thx.
Jason-The regression of the input variables are a little heavier this season. There may be slightly greater parity as well. Once we get past week 8, the regression will gradually fade out, as we become more certain of various team qualities.
So the Skins-Bills game is in Toronto this week, but it's nominally a home game for the Bills.
Is this in the model, or is it one of those things like injuries etc. deliberately not modelled?
Brian - In that case, it piques my curiosity that while Buffalo and Washington both have GWPs of 0.53, the game is not projected as a coinflip on what you are considering neutral turf. Since your model does not consider injuries nor the fact that Toronto is very close to Buffalo and thus the game is probably somewhere in between a home game at Ralph Wilson and a game on neutral turf, I wonder where the 62% projection for Buffalo comes from. The only possible explanation I can think of is that Buffalo's strengths and weaknesses match up well against Washington's. Can you give any additional insight here?
I would not have expected Denver's WP to be as good as it is hosting Detroit. Same for St. Louis & Seattle.
Re WAS-BUF I'm leaving the HFA in. I think you're right that in reality HFA won't be as strong as it usually is for true home games, but I suspect BUF will have a greater comfort level with the environment.
And remember, when teams close in strength play each other, HFA has a bigger effect on the outcome.
"I think you are knocking on the wrong doors by listing "upsets the model likes".
if the bettings lines have Team A as a 53% favorite and your model has them at 47%, that's a difference, sure, and it would be an "upset".
but it's not nearly as big or significant a difference as Team A being a 53% favorite per betting lines and your model having them as 70% to win.
for instance this week, where your model has
New Orleans as a 59% favorite
and the moneyline has them as a 90% favorite
that's a gigantic difference, I think that's the stuff you should be paying attention to for a tracking basis, not "upsets"
I guess we found out who was right about that "gigantic" difference!
Another winning week for the model. It's on a roll.
Model was very wrong on Detroit-Denver
So any ideas why the model was so right on Pitts beating NE and on NO vs Rams? Curious to see how it predicted those so well.