A post on X that I made this week gained a lot of attention, so I thought it worth a follow-up with a longer explanation here.
The thread is here:
I built what’s called a neural network (explanation to follow). I used the 107 Division I transfers that Arkansas was known to have contacted during last offseason, as compiled by Ben Brandon. The Hogs ultimately signed seven of these players, while the other 100 went to different schools.
The neural network used 2023 stats and 2024 results for the 100 non-Hogs to look for patterns to see if could predict how Arkansas’ seven signees would do. A neural network is a type of classification model, meaning that it tries to put single observations (in this case, individual transfers) into different groups, or classes. The four classes I chose were “Superstar”, “Starter”, “Contributor”, and “Bench”, and they were based on the players’ xRAPM, which is the best single measure of how good a player is.
After the model was “trained” on the 100 non-Hogs, I applied to the seven Hog signees. The important note is that the model was not given any information about how the seven Arkansas signees did this season. All it knew was what they did last year at their previous school. Here were its predictions:
That’s pretty spot-on! You could say Tramon Mark slightly underachieved and Khalif Battle slightly overachieved, but the real issue was that Arkansas signed five players that the model believed would never contribute at all. And they haven’t made the team better. In fact, relying on them for any contributions at all is the reason Arkansas had such a bad year.
So what happened? It’s possible that Eric Musselman and staff whiffed on targets they really wanted and had to “settle” for most of this class. We know they really wanted Ron Holland, who went to the G League. Maybe they really wanted a better point guard like Tylor Perry or Aaron Estrada, and ending up with El Ellis and Keyon Menifield was not the plan.
But it’s worth noting that a lot of these guys signed pretty early in the portal process, so I don’t think Arkansas was left with the scraps. Holland aside, I think this was a misevaluation from Muss and staff.
The neural net was overkill
I did the neural net classification model because I could, but it really wasn’t necessary. The best predictor of a player’s future xRAPM is their past xRAPM. You don’t need a complicated model to tell you that. You can just look at past xRAPM and get an idea of how good a player is.
Per xRAPM, here were the best players on Arkansas’ contact list:
- Dame Adelekun +14.2
- Tylor Perry +12.5
- Hunter Dickinson +11.1
- Aaron Estrada +11.0
Adelekun is a bit of a diamond in the rough, as the former Dartmouth forward was not highly recruited in the portal. He signed with Loyola Chicago and just led the Ramblers to a regular-season conference title despite his xRAPM this year falling all the way to +7.4 (that’s still better than any Hog). The other three were known to be elite portal prospects, and – surprise! – they’ve all been excellent at their new schools.
According to xRAPM, South Carolina signed two players better than any Hog signee in BJ Mack and Ta’Lon Cooper (both contacted by Arkansas). Our model expected South Carolina to break out this season and they did. But their players didn’t really get better for the most part: Mack’s xRAPM was +4.5 last year and +4.7 this year. The Gamecocks just signed the right players.
Arkansas… didn’t. The Hogs’ transfers didn’t disappoint so much as they weren’t really that good to start out.
I left out Denijay Harris, who didn’t play enough minutes to get an xRAPM score (so he’s at 0). These six signees had an average xRAPM of +1.8 at their previous schools. Compare that to +3.1 for South Carolina and +5.1 for Alabama, then check the final SEC standings. Arkansas and Alabama both had their rosters cleaned out after last season, but Nate Oats signed good players in the portal and Eric Musselman signed mediocre ones.
So what was the plan?
I wrote about my concerns with this team way back in the season preview, and the Hogs were never able to address this issue:
The key is going to be if the offensive creators can also defend. The numbers are skeptical about the defense of Ellis, Battle, and Davenport. Theoretically, this could either be Eric Musselman’s best offensive or his best defensive team, but probably not both.
…
You can only play five guys at once. Arkansas’ depth and overall roster flexibility is incredible, but can they consistently put five guys on the floor who can score and defend at the same time? That’s going to be the question.
Battle, Ellis, Davenport, and Menifield all graded as good or very good offensive players and bad or very bad defenders. I figured that was intentional after the offense struggled in 2023:
One big theme of Arkansas’ offseason adds is offensive pop. Ellis, Battle, and Davenport all grade as poor defenders but elite scorers. My theory on Arkansas’ transfer strategy is basically this:
-Defense is easier to “coach up” than offense
-The raw ability to create shots is hard to coach, and thus needs to be recruited
Guys can be taught a lot, but the instincts required to be a true three-level scorer who can be decisive with the ball in hand is hard to coach.
