The truth about the Patriots' fumble rate
- Thread starter singaporesoxfan
- Start date
Would love to study this but boy is ST fumble data hard to come by. My best shot appears to be to extrapolate offensive fumble data from Burke's article and subtract that from total team fumble data (and subtract the rare cases of defensive fumbles after turnovers).Eddie Jurak said:
Why does it matter if the Pats are outliers on fumble frequency? They are outliers over the past few years on points scored and wins too, and we attribute that to coaching. I think they are a true outlier and we should celebrate that because they are well coached. How many fumbles the Pats have is the wrong question to ask. It tells you nothing about the ball because it's completely confounded with coaching.singaporesoxfan said:
The better question, which is intended to partially control for coaching and team and player effects (which is the question you addressed), is whether the same player, when they switch teams, has his fumble rate go up. Player is the same, ball is different. (There are still problems here, but to first order let's ignore them: e.g. teams don't release guys unless they are playing poorly, so the average quality of players that used to play for the Pats is lower than those currently on the team).
Your response shows why the answer given to this question is invalid. Removing Brandon Tate destroys the effect. Therefore, it is too dependent on small sample sizes - too cherry picked - and is likely statistically insignificant if you account for multiple comparisons.
Done. Fumble rates are uninformative to address this question. Finis. Nice rebuttal.
http://www.shoutingloudly.com/2015/01/27/deflategate/
Here's an article by a grad student, Bill Herman, analyzing both the Sharp analysis (that players fumble more when they leave the Patriots), and the WSJ analysis (that players fumble more when they are not on the Pats than when they are, looking at 2010-2014 numbers). Difference in players looked at by the WSJ includes Danny Amendola and Brandon Lafell, and it includes all the time before joining the Pats for Blount. The author dismisses the Sharp analysis for much the same reasons as I did, and then concludes that "The Journal data supports the claim of cheating, even if it does not show such an unlikely outcome that cheating is the only sensible explanation."
Sigh. Here's the key assumption that
Which is classic "begging the question": assuming players have natural fumble rates means you assume no coaching of ball security is possible, then when you find that a team has better ball security, you conclude nefariousness.
The other problem of course, is fumble data is so rare that one player can skew the entire data set. In the WSJ case, the entire difference in fumble rates between players on and off the Pats is that LGBT fumbled a shit-ton in his first two rookie years at TB (10 fumbles!)
I'd like to rebut his "statistical significance" nonsense. Can someone point me to a good link to an explainer on the misuse of significance?
Here's an article by a grad student, Bill Herman, analyzing both the Sharp analysis (that players fumble more when they leave the Patriots), and the WSJ analysis (that players fumble more when they are not on the Pats than when they are, looking at 2010-2014 numbers). Difference in players looked at by the WSJ includes Danny Amendola and Brandon Lafell, and it includes all the time before joining the Pats for Blount. The author dismisses the Sharp analysis for much the same reasons as I did, and then concludes that "The Journal data supports the claim of cheating, even if it does not show such an unlikely outcome that cheating is the only sensible explanation."
Sigh. Here's the key assumption that
Which is classic "begging the question": assuming players have natural fumble rates means you assume no coaching of ball security is possible, then when you find that a team has better ball security, you conclude nefariousness.
The other problem of course, is fumble data is so rare that one player can skew the entire data set. In the WSJ case, the entire difference in fumble rates between players on and off the Pats is that LGBT fumbled a shit-ton in his first two rookie years at TB (10 fumbles!)
I'd like to rebut his "statistical significance" nonsense. Can someone point me to a good link to an explainer on the misuse of significance?
DrewDawg said:
To provide a friendly counterpoint to this, in four years with New England he had approximately the same amount of carries as he did in two years with CIN. In those four years with NYE he fumbled zero times, with CIN he has fumbled five times.
Gambler7 said:
Yes he did. College football stat sites are bad at showing it, but he did.
http://espn.go.com/ncf/playbyplay?gameId=272510145&period=0
BJGE fumbled with the Patriots during a preseason game.
https://www.youtube.com/watch?v=JueRWjZ-Hms
Why isn't run blocking a factor? Football Outsiders ranked the Patriots #1 and #2 in 2010 and 2011, respectively, and BJGE went to a team with a markedly worse run-blocking offensive line (ranked #11 in both years in CIN). I can't see all of his fumbles online, but the first one I see on Youtube is him getting blown up behind the line of scrimmage. Anecdotal, yes, but I'm not sure how an under-inflated ball would help in that situation.
I apologize if it's listed somewhere, but if they are comparing players with and without the Patriots, why isn't it compared to how other teams' players perform when leaving their teams as well? Are the Patriots clearly outliers there?
Though why look at only 2010-2014? Why wouldn't you go back to 2006, when the rules change allowed teams to supply their own balls? Well, simply because it would make the stats look worse for the WSJ I would guess. Adding one more year to the sample size adds Ben Watson to the cohort (who you did not leave out, and not coincidentally your totals were far from extreme as WSJ), which would hurt the "analysis" quite a bit. He fumbled 6 times against 167 receptions for the Patriots, and just 1 time against 193 receptions for other teams. That brings the WSJ number of "w/Pats" down from 185.3 to 117.8, and raises the "other teams" up from 77.3 to 82.3. In that scenario, one extra fumble every 271 touches would result in an identical rate to that group's with the Patriots. That's a bit over one a year, which could easily be attributed to an organizational emphasis on ball protection.
