Welcome, oh great and wonderful college football fans, to episode 689 or a review of how the GCR actually works and the metrics we calculate. Throughout the season, we’ll talk about most of this content, but we felt it might be good to have a reference post for our approach.
…..
When talking to people about the GCR, we get the most surprised reaction to our inputs to the algorithm. For football, we only use game location (away, home, neutral site), the final score of the games, and strength of schedule (SOS). That’s it (note: most computer ranking systems, like Sagarin or Massey, only use final score and some form of SOS calculation). FPI, and others like it, do use individual player metrics, along with in-game stats, to do their calculations. While we think it would be intellectually fun to have that detailed level of data to create a different algorithm, we are a bit ingrained with what we have although we do continue to improve it every year. For basketball, we only use the final score of the games. The primary reason for the difference is the number of games played. With just 12 regular season football games, we have to build the adjustment for team differentiation. With ~30 regular season games, basketball naturally normalizes without the adjustment. Two big rules for the GCR: 1) every metric is recalculated based on the most current information regardless of when past games were played, and 2) future games have zero impact on rank, strength of schedule, or any other metric.
…..
The key metric in the GCR ranking is the Score. For football, we currently do not have a normalization to a maximum value, but in Men’s and Women’s basketball, we normalize so that the top ranked team has a Score of 100. That could change for football next season. For all three, the Score is made up of two values: the performance and SOS metrics. Performance takes into consideration whether a team won or lost and the point differential. SOS takes into consideration the value of the team played (that team’s Score) along with that opponent’s opponents’ SOS (how tough is their schedule). In football, we add a location factor. As the season progresses the ratio between the two factors change as both normalize (central tendency). The only other factor (and the only “bias” we have) is top-tier conferences (Core 4 in football, Majors in basketball) have no adjustment to their values, while other conferences (Group of 6 and, more so, FCS in football, Mid-majors in basketball) have small adjustments to SOS (or rather to the teams that play them) that become less relevant as SOS normalizes over the season. For both football and basketball, we have an RPI value, but we don’t use the football version currently (still working on how to incorporate it). In basketball, we use the value as a predictor: if both RPI and Score suggest Team A will win the game, we predict Team A. If they disagree on the winner, we issue a no prediction.
…..
When we have an Evaluation (Top 266 for football, Top 363 for Women’s basketball, or Top 365 for Men’s basketball), the rank is solely on the Score. We publish these rankings once per week but update them daily. We use that most current value to predict the winners as well as who should cover the Vegas spread. We track our Prediction Accuracy on those games. We’ve published the accuracy in total each week (separating FBS and FCS for football) but this year, we’re providing more information. In football, we are showing week-by-week success rates by Game Category – a term we created to show the seven possibilities: 1) Core 4 vs Core 4, 2) Core 4 vs Group of 6, 3) Core 4 vs FCS, 4) Group of 6 vs Group of 6, 5) Group of 6 vs FCS, 6) FCS vs FCS, and 7) FCS vs non-Division 1. We most likely will show basketball in 5 Categories: 1) Major vs Major, 2) Major vs Mid-major, 3) Major vs non-Division 1, 4) Mid-major vs Mid-major, 5) Mid-major vs non-Division I.
…..
We also have a progress report. Historically, we show monthly conference comparisons, based on mean Score, to rank from first to worst. In addition, at least for football, we are showing the simple win/loss record for each conference on the weekly Evaluation. We may have something similar for basketball, but that is still TBD as we plan the season. Occasionally, we see something in the data that interests us and will post a special Explanation to share it.
…..
One way to measure offensive success is to measure yards gained and divide by potential yards gained. For example. If a drive starts on a team’s own 25 and they drive 50 yards, the value would be 50/75 or .667. The fault in that measure, is that teams do not win games based on yards gained. That drive could have ended in a turnover which (other than field position) negates that drive. Or it could have ended in a field goal (or a missed one). Theoretically, a team could go 50 yards, get stopped on the 25, and score 0 points on every drive. Another factor in judging a team’s offense is relative. There are teams early this season who are averaging 60 points a game, but their competition has been…not so much. A few years ago, we tried to tackle the question “how do we value an offense when we only use the final score?” Certainly, we could have pulled in more data, but that was not the plan, unless necessary. We finally came up with a metric (along with its defensive counterpart) that worked for our purposes.