I think Muss overinvested in individual scorers and sacrificed rebounding and defense. I think he thought he could coach up bad defenders. Unfortunately, he thought wrong:
Muss’s offensive scheme has some issues, namely that it is so isolation-heavy that it can only work for guys with the ball in their hands and can create off the dribble. That means that guys who got the keys to the offense (Mark, Battle) were awesome, but the guys who didn’t looked bad, because they had no other way to contribute. Outside of Mark and Battle in isolation, the only other offense was Makhi Mitchell on pick-and-rolls. There were no pass-first guys, no rim-runners, no catch-and-shoot snipers, and no offensive rebound gurus.
But while the offense was a modest disappointment, this was a tournament team if the Hogs had the standard Muss defense. Instead, we got the worst Muss defense in his five years by a very large margin. But the problem wasn’t that these guys regressed; it’s that they weren’t good defenders to begin with.
The two-part failure
This year almost seems like an overcorrection from last year’s “let’s load up on the most raw talent we can recruit” strategy. That team had guys with superstar potential (Anthony Black, Jordan Walsh, Nick Smith, Ricky Council) but lacked specialty role players. By focusing on guys with more developed (but more narrow) skillsets, the Hogs couldn’t out-talent opponents like they had in years past.
To be clear, a lot of these transfers that were signed for specialty roles were supposed to be supporting the returners, but those all regressed badly:
I don’t know if Brazile comes back or not, but this has been two years that the Hogs planned to build around him and it hasn’t worked out. And it’s (mostly) not his fault: he got hurt last year, and I suspect that the lingering effects of an ACL tear have ensured that he’s not the same player he was pre-injury. If the plan was to have him be the stud, averaging 12 & 7 a game, spacing the floor on offense and erasing stuff around the rim on defense, then some of the “transfers as role players” stuff makes sense.
You have to start with at least one superstar and at least two guys who are elite at both ends and can play 36+ minutes per game without ever being a liability. Muss likes having the one go-to guy who never comes out of the game: Mason Jones in 2020, Moses Moody in 2021, JD Notae in 2022, Anthony Black in 2023. But the Hogs didn’t have that guy in 2024, though Mark was close at times. I think it was supposed to be Brazile, so I’d say the first issue was Arkansas didn’t clearly establish its stars before it started pursuing role players. I do wonder if Holland and Grant Nelson were both scared off by the prospect of playing second fiddle to Brazile, and now in hindsight we see that either would be a massive upgrade. Then you get some good role players to create a balanced roster.
The second issue, then, is that Arkansas didn’t diversify its role players enough. None of the returnees or portal signees were good offensive rebounders last year, so this has been Muss’s worst offensive rebounding team. None of them were good at getting steals, so this team has struggled to force turnovers. And none of them were great passers, so this team doesn’t have many assists. I think they focused way too much on guys who can create shots off the dribble (guards) and guys who can clear defensive rebounds (forwards) and they’ve had issues in all other places.
So what’s the fix?
The two-part failure demands a two-part fix. First, make sure you have a star or two. The Hogs didn’t have that in 2024. Battle, if he returns, might could be a more low-end star on a very deep team, like Moody was in 2021. But at a minimum you need one more all-around great player. Could it be either of the 4-star freshmen? Moody, Anthony Black, Brandon Miller, and Reed Sheppard are among the many recent proofs that freshmen can be the stars you need. I don’t know if Jalen Shelley or Isaiah Elohim have the potential to get to that level; my gut would say no, so the Hogs need to win a big portal battle to land a big fish.
The second part is that role players have to help you win without shooting. Offensive rebounders and passers are both good to stock up on. And players targeted should not be terrible defenders. You might can coach them up a little bit (see Stanley Umude in 2022) but I don’t think you can create a good defender out of a bad one in one season.
We’ll keep a close eye on what Arkansas does this offseason. I expect a very different strategy in the portal.
Thanks for reading! Be sure to follow us on Twitter and on Facebook.
The latest from Fayette Villains, straight to your inbox
Enter your email to subscribe and receive new post alerts and other updates. You can unsubscribe at any time.