Going back to 2008 allows you to add David Thomas to the cohort. This brings the w/Pats number from 117.8 to 119.3, and "other team" total from 82.3 to 85.9. Hell, simply adding Julian Edelman to the group will bring the w/Pats touches total to 1936, quite close to the "other team" total of 1975 (yes, I'm aware he's only been on the Patriots), and suddenly the "with Patriots" cohort fumbles once every 62.5 touches, while the "other team" cohort once every 85.9 touches. Yes, I'm moving the goalposts here, but that's the point: choosing 2010 as a starting date seems arbitrary, gives a smaller sample, and appears to have been done so to screen out Ben Watson, who makes the shocking "185.3 touches" figure look a far more reasonable 117.8.
https://www.youtube.com/watch?v=JueRWjZ-Hms
Why isn't run blocking a factor? Football Outsiders ranked the Patriots #1 and #2 in 2010 and 2011, respectively, and BJGE went to a team with a markedly worse run-blocking offensive line (ranked #11 in both years in CIN). I can't see all of his fumbles online, but the first one I see on Youtube is him getting blown up behind the line of scrimmage. Anecdotal, yes, but I'm not sure how an under-inflated ball would help in that situation.
I apologize if it's listed somewhere, but if they are comparing players with and without the Patriots, why isn't it compared to how other teams' players perform when leaving their teams as well? Are the Patriots clearly outliers there?
Though why look at only 2010-2014? Why wouldn't you go back to 2006, when the rules change allowed teams to supply their own balls? Well, simply because it would make the stats look worse for the WSJ I would guess. Adding one more year to the sample size adds Ben Watson to the cohort (who you did not leave out, and not coincidentally your totals were far from extreme as WSJ), which would hurt the "analysis" quite a bit. He fumbled 6 times against 167 receptions for the Patriots, and just 1 time against 193 receptions for other teams. That brings the WSJ number of "w/Pats" down from 185.3 to 117.8, and raises the "other teams" up from 77.3 to 82.3. In that scenario, one extra fumble every 271 touches would result in an identical rate to that group's with the Patriots. That's a bit over one a year, which could easily be attributed to an organizational emphasis on ball protection.
Going back to 2008 allows you to add David Thomas to the cohort. This brings the w/Pats number from 117.8 to 119.3, and "other team" total from 82.3 to 85.9. Hell, simply adding Julian Edelman to the group will bring the w/Pats touches total to 1936, quite close to the "other team" total of 1975 (yes, I'm aware he's only been on the Patriots), and suddenly the "with Patriots" cohort fumbles once every 62.5 touches, while the "other team" cohort once every 85.9 touches. Yes, I'm moving the goalposts here, but that's the point: choosing 2010 as a starting date seems arbitrary, gives a smaller sample, and appears to have been done so to screen out Ben Watson, who makes the shocking "185.3 touches" figure look a far more reasonable 117.8.
Eightsworth - you brought up something that I hadn't even considered. I saw the initial reports dating the change in the ball handling rules to 2006, so I assumed they took effect only in the 2007 season. Should have confirmed that they took effect in 2006.
Sharp's entire analysis is premised on something happening in 2007 that changed the Pats' fumble rates, and attributes it to the rule changes. Well, it's clear that the rule change was already in effect by then. On the other hand, what did happen in between 2006 and 2007 was that Brady met Gisele. Maybe she caused his grip on his balls to improve.
Sharp's entire analysis is premised on something happening in 2007 that changed the Pats' fumble rates, and attributes it to the rule changes. Well, it's clear that the rule change was already in effect by then. On the other hand, what did happen in between 2006 and 2007 was that Brady met Gisele. Maybe she caused his grip on his balls to improve.
I'm finally looking at the Sharp article, and someone is using Brady's declining fumble rate as further proof of the Patriots cheating post rule change. From 2001-2005, Brady fumbled once every 1.7 games. From 2006-2014 he fumbled once every 2.6 games. Damning!
Wait...shouldn't the other guy who pushed the change be compared as well? From 2001-2005 Manning fumbled once every 2.8 games. From 2006-2014 he's fumbled once every 3.9 games.
Quick! Grab the pitchforks and meet me at the Manning household! He's clearly deflating his balls, too!
Question: does this "Sharp" guy remove road games played in domes, as he completely doesn't include "dome" teams? Why aren't they included, or at least their games on the road not played in domes? Also, does he think that teams finally decided to deflate balls at home in 2006, when prior to that they could do whatever they wanted to them? When was the PSI rule put into place? Did the fumble rate go up after?
Wait...shouldn't the other guy who pushed the change be compared as well? From 2001-2005 Manning fumbled once every 2.8 games. From 2006-2014 he's fumbled once every 3.9 games.
