Read on Twitter
Working on Opponent-Adjusted EPA but wanted to share this to illustrate why we use EPA over Yards
Here we are trying to predict the point differential for a given game using:
EPA/Play of prior games
EPA/Play from last year
EPA/Play of prior games and last season
Same for yards
Here we are trying to predict the point differential for a given game using:
EPA/Play of prior games
EPA/Play from last year
EPA/Play of prior games and last season
Same for yards
After a few weeks into the season and we find out more about teams EPA/Play has more predictive power than Yards/Play
Also important to note how generally unpredictable point diff. is
This is also true of the NFL, as you can see in @PFF_Moo's thread
https://twitter.com/PFF_Moo/status/1098244584633700352?s=20
Also important to note how generally unpredictable point diff. is
This is also true of the NFL, as you can see in @PFF_Moo's thread
https://twitter.com/PFF_Moo/status/1098244584633700352?s=20
The big reason that comes to mind as to why it would be lower in college would be the high turnover of rosters each year and a much wider talent gap in a given game between 2 teams (~130 in NCAA vs. 32 in NFL)
It's also interesting how long one can hold onto their priors from last season as the "EPA Previous Games" curve doesn't begin to converge with the "EPA Previous + Last" until about week 12
A product of easy early season/non-conference schedules for most teams?
A product of easy early season/non-conference schedules for most teams?
This is a lot more stats-heavy than most of the content we posted so if you have any questions feel free to reach out
And thanks to @PFF_Moo & @gberg1303 for their public work on the topic in the NFL, I tried to adapt their methodology in this work
And thanks to @PFF_Moo & @gberg1303 for their public work on the topic in the NFL, I tried to adapt their methodology in this work
