Indianapolis, New Orleans, and San Diego are still the big three headed into the divisional round games. My system is not very high on Minnesota and Arizona, and it could very well be wrong about those teams. But it's worth looking at why they don't fare well in the projections.
Let's look at the efficiency stats as of week 16 for the teams still alive. (The projections below include performance from the wildcard games but exclude week 17.) We'll see why the projections shake out the way they do.
TEAM | OPASS | ORUN | OINT% | OFUM% | DPASS | DRUN | DINT% | PENRATE |
ARI | 6.6 | 4.1 | 2.7 | 1.5 | 5.8 | 4.6 | 3.5 | 0.42 |
BAL | 6.4 | 4.5 | 2.6 | 0.8 | 5.9 | 3.5 | 4.3 | 0.56 |
DAL | 7.3 | 4.8 | 1.6 | 0.8 | 6.0 | 4.0 | 2.1 | 0.43 |
IND | 7.6 | 3.6 | 3.0 | 0.4 | 5.5 | 4.2 | 2.9 | 0.27 |
MIN | 6.9 | 4.1 | 1.4 | 0.9 | 6.1 | 3.9 | 2.0 | 0.38 |
NO | 7.9 | 4.6 | 2.1 | 0.7 | 6.2 | 4.5 | 4.7 | 0.39 |
NYJ | 5.7 | 4.5 | 5.6 | 1.2 | 4.8 | 3.7 | 3.3 | 0.35 |
SD | 8.1 | 3.3 | 1.9 | 0.3 | 5.8 | 4.5 | 2.8 | 0.30 |
NFL Avg | 6.2 | 4.2 | 3.1 | 1.0 | 6.2 | 4.2 | 3.1 | 0.39 |
First, let's look at Arizona. They face two road games on the way to the Super Bowl, which hurts their overall chances right off the bat. Second, while they are known for their potent offense, they turn the ball over more frequently than any other NFC contender. But their biggest problem is that their own offense pales in comparison to the phenomenal Saints' offense. The Cardinals' defense is only slightly better than average on net. Lastly, the Cardinals faced the 3rd easiest schedule this season, enjoying two games against the rest of the woeful NFC West.
Minnesota is much the same story. They've had the 2nd easiest schedule in 2009, but still failed to put up any impressive numbers. Despite the hype around Adrian Peterson, their running game is very average. Minnesota has been winning because of two things: smart passing and luck. Favre has protected the ball very well this year, bouncing back from a terrible 2008. His interception rate is among the lowest in the NFL, and he's still generating above average passing efficiency. Minnesota has benefited from some timely miscues by opponents--missed field goals and penalties at critical times that have saved 2 or 3 wins for them.
With that, here are your playoff probabilities going into the divisional round games.
% Probability
AFC | Div Round | Conf Champ | SB Champ |
IND | 76 | 48 | 28 |
SD | 72 | 34 | 19 |
NYJ | 28 | 10 | 4 |
BAL | 24 | 7 | 3 |
% Probability
NFC | Div Round | Conf Champ | SB Champ |
NO | 82 | 58 | 30 |
MIN | 43 | 13 | 4 |
DAL | 57 | 24 | 11 |
ARI | 18 | 5 | 1 |
Don't know where to go with this. Is there an area for general comments?
Reason I ask is because how difficult would it be to take the data from Pro-Football-Reference and trace team efficiencies back to 1980??? Defferentiating fumbles may be a problem.
Would coefficients need tweaking over different eras?
When you say the Vikings had the 3rd easiest schedule did you find that by their opponents' record in '09 or their opponents' GWP? If it's record, did you take out the impact the Vikings had on that record (ie Packers were 11-3 against everyone else)?
James. I believe it's opponent GWP.
Correct-GWP.
Jero-It's not too difficult, just a little time consuming. I spent a good chunk of my Wednesday evening doing just that, and only made back to '98. Yes, the eras are going to be different, so there is a trade off in going back too far in exchange for a larger data set. I had originally stopped at 2002 because that's when the current playoff format started, and the original goal of the research had something to do with home field advantage (if I remember correctly). I'd like to take this further and see if I can't build a reasonably solid regression model around the playoffs and compare it to the regular season.
