Switching it up... ON A MONDAY.
These aren't the power rankings. Which you probably could tell based on the conspicuous absence of the numbers 1-32 running down the left hand side of the page. Instead, this is an introduction to the new and improved method I'll be using. Before, I didn't have all the stats. Now, I do. I'll be using play-by-play data for every NFL game (more stats) and using a more robust regression model for handling opposing strength (better stats).
Most importantly though, this blog post represents something bigger than just a new model. It represents me desperately buying as much time as possible before you realize that I started writing this model the Friday after the NFL season started, and that it's taking longer than I expected to handle all of pro-football-reference's bugs in their play-by-play data (They don't give the offensive team for each play, so I have to rely on their expected points column to figure it out, but on kickoffs, they're inconsistent about which team they're showing the expected points for). My timeline is to finish the parsing-out-the-bugs part of the model tonight and then the regression part tomorrow, but I have to do a bunch of testing on last season's data to make sure everything's working right before I can publish, so I won't be hitting the weekly Tuesday deadline that I'd set for myself before the season.
But yes, this means that for every week starting week 2, the rankings will be going up on a Tuesday.
No comments:
Post a Comment