Pinned post
This fall I started a side project: building a college football rating system from scratch.
It rates every FBS team on two things, how good they are (adjusted for opponent and venue) and what they’ve actually accomplished. From there it ranks teams, predicts every game, and keeps a public record of how those predictions do.
The part I’ve enjoyed most is the testing. Every idea gets checked against five seasons of games before it goes in, and most ideas don’t survive. A few things I learned along the way:
Home-field advantage was overrated in my early versions. Home teams fell short of my predictions in every season since 2022, until I recalibrated.
Bye weeks, rivalry games, and “lookahead” games barely move the needle. The famous upsets are memorable, but on average favorites win as often as they should.
The AP poll knows something. Ranked favorites win more often than a pure numbers model expects, so mine now gives the poll a little credit.
Showing your misses builds more trust than hiding them. Every week I publish a report card of every pick, right or wrong.
I’m now sharing it all in a weekly newsletter: rankings, predictions, the report card, and whatever the data turns up. College basketball starts in November.
If you like sports, data, or watching someone grade themselves in public, I’d love for you to follow along: nickisratings.substack.com
Built in Python with public data from ESPN and CollegeFootballData.
It rates every FBS team on two things, how good they are (adjusted for opponent and venue) and what they’ve actually accomplished. From there it ranks teams, predicts every game, and keeps a public record of how those predictions do.
The part I’ve enjoyed most is the testing. Every idea gets checked against five seasons of games before it goes in, and most ideas don’t survive. A few things I learned along the way:
Home-field advantage was overrated in my early versions. Home teams fell short of my predictions in every season since 2022, until I recalibrated.
Bye weeks, rivalry games, and “lookahead” games barely move the needle. The famous upsets are memorable, but on average favorites win as often as they should.
The AP poll knows something. Ranked favorites win more often than a pure numbers model expects, so mine now gives the poll a little credit.
Showing your misses builds more trust than hiding them. Every week I publish a report card of every pick, right or wrong.
I’m now sharing it all in a weekly newsletter: rankings, predictions, the report card, and whatever the data turns up. College basketball starts in November.
If you like sports, data, or watching someone grade themselves in public, I’d love for you to follow along: nickisratings.substack.com
Built in Python with public data from ESPN and CollegeFootballData.
