Last week I released an article where I detailed an analytical approach to scouting freshman Wide Receivers. This is the Running Back edition.
I used a similar method that I used for my Wide Receiver freshman model. I created multiple different models which aimed to predict College Football Fantasy Points Per game using data sourced by local legend @bigWRguy. The model scores were then averaged on the same scale to finish with each player’s final score. Not all recruits from the past eight years are in the database,. Information on high school players can be difficult to access and time-consuming to compile. However, there is enough to be helpful in trying to make future predictions.
Data was split into a training set comprising of 70% of the data, and the remaining 30% was used to test the model. Only players who had a minimum of three seasons of College football played were included in the model creation process. This meant that recruiting classes only up to 2020 were used (except 2015 as missing data).
Five themes of variables were used across the three models:
- Recruiting grades
- Peak Miles Per Hour on tape
- Height and weight
- High School Production
- Strength of schedule
Where a variable was missing for a player, then they were allocated the average for that variable.
The R-squared correlation for the whole dataset (training + testing) from 2014-2020 (sans 2015) was 0.31. This means that the model score was able to explain 31% of the variation in College PPG. Although this is still very solid, it was below Wide Receivers, and overall I just found Running Backs more difficult to model. I think part of this subjective feeling, though, was my tendency to feel uncomfortable if a Running Back with a lower rated composite rating scored a high score. That comes from an unfair expectation in my head that College PPG = Draft Capital. I was training the model to the College PPG, and realistically, three stars do produce College PPG too! Just not always at high-profile programs, which limits their translation into the NFL, but again, draft capital wasn’t the aim, so I now feel better about the results. This is compared to the 247 Composite rating, which only has an R-squared value of 0.04 to College PPG, which is peanuts. Both the 2021 and 2022 classes so far have tested at an R-squared value of 0.16-0.18.
When compared by the average of the top 20 model scores per year, this year is projected to be the strongest year of the past eight years. Followed by 2017, then 2022, with 2018 being the weakest.
All-Time Model Scores
The Top 20 all-time of the model, similar to the Wide Receivers, has some hits and misses. Still, from a College Fantasy perspective, you would’ve been pretty happy. At the very least much happier than if you simply sorted by 247 Composite Rating (albeit, College Fantasy points isn’t 247’s goal, my understanding is that their goal is trying to project draft capital).
I thought it could also be fun to list out the top Running Backs who were below 0.9100 in the 247 Composite. I’ve used this number as a rough threshold for the Top 250 recruits of each Class. In this range, we’re definitely not expecting to feel confident in our chances of picking a player who is going to get strong draft capital, as the hit rates are very low. We obviously have two huge names near the top in Breece Hall and Jonathan Taylor. However, we’ve also go some other fun college producers on this list, such as Mohamed Ibrahim, Travis Dye, CJ Verdell, and Samaje Perine. If DJ Oliver even has half the college career of Hall or Taylor, then considering the cost to draft DJ Oliver in supplemental drafts, then you’ll be doing cartwheels.
Model Scores for the 2023 Class
0.9100″ width=”685″ height=”1024″>
Justice Haynes is the runaway RB1 for the 2023 iteration of this model. With Jahmyr Gibbs headed to the NFL and only really Jase McClellan ahead of him, I could see Haynes very quickly crushing for Alabama. Robinson, Owens, and Jackson are all well-liked by Campus2Canton. This model has Cedric Baxter as quite a large fade, considering he’s Campus2Canton’s consensus RB1 and also the RB1 for On3 and 247. Baxter’s High School Production was low, especially his receiving market share, and his efficiency wasn’t anything special. It’s important to remember that this model (like all models) isn’t infallible. It will get things regularly wrong; however, so do rankings. It will be interesting to see how the next few years play out for Cedric Baxter. At the very least, I’ll be drafting Haynes and Owens over him. I think I’ll simply shift to another position as opposed to making the decision between Robinson and Baxter.
In the top 250 recruits, the model has some wild discrepancies compared to the Campus2Canton consensus. Names such as Quinten Joyner (C2C RB24), Trey Holly (C2C RB40) and Dylan Edwards (C2C RB18), and Mark Fletcher (C2C RB33) are the possible options for where the model is going to make its money in this class. Fletcher is especially interesting, as On3 has him RB8. Trey Holly is small (5-7, 177-189, depending on the source), which could very well limit his NFL upside, but shouldn’t be a barrier to College production.
When limited to players who score lower than 0.9100 on the Composite (mostly three stars or lower), my favorite name by far is DJ Oliver, who I referenced earlier on. In addition to a top 3 all-time score for a <0.9100 player, he’s also 237lbs. and LeGarrette Blount’s nephew. Campus2Canton has him ranked RB36. The Official ran their 3-star Running Back show last week and also referenced the rest of the model’s Top 5 in this category: Jamarion Wilcox, Keyjuan Brown, Jordan Louie, and Keith Willis Jr, plus Dawson Pendergrass a little further down. All really great darts to throw at cheap prices in supplemental drafts.
I hope you enjoyed both this article and the Wide Receiver article and they help you in your supplemental drafts. Or if they simply help you enjoy following some recruits through college.
For those looking for Quarterback and Tight End models, unfortunately, I have to apologize. I won’t be pursuing any for the simple reason of they’re too difficult.
As always, tag me in the Campus2Canton discord, or reach out to me on Twitter if there are any other recruits you’d like to know some model scores for, and I’ll be happy to tell you them. I wish I could post them all here but a slab of 53 different players looks a bit much.