Quote Originally Posted by pbmax View Post
Its come under a bit of fire in the analytical community.

http://regressing.deadspin.com/why-t...mos-1681805710

I haven't gone through it all, but the first version of the article looked at lost fumbles which is problematic because fumble recovery is largely a matter of luck. He corrects this in an update to include all fumbles.

Second, he excludes Dome teams in some data. It has been argued that the Patriots are less outliers if you just exclude dome games, rather than teams.

Several of the probability numbers are miscalculated. Instead of using X, they use 1/X, and try to plot it against a normal distribution curve. Except no one know if 1/X has a normal distribution.

That said, Advanced Football Analytics (formerly Advanced NFL Stats) thinks despite Sharp's poor analysis, there is still enough of an outlier there to be worthy of investigation.
The article I read was apparently an updated one. It included all fumbles, not just fumbles lost. His play count seemed solid. You really don't have to look beyond those data points and the resultant "plays/fumble" calculation to see that something seems wrong, especially when the abrupt change after the rule modification is considered. How he subsequently massages and manipulates the numbers is mostly to generate pretty graphs. The raw numbers tell the story.