…..
While we don’t publish this every week due to space constraints, we track it constantly, for football only. We call it Offensive Efficiency Rating (or OER) which compares a team’s score vs each opponent’s average, normalized to 100. For example, if Team A beats Team B 35-10 and Team B normally gives up 35.0 points per game, Team A’s OER, for that game, is 100. If Team B normally gives up 17.5 points per game, it’s 200. If Team B normally gives up 70.0 points per game, it’s 50. It sounds simple, but each team’s OER is then the average of each game’s OER. The complication is Team B’s average points allowed per game, in the example above, will change each time they play, which directly impacts every opponent’s OER value. This value does not care whether a team is Core 4 or FCS (although non-Division I teams are ignored), which means, the most effective offense (highest OER) may be an FCS team, if they consistently score more than their opponents normally allow. One important feature in the math: James Madison put 87 on Wagner last weekend. That 87 is included in Wagner’s average points allowed which lessens the impact of runaway scores. Here are the current Top 10 OER so far:
| Rank | Team (Record/Rank) | OER Value |
| 1 | Arizona St (1-1/65) | 252 |
| 2 | Army (1-1/99) | 212 |
| 3 | Temple (1-1/33) | 206 |
| 4 | Virginia (2-0/11) | 198 |
| 5 | NC St (1-1/76) | 195 |
| 6 | TCU (1-1/64) | 190 |
| 7 | Tennessee (2-0/14) | 189 |
| 8 | Maryland (2-0/16) | 187 |
| 9 | Northwestern (1-0/67) | 185 |
| 10 | Auburn (2-0/23) | 182 |
…..
Arizona St beat Morgan St 70-7. The Bears beat NC A&T 17-7 and a non-D1 team (Lynchburg 55-0), which does not count in the factor. Then they lost to Texas A&M 20-48 who scored 50 on Missouri St. Put those values in the algorithm and they have the most Efficient Offense so far at 252 (they score 25.2 points for every 10.0 points their opponents normally allow). By season’s end, an OER of 150 is exceptional.
…..
The last metric is the opposite of OER, the Defensive Efficiency Rating (DER) which calculates how the defense (which includes special teams and turnover points) handles opponents. If Team A allowed 10 points to Team B, that sounds pretty good. Team A should win a lot of games if they do that over and over again. But what if that is Team B’s 4th game and the first points scored all season? Team B’s average is 2.5 per game, so Team A’s DER is just 25 (they should have given up 2.5 points but gave up 4 times that much – 1/4 = .25). If Team B averages 30.0 points per game, the DER is 300. Again, the GCR recalculates after every game played so the DER would adjust as Team B improves or declines. Here are the current Top 10 DER so far:
| Rank | Team (Record/Rank) | DER Value |
| 1 | Virginia (2-0/11) | 577 |
| 2 | North Carolina (2-0/37) | 506 |
| 3 | Penn St (2-0/9) | 417 |
| 4 | Maryland (2-0/16) | 364 |
| 5 | South Carolina (2-0/25) | 356 |
| 6 | Kansas St (2-0/35) | 345 |
| tie | Texas A&M (2-0/7) | 345 |
| 8 | UCF (1-1/50) | 333 |
| 9 | Pittsburgh (2-0/13) | 331 |
| 10 | Notre Dame (2-0/3) | 308 |
…..
Virginia beat NC St 34-8. State then beat Richmond 73-0. Then VIR beat Norfolk St 59-3. NFST beat a non-D1 team (Winston Salem 56-0) which does not count but lost to Old Dominion 31-10. The Cavaliers boast a video game level DER of 577 (they give up 10 points for every 57.7 points their opponents normally score). By season’s end, a DER of 150 is exceptional.
…..
That’s it for today. Tune in Thursday for a one-game Expectation as we kick off Week 3. Thank you for reading, commenting, and sharing with others. If there are other cuts of the data you’d like us to review or any suggestions for improvement, we are all ears, JoJo and G.
