Roster turnover and next season's rating
By George Boyle · Updated Nov 9, 2026 · 6 min read
Every team's preseason rating starts from last season's final number, pulled most of the way back toward the Division I average. The pull is the same whether a team returns its whole roster or loses it to the NBA, graduation and the transfer portal. This study tests whether knowing who left and who arrived does better, on a season the test never saw.
- Effect size (production, pooled)
- +3.4 vs. −1.6 pts
- 80%+ returning vs. under 25% returning
- Verdict
- Real, not yet precise enough
- needs more seasons before it can be used
- Sample
- 1,446 team-seasons
- 4 season pairs, 2022-23 through 2025-26
- Unseen season
- 2025-26 only
- 364 team-seasons, never used to fit anything
The question
The current preseason rating is simple on purpose: take a team's final rating from last season and pull it 80% of the way back toward the Division I average, whatever happened to the roster over the summer.
This study tests richer versions against that flat baseline: one that also knows what share of last season's minutes is still on the roster, one that knows what share of last season's production is still there, and one that also credits transfers coming in.
This is a measurement, not a model change. The preseason build stayed exactly as it was while this ran.
Does it beat the flat baseline on a season it never saw?
On the seasons used for fitting, adding roster information keeps improving the fit. The richest version explains noticeably more than the flat baseline.
On the one unseen season, every richer version does worse than the flat baseline, and worse than a plain single-number refit of it.
| Model | Error, unseen season (pts) | Error, fitting seasons (pts) | R², fitting seasons |
|---|---|---|---|
| Flat 80% regression (no fitting at all) | 5.343 | — | — |
| Same idea, refit by regression | 5.352 | 4.975 | 0.560 |
| + returning minutes share | 5.483 | 4.851 | 0.582 |
| + returning production share | 5.457 | 4.791 | 0.592 |
| + returning production + transfers in | 5.378 | 4.611 | 0.622 |
Every richer model fits the seasons it was built on better and predicts the unseen season worse than doing nothing extra at all. That gap is not a bug. It is the whole finding.
The relationship is real; three seasons don't pin it down
Refitting the same two-term model (last season's rating plus returning minutes share) on the unseen season by itself keeps the same sign and stays statistically significant. This is not a chance correlation in the fitting seasons.
What changes is the size. The unseen season's own coefficient is roughly 58% as large as the one fit on the three earlier seasons.
| Fit on | Coefficient | Standard error | p-value |
|---|---|---|---|
| Fitting seasons, 2022-23 through 2024-25 | +7.338 | 0.963 | < 0.001 |
| Unseen season alone, 2025-26 (for comparison only) | +4.288 | 2.055 | 0.038 |
What returning production is worth, in the data
Group every team-season by how much of last year's production is still on the roster, pooled across all four season pairs, and the pattern is clean and rises almost step by step.
Teams that returned almost nothing got worse the next season, on average. Teams that returned almost everything got better.
Plugging the larger coefficient from the fitting seasons into the unseen season overcorrects for teams that returned the most and undercorrects for teams that lost the most. That is exactly why the models above score worse there.
The relationship isn't fake. Three fitting seasons, one of which (2022-23) still carries pandemic-era roster noise, don't yet pin its size down precisely enough to use as a correction.
| Returning production | Teams | Avg. rating change next season |
|---|---|---|
| Under 25% | 409 | −1.60 pts AdjEM |
| 25–35% | 226 | −0.24 pts |
| 35–50% | 385 | −0.26 pts |
| 50–65% | 262 | +1.67 pts |
| 65–80% | 117 | +2.45 pts |
| 80% or more | 43 | +3.38 pts |
Why preseason ratings don't adjust for this yet
The verdict: the signal is real and points the same way everywhere it was checked, but three fitting seasons are not enough to ship it as a correction to the live preseason rating.
The plan is to re-check once a fifth and sixth season are available. Any change would need its own test, written down in advance, before it touched a live rating, the same as every other model change on this site.
Where it is safe to use today is as context, not a correction: which players a team lost and gained, shown on the Ratings and Matchups pages and framed as "here's what has historically happened to teams like this", not "we adjusted the number for it".
Method
how this was measured
1,446 team-seasons across four consecutive-year pairs, where the team was Division I in both seasons of the pair.
"Returning" is measured from the next season's own box scores. That is a mild look ahead a backtest can't avoid, since it needs to know who actually played, not who was listed in the preseason. A live preseason build uses the current roster, which is noisier this early because not every transfer is official yet.
Production is a simple weighted box-score value: points, plus half of rebounds, plus assists, steals and blocks, minus turnovers and missed shots.
Pooling across four pairs counts many of the same programs more than once, so the effective sample behind the group table is smaller than the raw team-season count suggests. That is a reason for caution, not a reason to distrust the direction of the pattern.
See it live
Written by George Boyle, who builds The Sport Stack. Corrections and questions: hello@thesportstack.io. Who runs this.