Interceptions vs. Fumbles as a Winning Stat
NFL stat maestro Bud Goode said long ago that interceptions are three times more costly than lost fumbles and that lumping lost fumbles with interceptions as "turnovers" made as much sense as lumping doubles and homers together under "hits" in baseball.
Of course, this is incredibly counterintuitive. Goode never came up for a reason for the discrepancy. But he speculates that lost fumbles are actually a "power" stat because teams that win more run more in the second half and thus, presumably, fumble more. More important, he said, the conclusion is consistently supported through regression analysis.
Now, I'm no statistician. But I do know my way around a scatterplot. I understand the importance of correlation. But outliers can distort correlation coefficients. For simplicity's sake, let's look at net interceptions, which we track in our Stat Power Index, largely inspired by the work of Mr. Goode, and net fumbles. We'll examine the 10 best and worst teams in each net category and compare records. If lost fumbles are so much less meaningful than interceptions, we should see a poorer correlation to success and failure as reflected in the won-loss records of the leaders and trailers in each group. Because one year can distort, we'll look at 2004, too.
First 2005. The 10 best teams in net fumbles (recovered vs. lost) had a .642 winning percentage. The 10 worst, .422. In net interceptions, the 10 best teams last year had a .675 winning percentage, the 10 worst, .324. Advantage net interceptions, but that does not seem to correspond to a 3-to-1 ratio in relative importance in favor of interceptions.
I understand that Mr. Goode wasn't tracking the net stat, but rather each specific instance of a lost fumble or interception in the context of a game. Still, he nets other stats, such as YPA, so why not do the same with lost fumbles and picks?
Was 2005 an outlier year? In 2004, the 10 best teams in net fumbles had a .585 winning percentage, the 10 best in net picks .675. That's much more significant on the winning side. But net picks did not do as good a job in predicting losers as in 2005, with the trailers in net fumbles coming in at .375 and trailers in net picks actually having more success with a winning percentage of .400.
I was shocked to find that 6 of the 11 teams that at least tied for the top 10 spots in net fumbles repeated the feat in 2005 (when two teams again tied for 10th place). Only four teams that led in net picks in 2004 repeated the feat in 2005. I would have guessed that lost fumbles were more a product of luck than interceptions. But that conclusion is not supported by this small data set. On the other hand, the Saints and Jets were tied for first and third in net fumbles in 2004 (plus-10 and plus-9, respectively), but LAST in 2005 (-10 and -12, respectively). A swing of 20 turnovers can cause a radical swing in won-loss record. Giants fans should be wary, as they led the NFL last year in the stat, at plus-12.
Of course, this is incredibly counterintuitive. Goode never came up for a reason for the discrepancy. But he speculates that lost fumbles are actually a "power" stat because teams that win more run more in the second half and thus, presumably, fumble more. More important, he said, the conclusion is consistently supported through regression analysis.
Now, I'm no statistician. But I do know my way around a scatterplot. I understand the importance of correlation. But outliers can distort correlation coefficients. For simplicity's sake, let's look at net interceptions, which we track in our Stat Power Index, largely inspired by the work of Mr. Goode, and net fumbles. We'll examine the 10 best and worst teams in each net category and compare records. If lost fumbles are so much less meaningful than interceptions, we should see a poorer correlation to success and failure as reflected in the won-loss records of the leaders and trailers in each group. Because one year can distort, we'll look at 2004, too.
First 2005. The 10 best teams in net fumbles (recovered vs. lost) had a .642 winning percentage. The 10 worst, .422. In net interceptions, the 10 best teams last year had a .675 winning percentage, the 10 worst, .324. Advantage net interceptions, but that does not seem to correspond to a 3-to-1 ratio in relative importance in favor of interceptions.
I understand that Mr. Goode wasn't tracking the net stat, but rather each specific instance of a lost fumble or interception in the context of a game. Still, he nets other stats, such as YPA, so why not do the same with lost fumbles and picks?
Was 2005 an outlier year? In 2004, the 10 best teams in net fumbles had a .585 winning percentage, the 10 best in net picks .675. That's much more significant on the winning side. But net picks did not do as good a job in predicting losers as in 2005, with the trailers in net fumbles coming in at .375 and trailers in net picks actually having more success with a winning percentage of .400.
I was shocked to find that 6 of the 11 teams that at least tied for the top 10 spots in net fumbles repeated the feat in 2005 (when two teams again tied for 10th place). Only four teams that led in net picks in 2004 repeated the feat in 2005. I would have guessed that lost fumbles were more a product of luck than interceptions. But that conclusion is not supported by this small data set. On the other hand, the Saints and Jets were tied for first and third in net fumbles in 2004 (plus-10 and plus-9, respectively), but LAST in 2005 (-10 and -12, respectively). A swing of 20 turnovers can cause a radical swing in won-loss record. Giants fans should be wary, as they led the NFL last year in the stat, at plus-12.
Entry #115611179337293523 posted by Catherine Salfino at 5:34 PM
<< Home