Win probabilities for week 5 NFL games are listed below. The probabilities are based on an efficiency win model explained here and here with some modifications. The model considers offensive and defensive efficiency stats including running, passing, sacks, turnover rates, and penalty rates. Team stats are adjusted for previous opponent strength.
Last week the predictions were 10-3. But if you are a new reader, don't expect the model to be that accurate every week. On average it will be between 70-75% correct, and some weeks it can go 6-8. Over the past 2 seasons however, it's been the most accurate prediction system I can find, but only slightly more accurate at picking winners than the spread.P WIN GAME P WIN 0.15 
0.85 0.79 
0.21 0.53 
0.47 0.68 
0.32 0.15 
0.85 0.56 
0.44 0.59 
0.41 0.54 
0.46 0.31 
0.69 0.42 
0.58 0.28 
0.72 0.12 
0.88 0.47 
0.53 0.30 
0.70
Week 5 Game Probabilities
published on 10/01/2008
Subscribe to:
Post Comments (Atom)
have you ever tried to include a situational variable? i do not know how you would quantify it, but for instance: last week denver was coming off of two hard fought, shoot out victories playing against kc who was winless on the season.
I honestly dont understand why NE is a fav going to SF anyway. Same with the AZ-Buffalo game as a coin flip. Time will tell but teams heading two time zones away dont usually do too well.
Actually, Football Outsiders did a study on this in their most recent annual and found that the time zone change doesn't really matter. I am going off the top of my head here, but I'm pretty sure it was for all time zone travels. Feel free to correct me if I'm wrong.
Josh I have no idea if that is true. Although there is probably no real difference since playing on he road anywhere is always tougher.
Anonymous, that's all well and good, but said variable would have to be completely subjective. And what's a "hard fought game", exactly? What if you're coming off one? Or three?
There's just no way to accurately quantify such a factor.
Edward, I understand. I guess it would have to come down to what research showed about one 'hard fought' game as opposed to two or three etc.
What is a 'hard fought' game? I don't know exactly. I feel as though research may be able to quantify the variable to some extent. I was just throwing the question out there to see if the thought had crossed your mind(s).
Okay, so I'm not statistics whiz, but I get the general idea of the basics.
However, I'm having a hard time converting the regression model into a probability. If you plug in the efficiency ratings for week 4 into the regression calculation (multiply the values by the coefficients and add the intercept, you get a result that should basically be a value between 0 and 1 (AWin).
I'm coming up with a variety of results, and how do you get it into the Pwin final value of probability here?
I absolutely love this site and think all of this analysis is awesome, and would love to understand this basic calc more.
Thanks!
Aaron,
The model is a logit model. When you multiply the values by the coefficients, you take that answer and put it into this:
1 / (1 + e^-z)
where z = the results you got from multiplying the values and coefficients. I hope that clears it up for you.
also, the 11-2 record for last week is incorrect. cincinatti was a loss (even though palmer did not play), denver was a loss, and arizona was a loss.
Right. How did I miss that? Fixed.
Does your model factor in the fact that Rodgers is injured? If not, I'm surprised that Atlanta is favored.
Thanks Josh! I get it now...sorta. Now it's time to dust off those old stats books...
Much appreciated!
Once again a nice record for the week (considering the upsets). You matched the winning record in our pool of 38 people (as you also did last week). In my wisdom, I thought your model was wrong on GB and BUF. Shows what I know...
After looking at your results from last year, I decided to follow the picks that you and I like and I'm winning a good amount of money so keep up the good work and don't change a thing. Its up to the people to decipher how much and injury or momentum factors in to your probabilities