104: Dispersion
A look at teams that three different college football preseason power ratings agree and disagree on.
Toe meets leather one week from today.
Only a sport as magical as college football could have a Week Zero. There are seven games (five with at least one FBS team) next Saturday.
`Tis the season for spinning a story.
I do think college basketball analytics or advanced stats have never been more main stream. Torvik and kenpom are household names. After all, WAB is finding its way1 to Team Sheets next season.
College football might not be far behind. This season, we get an actual playoff that features around nine percent of FBS teams instead of around three percent. It’s not exactly the BCS, but there are a growing number of public preseason power ratings.
This is a good thing for the sport in my mind. We have lots of people trying to figure out how to rate 134 teams that play about 12 games per season in mega-conferences with systemic advantages and tons of uncommon opponents.
Three power rating systems that I find useful are SP+, FEI, and KFORD. We’re going to compare the preseason power ratings across these three models to see where they agree and disagree the most.
First, some words about each rating system:
SP+: from ESPN Staff Writer, Bill Connelly. SP+ is a tempo- and opponent-adjusted measure of college football efficiency. You can find it behind the ESPN+ paywall, it’s worth it. See the latest rankings released early this week, the methodology is based on returning production, recent recruiting, and recent history. Connelly even tries to rank FCS and NAIA teams.
FEI: from Brian Fremeau. There is a treasure trove of data on bcftoys.com, and Fremeau measures drive efficiency. FEI does an incredible job on pointing out how field position matters. This point of view can help you understand how football is complementary.
KFORD: from Kelley Ford. Find power ratings, regular season wins, and watchability charts at kfordratings.com. Big fan of Ford’s visualizations, the regular season wins required to get a CFP at large-bid is tremendous.
These aren’t the only preseason rankings, but the only three I focused on to start. I know Beta Rank, Massey-Peabody, and cfb-graphs.com2 are super useful too.
10 teams that SP+, KFORD, and FEI agree on the most
See more about how to read this chart in the footnotes3. While we’re using power ratings from the comparison, we’ll also use rankings to explain the differences because its a bit easier to grok or faster to understand.
The teams on this chart feel stable. Utah and Notre Dame are the most similar across these three models. FEI is highest on both teams, but the other systems are not far behind.
While those two teams are both well-above average, it’s also fascinating how all three systems are also similar for below-average teams like Arkansas State and Georgia Southern.
Clemson jumps out at the end of the list hovering around the top-ten in all three systems. The Tigers are projected to be two-touchdown underdogs in Week one against Georgia4.
10 teams that SP+, KFORD, and FEI disagree on the most
This is a fascinating list of teams. Obviously, Colorado leaps out at you. The Deion discourse is a race to the bottom5. The Buffaloes are as high as 43rd (KFORD), but rank 60th (SP+) and 75th(FEI) in other systems.
Michigan, the defending champs, show the third highest dispersion across the three models. FEI has the Wolverines ranked as high as 4th, while KFORD has Michigan ranked at 12th. The Wolverines are a good example why we’re normalizing the power ratings. It shows how unusual a team's rating is within each system, rather than just its position ranking in order.
Air Force shows the highest dispersion, although it’s entirely due to its 45th FEI ranking compared to its identical ranking of 88th in SP+ and KFORD.
This chart also feels like it supports evidence of these models factoring in transfers as Ole Miss and Miami are highly dispersed. FEI is way lower on both teams, so perhaps that model is doing something different?
And FEI is way higher on Washington (ranked 11th), while SP+ (32nd) and KFORD (36th) are lower. Washington has a new coach and lost a ton of production.
How to think about the differences
There are a ton of unknowns in college football. The transfer portal, coaching changes, injuries, new conference opponents, and an expanded playoff.
A ton of possible things to factor into trying to predict a sport ripe for randomness.
These preseason power ratings are a starting point6. It’s something to agree or disagree with, and you can also compare these metrics to the market or your favorite sportsbook.
You can find the code to generate this data here, complete with matching team names across all systems7.
Bonus: ACC dispersion
North Carolina makes an appearance as a team that is similar across all three ratings hovering around the top forty. UNC is a perfectly above average to average football team.
FEI is about 14 spots lower in its ranking of NC State than SP+ and KFORD. Again, perhaps, this suggests transfer portal or how each model takes a transfer portal QB into account.
A reminder that Notre Dame is about ~63 percent (-170 odds) to make the college football playoff and about ~41 percent (+142) to win more than 10.5 games. This means the Irish likely still make the playoff at a 10-2 record.
My curiosity is do the Irish make the playoff with a 9-3 record? Notre Dame plays five ACC teams, and only one (Georgia Tech) away from home. A loss to an ACC team would really dial up the propaganda about the strength of conferences.
If you think college basketball was bad, the playoff committee probably brings more shenanigans when it comes to earning a spot given there are only 12 total teams.
If a 9-3 Notre Dame team misses the playoff, how many ACC teams get in?
If a 10-2 Notre Dame makes the playoff, how many ACC teams get in?
¯\_(ツ)_/¯
Thanks for reading this far, and please subscribe if you haven’t already. If you’re looking for the full list of comparisons across all 134 FBS teams, you can find it here.
Skeptical of deriving WAB from NET, but who knows? People are going to reverse engineer NET, which is good.
Eckel ratio from Parker Fleming is a pleasantly stupid stat. His words, not mine. I think its a great way to get a sense of game control and how offense can help defense, and vice versa.
This uses the actual ratings rather than ranks because it preserves more information about the relative strengths of teams.
Range is the difference between the maximum and minimum ratings. It just measures the spread of the data.
Standard deviation is the average distance between each rating and the average of all ratings.
Average is the sum of all ratings divided by 134 (number of FBS teams).
The lower these numbers, the more agreement. And the higher these numbers, the more disagreement.
The Tigers won 10 or more games for 12 consecutive seasons before winning nine games last season. In the second half last season at Duke, Clemson finished five of its six drives in Duke territory and did not score any points in that half. Zero.
Not sure we need to try to reconcile Deion Sanders the player, the person, and the coach. The coverage is exhausting. It’s perfectly fine to not have a take on it.
Ted Knutson has a great video talking through how to model soccer (fútbol) across the Premier League. The Transfer Flow looks legit. Also, EPL is back.
Every time I see the naming conventions switch between North Carolina State or N.C. State or NC State, I sort of lose my mind.