Reconsidering Team Talent Composite: A Note

Fleshing out some thoughts on how we value talent in college football.

Every year, 247 Sports releases a team talent composite. Check it out here.

The team talent composite does some fancy weighting to prioritize top talent, providing a broad brush representation of where talent is concentrated in college football.

The Top 5 from the last three seasons are here as follows:

2024: Alabama 1018.28, Georgia 1008.21, Ohio State 998.62, Texas 953.95, Clemson 924.63

2023: Alabama 1015.43, Georgia 977.80, Ohio State 974.79, Texas A&M 926.20, Clemson 917.88

2022: Alabama 1016.83, Georgia 989.89, Ohio State 983.26, Texas A&M 948.02, Clemson 943.04

It should not be a hot take that the House That Saban Built holds the most talent in college football, and that Kirby Smart’s Georgia is hot on the Tide’s heels. It could be noted that the gap is closing (26.94 in 2022 as compared to 10.07 in 2023), and there is evidence to support that, but also small sample size caveats abound, in addition to the one sided nautre of roster composition. Alabama currently has 14 5 stars and 61 4 stars, and there are only so many five stars and roster spots to go around.

What I want to attempt to answer in this quick post is a two-fold question:

Can we recreate the team talent composite rankings, and can we imagine a weighting system that accounts for positional value to provide more information about the distribution of talent?

As to the first question, I’ll employ my friend Bill/collegefootballdata.com to pull the 2024 ratings for players and see what we can do with pre-scraped data. In another life, one of lazy grad school afternoons and fewer screaming children in the night, I built a 247 scraper, but they moved things around and it’s no longer maintained. Bill’s work at CFBData will more that suit our purpose: if you’re just getting into college football analytics, although ESPN’s public data is tragically inconsistent, his website is still the best resource for getting public data out there.

As to the second question, not all talent is created equal. We know that in the NFL, premium positions exist: your edge rushers, quarter backs, wide receivers, left tackles. In college, with a nod to Ian Boyd’s Space Force Theory, we have some ideas about what kind of personnel matters, but in general it’s a much more open question. The stylstic heterogeneity that so beleagures college football is exactly what makes it beautiful, but it also makes the idea of “premium positions” somewhat more fungible than in the NFL.

For the purposes of this post, I’ll grab the data from Bill’s collegefootballdata.com, I’ll try to recreate as closely as possible what I can with the team talent composite, and then I’ll try two alternate positional weighting schemes to see how they inform our understanding of who is accumulating what talent.

Gory Details

First, we’ll let the current NFL franchise tag numbers dictate our positional weighting. I’ll grab those from Over The Cap. We’l take those salary numbers and divide them by the total to get weights based on NFL compensation. This is far from perfect, but it’s an informative and useful starting point as we welcome the advent of revenue share, roster management, and ultimately salary caps into college athletics.

The second weighting scheme we’ll derive with a little power rating approach, using the weights from my “Vegas Style” unit power ratings, in line with what a lot of sharp folks in the industry use. I’ll again divide those by the whole, and get the weights.

Here they are:

Now, as for the recreation of the initial team talent composite. I grabbed the rosters from collegefootballdata.com, I joined them to recruiting rankings since 2016, and then sorted by recruiting ranking and applied the weighting and summed. Here are the results, compared to the actual values:

That’s not bad! Using a public scraper and a couple lines of code, I’ve got something that has a correlation coefficient of .965 to 247’s team talent composite.

The top teams in 2024 are:

Alabama 1006, Ohio State 1006, Georgia 1005, Texas 1000, Oregon 977, with composite top 5 finisher Clemson at 6, with 995 points.

For a few minutes on a Friday night, I’m pretty good with that! Let’s go ahead and apply our NFL weights and our power ratings weights and see what we find.

We’ll keep this to looking at the top 5 and how they shuffle, as this is just an introductory note:

Franchise Tag Weights:

Georgia 742, Ohio State 709, Texas 682, Alabama 641, Clemson 621

Vegas Power Rating Weights:

Georgia 767, Ohio State 730, Texas 715, Alabama 645, Clemson 641

Interesting! In both schemes, we see Alabama not only lose the lead among this year’s top 5 teams, but in fact fall to 4th. Why might that be? Well, in the franchise tag rankings, the top 5 positions are QB, LB, WR, DT, and DE. Let’s see how Alabama stacks up in those positions (again noting that our data will not be 100% perfect and complete):

I think the first thing we can say is that the LB/DE divide is not well captured in the recruiting data, just looking at the numbers. That’s no worry, it’s a known unknown and can be fixed in future analysis. Note also Alabama has 17 5 stars on their roster, according to 247 Sports, but we’re listing 15 here. Again, a problem, and might explain some of the differences, but not solely responsible.

Let’s set aside data concerns for a few minutes and discuss this as an exercise in roster building. To the extent that a franchise tag weight or a power rating weight captures the position groups most important to winning, we see some interesting concentrations of talent that drive the dip from Alabama from 1st to 4th. First of all, what position group do you not see listed here? Defensive back. Alabama is loaded at that position, but here’s where the positional weight shines. While there are certainly concerns about weak link systems and the secondary, this method suggests diminishing marginal returns to wins of concentrating talent in the secondary.

Secondly, look at the OL build in the 4 star range. Alabama has 8 4 star offensive linemen, whereas Georgia and Ohio State both have over 10. That alone might be a driver of the discepancy, as those teams are deeper along the offensive line. Finally, we can explain a little bit of Ohio State’s jump by the difference in wide receiver talent. Sitting with 4 five stars on the roster, the Buckeyes stand alone in attracting elite talent at the impact position of wide receiver.

Conclusion

So, what did we accomplish here? Well, one, burned the time until the Syracuse and Stanford game started. Additionally, I came up with a pretty decent recreation of 247’s Team Talent Composite using public data. Finally, we applied alternative weights to explore how positional importance can help us understand roster construction and give us a more refined idea about how talent might translate to wins.

Future research into this subject might consider experience and recruiting rating as well as positional importance, to devise an idea of the “bulk” of talent and where a team is in its development cycle. That is extremely important for coaches and general managers to plan around when to add from the portal and try to put together a truly special run.

Conclusions from this initial piece should be held lightly, but I hope you enjoyed a little musing on some potential strategies for roster construction.