Thursday, August 31, 2017

Football is back! (Week 1)

I know technically football was back last saturday, but c'mon. Today things start in earnest. As I write this I'm watching Oklahoma State alllmost block a Tulsa punt. And I'm in Paris and it's the middle of the night!

College football is back, and my model is back. Early in the season last year I had a serious life situation that was completely disruptive of my hobbies, college football modeling included. I'm excited to be back.

The changes this year are fairly minimal. I've added UAB (welcome back) and Coastal Carolina (welcome!) to the model. I've added the Big-12 title game, and I've done some refining around how the model use a team's season performance to predict its CFP rating. I also leaned on Massey's ratings to smooth over some edges around how FCS teams are rated (while I don't model FCS football, I do give ratings to each FCS team who plays an FBS team).

Lastly, as always I'll experiment with new ways to share/present/communicate college football data.

The model hasn't changed much, but let's go over the basics.

The Data Viz

  • Top 25
    • On the left hand side I'm adding a Top 25 section. I'll experiment with the look/feel as the season goes on, but for now it's simple - team and rating.
  • Team Dashboards
    • If you've been following on twitter (@actuarygambler) you've seen preseason dashboards tweeted out for each FBS team over the last week. If you follow the link here or or the top right you can make one at any time for any team. I update them after each week.
  • More to come I expect
  • Games of the week
    • Something we all love about college football is that every regular season game is meaningful. A loss in week one could have season-long implications. In that spirit, I've added a snapshot for each of the 10 most watchable games each week that shows each team's chance to win, as well as how winning or losing will impact their chances to make the CFP and a bowl game
  • Top Games
    • The first year I built a model, one of the first metrics I created was Watchability. It's designed to capture how likely a game is to be close and well-contested. It's based on how good the two teams are and how likely the game is to be close. It's a prediction, which means it doesn't always work out, but over the years it's opened my eyes to games I might not have thought to go check out.
    • This table show the 10 most watchable games in each week. It's going to be an awesome week 1 with serious postseason implications.
    • Each team is shaded green according to their chance to win, the darker their shading the better their chance to win. Most of the "Top Games" are pretty even matches, as expected.
  • Week Schedule
    • This table is similar to Top Games, except it has ALL the games, including matchups of the titans, such as Presbyterian @ Wake Forest

The Model
  • Basics
    • Each team has a rating between 0 and 1. Game odds are calculated using a team's rating and the rating of its opponent
      • If both teams have the same rating, each has 50% chance to win
      • If a team is rated 0, they never win. If a team is rated 1, they always win
      • The full college football season, including regular season games, conference championship games, bowl games, and the CFP, is simulated using the team ratings
    • As the season plays out, team ratings are adjusted based on how their on-field performance compares to modeled expectations

    No comments:

    Post a Comment