James, the Vikings actually had the second easiest schedule. And that includes your adjustments.
http://www.cbssports.com/mcc/messages/thread/19365276
Brian, maybe you could look at that list too and investigate the correlation between SoS and playoff chances a little.
How do the Colts have a 78% chance of winning while the Ravens have a 24%. That doesn't make sense.
Sorry. The whole line for IND was slightly off. It's 76, not 78%. Fixed now. Thanks.
Where did you get the updated strength of schedule for 2009? I've looked all over the net for it but have come up empty.
Interesting that Balt and NY combined have a greater chance than Minn and AZ combined to win the Super Bowl.
When a team has the edge in both Offensive and Defensive Yards per Rush in a Divisional, these teams are 20-9 ATS (69%) since '93, and 17-0 ATS (100%) over the last 11 seasons!
Wow. What about both passing, YPA and def YPA?
I don't think these calculations account nearly enough for the variance in the underlying statistics, which is why you have the fairly ridiculous skew effect going in the individual probabilities.
Anonymous: the fact that Dallas has almost the same chance of winning the Superbowl as Minnesota, Arizona, Baltimore, and New York combined should be a pretty solid indicator that the model is seriously flawed.
Interesting stat, Anon. Nice time to point it out, too, because that 11-year streak will break if Indy, San Diego, or Arizona win this weekend.
Ben-Are you aware of how accurate this model is?
Do you have confidence intervals for your probability estimates? I think that would be interesting.
By what metric are you measuring its accuracy? I know that the results it produces are wildly different from what the betting markets suggest. I generally don't think those kinds of overlays exist, and if you think they do, you're probably underestimating the analytical skills of the market-moving sports-bettors.
On top of that, without seeing any results, I can directly observe an obvious flaw in the model, in that it doesn't sufficiently account for variance in the underlying statistics. The result you would expect from this error is overstating the favored team's chances, with the error becoming greater and greater the more of a "mismatch" the game is. Lo and behold, that's exactly the pattern I see in these predictions, as compared to the betting markets and common sense.
Ben-You're mistaken. The logistic regression model underlying the projected probabilities does account for variance in team performance.
And actually, the individual game predictions are slightly, but consistently, more accurate (and better calibrated) than betting markets.
The error in the betting markets for eventual Super Bowl champion is caused by the conjunction fallacy. People (bettors) consistently overestimate the probabilities of serial events. For example, most people think that the odds of two 0.40 serial events is not 0.16 but closer to 0.20. Not everyone, but enough of them do.
The odds makers have to reconcile these inefficiencies and produce odds that balances wagers and includes 'the vig.' They often don't accurately reflect their best estimates of the actual outcome.
It's not that I think you can't beat the betting markets, it's that I don't think you can beat them by that much. And again, I think you underestimate the analytical capabilities of the market-moving sports bettors, who have done some of the best statistical analysis I have ever seen. Also, your claim that odds makers have to set lines that reconcile the natural "inefficiencies" of sport-bettors is simplistic. Whatever irrational tendencies the bulk of sports-bettors have in particular matchups is largely offset by the fact that the market players will put up more money in those situations to take advantage of them.
Also, I've looked at the details of your model and stick by what I said, that I don't believe it *sufficiently* accounts for variance.
"I've looked at the details of your model..." That's nonsense. This model is more accurate than Vegas, despite being ignorant of things like injuries to starting QBs, plus it's almost perfectly calibrated. Teams given a 0.60 probability to win do win 60% of the time, teams given a 0.70 probability win 70%, etc.
Move along.
Go Burke!
Hey could it be possible to allow comments on the WP pages for different games? I always have them on when watching NFL games and it would be fun to have a little discussion forum for smart, stat loving football fans!
I said that I believe you can beet the market writ large, but that when you see massive overlays relative to the market -- especially in a predictable pattern -- you should be skeptical.
I would be interested in seeing your calibration data, especially as compared to an identical calibration analysis of high-volume sports market "predictions" -- especially for games with considerable favorites, where I think your model overstates the chances of the better team.
