Quote:
I'm not arguing that points per play or yards per play should be what is used to rate offenses, because I haven't looked enough into it to form an opinion. I'm using these stats to rate defenses.
1- It's either good for both or not. The calculus of did they move it or score really doesn't change much. It's either giving a good picture answering those questions or it isn't. The offense/defense split makes little sense.
Quote:
I don't think it's a coincidence that the top 3 defenses in allowing pts per play and yards per play are the same top 3 defenses ranked by S&P and were widely considered the top 3 defenses in CFB this past year (Alabama, Clemson and Wisconsin).
2- So a system based on per play efficiencies falls in line with per play stats? Not exactly shocking...
Yards per drive and yards per game has Michigan mixed in with those three.
3- However that once again biases against style of play. Virginia Tech had outstanding YPD defense ranking 11th and were very good all year. However they tend to be aggressive and gave up median or worse numbers of 20+ yard plays as they roll the dice a lot. 27th in yards per play defense which isn't indicative of how good they were. They were also 11th in tackles for loss.
Quote:
I would agree with your example on the 13 plays if that's all the offense got in a game, but most teams run 65+ plays per game which gives you way more data points. All those data points per game and consequently over the course of the season will help balance out the extreme outliers in terms of plays like the one from your example above.
My example was an extreme one to prove a point. Since you seem to be ignoring that long plays have bigger per-occurrence impact and more frequency... lets add actual numbers here.
4- Typical plays go for handfuls of yards which means a long play is giving a big impact to any average. Typical drives go for 33 yards. Outliers have a much bigger impact on per play as even a 99 yard drive would have only about a 5.5 YPD impact over a 33 YPD average figure over the course of a typical game with 12 drives but drives that long are rare. Touchdown drives are only 27% of estimated possessions. Taking a long drive of 75 yards would only be 3.5 per drive above average usually.
Let's apply that same math to YPP. A typical offense has 70 plays. One additional 65 yard gain can take a single game yards per play statistic and take it from a median yards per play figure to a number on par with a top 19 season ypp figure when it's impact is divided out over 70 plays. Add in that a team who had a median number of 20+ plays like Nebraska had roughly 5 of them per game and it's easy to see how the stats skew and skew and skew based on how many big plays are hit and especially if they go 30+, 40+, etc...
Quote:
Those skewed plays are not likely to be a big factor when you consider the sheer number of plays that will used in the statistics.
5- My example showed it IS a big factor. For instance 3-4 long plays really skewed YPP in BU vs TCU 2017. Yards per play was 6.24 which is on par with what the 29th best offense in the country in YPP averaged for a season. Baylor had some good plays but was nowhere near as good that day as the stat would suggest. Yards per drive had them at 26.2 which was much closer to reality where Baylor had 15 drives with 2 field goals and two long TD's to show for it while giving up tons of sacks.
Quote:
By contrast, teams might only get 6 or 7 drives per game and to me, that is a much easier to manipulate that statistic because there are much fewer points of data than there are on a per play basis.
6- You quite literally are off by double. Typical games are 12 drives. So your argument just lost half it's strength as there's double the sample size to dilute any outliers.
7- Outliers
need less diluting with per-possession stats. A team driving 60 yards for a TD is still only 81% more than what a median drive would be. Even a mundane ten yard gain is 78% higher than median play yardage. (5.61) Considering a median team will have roughly 13 plays of 10+ a game... that stat is gonna skew and do so quickly with any marginal variation from that.
A team getting a modest 20 yard gain? 257% more than median. 30 yards? 435% of median YPP. If they hit just a couple bigger plays the stat gets really slanted.
Play by play stats are far more variable and skew easily and the play counts aren't enough to balance it out.
Quote:
A team can be trying to run clock at the end of games or kneeling in victory formations and that will skew numbers on a per drive basis much much more than they would on a per play basis.
8- I omit obvious kneeldown drives from the stats. They bias nothing as a result of that. As far as running clock, they are still trying to move the ball if only to make first downs and bleed clock.
Quote:
I agree with your last point. Fewer busts, particularly by having improved safety play, should make the defense much better. Just how good that is remains to be seen.
I think this unit has the potential to surprise people but they have to get the mental screwups under control. Stuff rate was 25th nationally which is encouraging if they can get the mental part down.