Quick! Grab the pitchforks and meet me at the Manning household! He's clearly deflating his balls, too!
Question: does this "Sharp" guy remove road games played in domes, as he completely doesn't include "dome" teams? Why aren't they included, or at least their games on the road not played in domes? Also, does he think that teams finally decided to deflate balls at home in 2006, when prior to that they could do whatever they wanted to them? When was the PSI rule put into place? Did the fumble rate go up after?
No the initial article throws out all games played by "dome teams", it includes any games that "outdoor teams" played in domes. This is a serious flaw that has been pointed out in his data.
The reason he picked after 2006 is that is when the current rules about doctoring the balls, lobbied for by Brady and Manning, went into effect.
The rule on the PSI has been around since the 30's and hasn't changed at all.
The reason he picked after 2006 is that is when the current rules about doctoring the balls, lobbied for by Brady and Manning, went into effect.
The rule on the PSI has been around since the 30's and hasn't changed at all.
Byrdbrain said:
No they didn't. They went into effect in the 2006 season itself, as mt8thsw9th's link and this Washington Post article show. So his data sets really should be 2000-2005 and 2006-2014. Not sure how much that changes his results, but his central claim is: "fumble rates for the Pats went down starting in the 2007 season onwards, so something must have happened in the 2006-2007 offseason" and attributes that to the rule change. But he was wrong about when the rule change actually took place.
Sharp excludes road teams entirely. Here are some more charts including those teams
% change of fumbles lost from 2000-2006 compared to 2007-2014
http://fat.gfycat.com/FoolhardyHandmadeBorderterrier.gif
The top three teams are NE, NO, and ATL: three teams with pocket QBs, prolific passing games, good or better pass blocking, and lesser reliance on RBs.
Sharp's points on players only doing this with NE is kind of blown up when you excude ST touches and Brandon Tate specifically. This chart is flawed as it includes Wes Welker's ST turnovers with NE while excluding them elsewhere. The difference does exist in the 30% range.
https://pbs.twimg.com/media/B8U_mNWCUAAWCMM.jpg:large
There is a good Reddit thread on these articles too:
http://www.reddit.com/r/dataisbeautiful/comments/2ttlvy/the_new_england_patriots_mysteriously_became/
It's been stated, but BJGE accounts for half the difference between Pats Players on NE v. off NE when you exclude ST touches.
% change of fumbles lost from 2000-2006 compared to 2007-2014
http://fat.gfycat.com/FoolhardyHandmadeBorderterrier.gif
The top three teams are NE, NO, and ATL: three teams with pocket QBs, prolific passing games, good or better pass blocking, and lesser reliance on RBs.
Sharp's points on players only doing this with NE is kind of blown up when you excude ST touches and Brandon Tate specifically. This chart is flawed as it includes Wes Welker's ST turnovers with NE while excluding them elsewhere. The difference does exist in the 30% range.
https://pbs.twimg.com/media/B8U_mNWCUAAWCMM.jpg:large
There is a good Reddit thread on these articles too:
http://www.reddit.com/r/dataisbeautiful/comments/2ttlvy/the_new_england_patriots_mysteriously_became/
It's been stated, but BJGE accounts for half the difference between Pats Players on NE v. off NE when you exclude ST touches.
More specifically, if you look at Pats players on NE vs after they left, BJGE's 3 fumbles in 2 games in 2009 accounts for ALL the difference. Not sure how much you can draw conclusions when a bad case of fumbleitis can swing the data so much.Fishercat said:
This analysis makes the same statistical mistake that every other analysis has made this far. The proper unit of replication is the individual player. The sample size is the number of players with offensive touches. The correct analysis is a repeat measures ANOVA with player included as a random effect.singaporesoxfan said:
...which is just a formalized way of saying the same thing that has been stated repeatedly in this thread. Imagine you have 10 players who were all on the Patriots and then they all left with enough touches to do the analysis. Say 9 of them had decreased fumble rates when they left, and the 10th went to play for a team whose quarterback liked slathering his ball in Crisco before every game. Say also that the Cincinnati Criscos were dependent on this 10th player for all running plays. That 10th player's fumble rate would increase dramatically, and swamp the other 9 players. You don't get to group all of that into one analysis and treat all fumbles the same. Each fumble the 10th player commits is not an independent event. There are factors both at the team level (stadium type, typical weather conditions, defenses faced, offensive schemes) and the individual player level (aka "talent") that will contribute to the totality of fumbles for each player. Because individual fumbles for a single player are not independent, you're overestimating sample size and therefore whatever test you run is detecting significance where there is none. It's pseudoreplication.
Slate has another article up, purporting to get to the bottom of the fumble numbers, and does not even mention the special teams issue.
This was a disappointing article in a variety of ways. I like the author (Jordan Ellenberg) quite a bit but his treatment of this issue was poor.drleather2001 said:
Oh, by the way, the Super Bowl Champions handled the ball 72 times against Seattle. Zero fumbles.
His writing regarding rational points on high dimensional varieties is better.Eddie Jurak said: