Note: The web scraper I was using to gather data from various sites is broken, so I stopped updating October 24, 2016. I'll look to resume this rating after I figure-out a more reliable way to get statistics.
An attempt to rate college football teams (good).
This rating system was created for my own enjoyment, i.e. I enjoy college football and I enjoy lists. Much of the "information" presented here is anecdotal, at best. That's intentional. Football is simulated warfare + shoulder pads. Any attempt to seriously quantify it should be met with tremendous skepticism.
That being said...
goodrating is an attempt to accurately rate college football teams using the fewest number of statistics as possible. Like people, statistics have biases. In limiting the number of statistical factors, I hope to control the effect of their bias.
But... how can a number be biased?
It's all about context. E.g. a team can have a 0.950 winning percentage, but that number doesn't account for the quality of teams they played while achieving that number. A high winning percentage could be an indicator of greatness, but it could also be an indicator of their opponents' collective weakness.
Ok, what are you going to do about it then?
Glad you asked!
Here are the four main pillars of goodrating:
- Winning percentage
- Offense rating
- Defense rating
- Player rating
A lot of rating systems will adjust a team's winning percentage based on their "strength of schedule (SOS)," i.e. how good were the teams they played? I don't like using SOS, though, because although it rewards successful teams that play difficult schedules, it also tends to reward bad teams that play difficult schedules. Instead, I will attempt to adjust their rating based on the team's cumulative performance versus expectation, aka "the spread."
If a team wins a lot of games, but not as convincingly as they should, they are adjusted down. If a team has an average record, but a lot of close losses, they are adjusted up, and so on.
For now, this is purely a team's average yards per play. There is a wide variety of offensive styles in the NCAA, but all of them have the same goal: gain a whole bunch of yards.
Much like the Offense rating, this is based the team's average yards allowed per play.
This is the team's combined recruiting scores during the previous four years, as provided by the 247 Sports' Composite Rankings. Although the scores are entirely based on projections, there is a strong correlation between recruiting well and winning football games. See the recent SBNation article The 2016 Blue-Chip Ratio: How close is your CFB team to having title-level talent? for more explanation on that. (Spoiler: Alabama gets good players)
What about teams that play each other?
Good observation! We don't track specific game outcomes. That's because head-to-head results don't matter in this rating. This is an attempt to quantify the "goodness" of team, not a predictor of game outcomes. Therefore, it is possible that a team will be rated above a team that beat them. Sorry.
Each statistical category, is normalized to a 100-point scale. For example, if the best winning percentage in all of football is 0.9, then the team(s) with that winning percentage get a score of 100. After each score is calculated, we apply some weights. Here's how the weights are currently distributed:
- Winning percentage -> 35%
- Offense rating -> 20%
- Defense rating -> 20%
- Player rating -> 25%
And before you ask, yes, those weights are complete arbitrary!
After adding the weights, we simply add the four scores and rank accordingly, high-to-low.
This part is still very much a work-in-progress. Currently we are adjusting based on each team's cumulative "Against the Spread" performance plus their "Strength of Schedule" rating provided here and here, respectively. We cap each score at 5.0. There is currently no cap for negative scores, because I'm cruel.
Just like any statistics-driven system, the rating will (theoretically) improve as the sample size increases. This means that there is a strong possibility that the ratings will be less-than-accurate during the first few weeks. Please be patient (also, see the note about this being for fun at the top).
If you have any questions/comments please feel free to open an Issue or email me at davenich [at] gmail.