[CTcard]

 Sark was significantly better. Full stop.

Well, I am certainly not going to make any more effort to convince you anything about Helton or Sark. Will try to say a few words about choosing stats and errors.


First, what is a good stat for the comparison we are trying to make?  
Yards per play has some advantages - the numbers are easily available and do make an adjustment for pace of play. 

YPP has some disadvantages - primarily that it is very heavily influenced by style of play. 
As an example, for most purposes I'd would rate a 1 play 80 yard drive as equally effective as a 10 play 80 yards drive, but of course they are drastically different in yards per play. 

YPP is worse that total yards as a comparative stat for two teams within one game (since the two teams typically have about the same number of possessions). Here it's not so bad, but there is better.

I would advocate for per game FEIs available at Brian Fremeau's site, bcftoys.com. 
FEI is drive based, which I think is the best normalization for efficiency - and in general there is a ton of interesting stuff on this web site so it's worth looking at.


What difference is meaningful? 
I have the numbers to do this with FEI, so I continue with that. 

Over the first 5 games of 2015 $C came in at an average FEI of 0.200 and over the last 9 at 0.121.* That would be the difference between #8 for 2015 and #26. [Which seems about right, as it sandwiches $C's overall ranking at 17.]

That doesn't seem a brobdingnagian difference to me, but might to you. More importantly, the difference between the two values if 0.08 but the standard deviation in the set is 0.23. Pretty clearly those two numbers are not reliably different from each other. 

*Fremeau uses a weighting of game FEIs to get season FEIs, and doesn't completely explain how. The numbers indicate that there is about a 70% scaling from a simple average to his weighted average, so I have used that above. 

Throwing out data
Concentrating on one season does keep the talent level as even as possible, but also makes a very small data set with big variation (as just noted). It also eliminates any development in Helton after first 1/2 season as head coach. 

Adding more data inherently means differences in the talent pool. While QB is the most important position, it isn't the only position and its certainly not clear that otherwise all-things-are-equal. 

There is no exact better here, but if one goes for the larger data set you cannot simply choose to eliminate 60+% of the games for one guy and have a meaningful measure.

[CTcard]

 Sark was significantly better. Full stop.

Well, I am certainly not going to make any more effort to convince you anything about Helton or Sark. Will try to say a few words about choosing stats and errors.


First, what is a good stat for the comparison we are trying to make?  
Yards per play has some advantages - the numbers are easily available and do make an adjustment for pace of play. 

YPP has some disadvantages - primarily that it is very heavily influenced by style of play. 
As an example, for most purposes I'd would rate a 1 play 80 yard drive as equally effective as a 10 play 80 yards drive, but of course they are drastically different in yards per play. 

YPP is worse that total yards as a comparative stat for two teams within one game (since the two teams typically have about the same number of possessions). Here it's not so bad, but there is better.

I would advocate for per game FEIs available at Brian Fremeau's site, bcftoys.com. 
FEI is drive based, which I think is the best normalization for efficiency - and in general there is a ton of interesting stuff on this web site so it's worth looking at.


What difference is meaningful? 
I have the numbers to do this with FEI, so I continue with that. 

Over the first 5 games of 2015 $C came in at an average FEI of 0.200 and over the last 9 at 0.121.* That would be the difference between #8 for 2015 and #26. [Which seems about right, as it sandwiches $C's overall ranking at 17.]

That doesn't seem a brobdingnagian difference to me, but might to you. More importantly, the difference between the two values if 0.08 but the standard deviation in the set is 0.23. Pretty clearly those two numbers are not reliably different from each other. 

*Fremeau uses a weighting of game FEIs to get season FEIs, and doesn't completely explain how. The numbers indicate that there is about a 70% scaling from a simple average to his weighted average, so I have used that above. 

Throwing out data
Concentrating on one season does keep the talent level as even as possible, but also makes a very small data set with big variation (as just noted). It also eliminates any development in Helton after first 1/2 season as head coach. 

Adding more data inherently means differences in the talent pool. While QB is the most important position, it isn't the only position and its certainly not clear that otherwise all-things-are-equal. 

There is no exact better here, but if one goes for the larger data set you cannot simply choose to eliminate 60+% of the games for one guy and have a meaningful measure.