And I'm sorry if I sounded so contentious that you have commentors "rooting" for you -- perhaps exaggerated further by the fact that I haven't commented here before -- but I assure you I am not a troll and I take these things very seriously.
People have a hard enough time distinguishing between 15-25% and 1% from where I sit.
Look at the NYG-NE Superbowl. It was what? 4-6 to 1, somewhere in there. I thought it was on the lower end of that range so I took the NYG. Then when the NYG win people pretty much all responded as though the NYG had a 1% chance. I realize that may have been because that IS where the suckers would have/did bet, and the professionals were holding the number up to reality. But it is really striking how far off most peoples subjective guesses are from the actual statistics.
Im very curious to see the increase of performance at the Vikings side, especially after the touchdowns of Brett Favre against the Dallas Cowboys.
www.sawfer.com
You say Minnesota won 2-3 games more than they should because of luck. I think your system says they aren't as good as their record so you search for things that you consider lucky.
Balt misses a FG to beat the Vikings - but in the same game Minnesota is stopped twice inside the 10 in the 4th quarter - was that all skill by Balt?
SF - late pass by Favre wins...SF blocks FG returns it for a TD to make it 14-13 instead of 13-7 or 16-7 - which play is luckier?
Pitt - Steelers get two TD's on int's and sack fumble...is that lucky?
Chic - Vikings miss a convert lose in OT - is that lucky for Minn or Chic
Minnesota beats two NFC East teams by 78-10 combined - two teams you say are better than Minnesota.
NYJ beat SD on the road - yet you have SD as the 2nd best team
Your system overrates the NFC East and AFC West teams. Those teams consistently got hammered when they played other decent teams (GB, Minn, NO, Arz, Balt, Pitt, NE, Indy)
Combined the NFC East against those teams mentioned and SD and Den - 4-12 negative 150+ pts
SD-Den against the teams mentioned 1-6 negative 68pts
I'm impressed how hard you smart folks work at these things - but some things aren't as complicated as some would like to make them.
I did give MIN about a 40% chance to beat the Cowboys. It took the Cowboys playing their worst, maybe 2nd worst, game of the year to hand MIN that win. If they played 10 games, I'd bet DAL wins 6 of 'em.
MIN had a very weak schedule this year, by any measure. They're good but overrated, and I'll stand by that. However, that doesn't mean they can't win the championship. Just look at the '07 Giants.
How often does a team you think will win 60% of time lose by 31?
Hand Minnesota the win. That kind of stuff happens. Good teams lose games by turning it over a ton...a bunch of fluke plays (take Arz losing to SF this year).
Weaker teams don't sack the QB 6 times and have 13 stops for negative yardage. Doesn't happen.
Weak schedule argument:
The Vikings did play a bunch of stiffs - for the most part they pounded those teams with the exception of the Bears on the road and SF at home.
But in 9 games against decent competition (GB, GB, Arz, Car, Dall, Balt, Pitt, NYG, Cinn) they were 7-2 plus 67 pts.
I am skeptical of Burke's model for my own reasons, and I admit I was not unhappy to see Minnesota and New York win and Dallas lose, but some of the criticisms here are off base.
It's not like Burke is guessing based on his subjective opinions of the various factors. It's a regression analysis, and it's perfectly reasonable to think it is flawed for not not being detailed enough (e.g., taking strength of schedule on a game by game basis) or for weighting some things too heavily (e.g., giving teams too much credit for having run well statistically). But what you can't do is look at this result or that result and say "but there was a blocked field goal, Minnesota just got unlucky against so-and-so" or "look, Dallas lost by 31, they couldn't have been the better team." That's like criticizing Google for failing to list the top fan-site in the top 5 results on a search for "Chuck."
Maybe you know better than the model in individual cases, or maybe you even have ideas about how the model can be improved, but to criticize a model with a solid track record b/c of individual factors you think it missed or b/c of individual results it failed to predict misses the point.
Uh ... where are the Falcons and the Patriots?
Uh...You're looking at a post from the 2009 season.