Cohesion Impact Analysis
Multi-level regression measuring cohesion's explanatory power beyond talent alone
12 seasons (2035-2049) | 53,000+ lineups | Updated 2026-03-19 09:02
Regression Results
Multi-level regression: Talent Only R2 vs Talent + Cohesion R2. Delta R2 = cohesion's unique explanatory power.
| Level | N Lineups | Talent R2 | +Cohesion R2 | Delta R2 | Coefficient | p-value |
|---|---|---|---|---|---|---|
Top Cohesion Pairs
Player pairs with the largest performance bonus beyond what talent predicts. Minimum 50 shared minutes.
| # | Players | Min Together | Net Rtg | Expected | Cohesion Bonus | Seasons |
|---|---|---|---|---|---|---|
Team Continuity vs Performance
Roster continuity (% of top-8 minute players retained year-over-year) correlates with net rating. r = , p = , n = team-seasons.
Denver Dragons History
| Season | Continuity | Retained | Net Rtg |
|---|---|---|---|
League-Wide Pattern
Teams retaining 75%+ of their top-8 minute players average net rating, vs for teams retaining <50%. Continuity alone explains ~6% of net rating variance (r=0.245).
Methodology
Data: 12 seasons of JBL lineup data (2035-2049, excluding 2034/2038/2045 due to missing data). 53,000+ lineups with 50+ shared minutes.
Cohesion metric: Average seasons played together for all player pairs in a lineup. Computed from historical roster data across 12 seasons.
Model: OLS regression. Stage 1: Net Rating ~ avg talent (CA). Stage 2: Net Rating ~ avg talent + cohesion. Delta R2 = cohesion's unique contribution.
Interpretation: A Delta R2 of 4-5% means cohesion explains 4-5% of net rating variance beyond what talent alone predicts. All results significant at p<0.001.
CJBL: College data shows no significant cohesion signal (class year is too weak a proxy for actual minutes played together).