Great post at PFF looking at the value of NFL draft picks (based on second contracts) under a hypothetical where you knew who the best players were. In other words, how much is the first pick worth relative to the second pick, third pick, etc., if you knew in advance who the first, second, third best players (by contract value) were? It’s a brilliant approach because it establishes a bound, and asks under the most favorable conditions, how much are top picks worth?
Some months ago I discussed the NFLs Big Data Bowl. It’s fascinating to see all the talented people go after these problems and see what the come up with. Here I thought I’d review the videos that the finalists made and offer some comments. The challenge this year was to make a contribution to understanding passing defense using a subset of tracking data which has the skill position players on offense and passing defenders.
The NFL season is not over, but I thought this would be a good time to collect my thoughts on roster construction in general and the Seahawks situation in particular. I’m excited to compare these ideas to what the team actually does over the next six months.
The Seahawks are in a fairly good position heading into the offseason. They have one of the league’s best quarterbacks, and quarterback is 3-10x more valuable than other positions (excluding specialists, who are even less valuable).
It’s pretty clear that passing is more effective than running for an offense. Expected points added (EPA) and win probability added (WPA) are both higher for passing than running. It therefore seems reasonable that a team should want to dedicate resources to making their passing offense and passing defense more effective, and that offenses should pass more and defenses should try to induce opposing offenses to run more.
In this post I’ll take a look at how well teams are doing in terms of balancing the effectiveness of passing and running.
The 2021 edition of the NFL’s Big Data Bowl is here. The challenge is to evaluate defensive performance on passing plays using player tracking data, without offensive or defensive lines.
Resources that I’ve found helpful :
Tom Bliss tutorial Definitions of events in the data How to animate a specific play R for NFL analytics What’s covered here:
SQLite database of the tracking data examples of animations using the best and worst (by EPA) Seahawks plays in the dataset.
I’m really enjoying playing with nflfastR.
There are a couple of nice quarterback measures in there - completion percentage over expectation (CPOE) and expected points added (EPA). I looked at all quarterbacks that had at least 100 passes each for CPOE and EPA (EPA has more plays available), there were 175 of them. This covers from 2006 forward through this past weekend’s games.
There were 15 that were top 10 on either measure:
The relatively new nflfastR package is pretty incredible. One command and you get a local SQLite database with 20 years of cleaned play by play data:
update_db( dbdir = “/home/michael/Documents/NFL/”, dbname = “nflpbpdb”, tblname = “pbp_raw”, force_rebuild = FALSE)
And you can run the same command to update with any new games. Incredible.
Playing around with it, one measure you get is completion percentage above expectation (CPOE), which takes into account the relative difficulty of each throw and the outcomes.
