Game probabilities for week 15 are up at the New York Times' Fifth Down.
This week I examine how some outlier teams can break the model, including this season's NE Patriots.
...Malcolm Gladwell’s book “Outliers” popularized the term in his title, but the exceptional people he featured in his best-seller were not actual outliers. If I spend five hours trying to learn to play the violin, I’ll stink. But if someone else spends 10,000 hours practicing violin, we would of course expect that person to be great. That’s entirely predictable and fits our model of practice makes perfect. A true outlier would be someone who spent only a few hours practicing, yet was still a violin prodigy.
Sometimes teams come along and break my model, and the 2012 Patriots are one of those teams...
Hello
Is Buffalo-Seattle calculated as a neutral field game?
No. Is that in Toronto this week?
Being willing to do anything for 10,000 hours to perfect it is an outlier temperament.
Why softening your stance Brian:
You need a more detailed analysis to substantiate your claims here. (larger sample size). We know NE cheated in the past. We know people only cheat when the reward is equal or greater than the risk. We can assume NE benefited in the past from cheating. Unless proven otherwise the most logical position is they are still doing so. It's easy to pick off opp. passes when you know their plays!
Can't we somehow use your expected play analysis to see
if NE how much of their over-performing is the result of success on high leverage plays?
Finally;
1. I mentioned weeks ago the importance of 'score effects' on the model. Basically NE gets ahead so other teams take more risks, exactly what we see in other sports (i.e. ice hockey).
2. If this is accurate THEN this means Carolina is even higher rated given their high OPASS???
Ouch?!
To the comment two above this one: seriously - cheating? I thought this was a site for more intelligent football analysis.
The cheating comment was interesting... that is a potentially legit way to explain an outlier.
By the way, the model called the Carolina upset over Atlanta last wee... wow.
Another legit way would be that NE plays nearer to the optimal strategy mix, so the same base payoffs of passing and running on defense turn into higher combined payoffs.
Another point would be, that NE has a scheme which is strongly relying on taylored gameplans. This tend to work better, when there is more film from which to derive such a film plan, and more familarity of the defense/offense with the differing concepts. This would lead to a steady increase in GWP over the season, which would not be adeaquatly captured by Brians "Whole Season" Algorithm.
> It's easy to pick off opp. passes when you know their plays!
On average, there are about 2 interceptions per game. We're in week 14, so the 15 interceptions the NE defense has is roughly normal - the turnover differential advantage is largely due to an abnormally low number of thrown interceptions.
I don't see people accusing the Bears with their 21 picks as cheaters.
> We know people only cheat when the reward is equal or greater than the risk.
That's the same as people only making +EV bets? The reasons that people do, or don't, cheat are much more complex than that.
Brian, in the 5th down column, you write:
"The Patriots’ defensive interception rate is among the best in the league at 3.1 percent."
That's 3.1 interceptions per 100 passes thrown?
"[The correlation between passing efficiency and defensive interception rate] turned out to be very weak, only 0.06 on a scale from zero to one (one being perfect correlation)."
Do you mean the r-squared?
The rate of interceptions as a function of score difference looks reasonably linear to me -- something like:
Chance of interception per pass = 2.7 % + 0.02% per point behind
- Is Buffalo-Seattle calculated as a neutral field game?
- No. Is that in Toronto this week?
Yes.
Brian - did you look at the rank correlation of offensive efficiency v defensive interception rate? That's usually a good way to test for non-linear correlation
(previous comment deleted for typos)
Do the folks who think that this is due to cheating realize that offensive plays are called into the QB...
Regarding the passing offense and defensive interceptions, and since when is "almost one standard deviation" considered evidence of a relationship? it is the 68% confidence level. I.E. entirely consistent with it being a normally distributed random variable.
Additionally, the bottom 10 being equal to the league average, implies that the middle 12 are below average (someone has to be below average don't they??).
Brian, combining this with the 0.06 correlation, i think you have convincingly proved that there is no relationship between passing offense and defensive interception rate.
Hey, wait a minute.
The hypothesis of the article is that the model undervalues the Pats, because the Pats consistently excel at interceptions.
However, the article then proves that the Pats get interceptions because they have a large lead. Therefore, the interceptions are not causing the wins, they are a result of the wins. (similar to the 'running the ball correlates with wins" argument.)
Therefore, the interceptions are not causing the wins.
Therefore, the hypothesis is incorrect.
Kulko;
Interesting points. However, the key point is NE is the ONLY team able to consistently outperform.
Surely other good coaching staff are doing what BB is . Granted maybe not quite as well. Shouldn't we see a few other teams with able to outperform. But alas we don't. My research is similar to what we find this performances by athletes on steroids - Clear numbers way beyond expected and removed from the group."Like they say fool us once....."
Is Brian now steering us away from potential evidence that NE are still cheating? After bravely making strong points to the contrary 4 years agao on this site? Hmmm..
I don't understand what this mystery about NE is, or why they are an "outlier". They are averaging 36 points a game. They have scored 56 touchdowns (over 4 per game). Their offense is amazing, of course they win games.
Their interceptions are average, 15, which is tied with many other teams in the range of 6th to 12th in the nfl.
At first I thought the model got this game right, then an article explaining why the Pats are a outlier and to take Pats.
The model got it right, the 49ers are the right play based on a huge regression to the mean for the Pats.
Interesting isn't it ?
He writes an article explaining why the Pats are an outlinier and then they are not.
He way, way, way over-reacted to last weeks big win by the Pats, he had no clue how to interpit that game, I did and because of it won.
Mike M sounds very proud of himself. If my analysis over-reacted to anything it was NE's +198 pt differential, and that their 3 losses were by a grand total of 4 points. I also left the game estimate as is, which turned out to have been the right call.