How it works
Methodology
The Kolesky Index combines three complementary rating systems into a single transparent, reproducible composite score.
Elo Rating System
A zero-sum system that adjusts after every match. Teams gain points for wins and lose points for losses, scaled by opponent strength and match importance.
R′ = R + K × (S − E)
E = 1 / (1 + 10(Rb−Ra)/400)
K = 32 × SoS multiplier
E = 1 / (1 + 10(Rb−Ra)/400)
K = 32 × SoS multiplier
Index Weight50%
Massey Rating
A linear algebra approach that solves a system of equations over the full schedule graph. Accounts for margin, score differentials, and offensive/defensive decomposition.
M·r = p
M = match matrix
r = ratings vector
p = point differential vector
M = match matrix
r = ratings vector
p = point differential vector
Index Weight30%
Performance Index
A composite of nine soccer metrics — goals, goals conceded, shots on goal, save %, assists, shooting accuracy, corners, and discipline — z-score normalized across all D-I teams.
PI = 0.4·GD/90 + 0.3·SoG/90
+ 0.2·xGD + 0.1·CS%
Normalized: z-score within cohort
+ 0.2·xGD + 0.1·CS%
Normalized: z-score within cohort
Index Weight20%
Strength of Schedule
A recursive measure — opponents' ratings weighted by recency and home/away context. Used as a multiplier in the Elo K-factor and as a blending modifier across all pillars.
SoS = Σ(opp_rating × w_i) / Σw_i
w_i = recency decay × venue_factor
Range: 0.80 – 1.30
w_i = recency decay × venue_factor
Range: 0.80 – 1.30
Index WeightModifier
Player Impact Score
Each player's score uses a role-based weighted scorecard on a 0–100 scale. Three sub-scores are blended: Strength of Minutes (SoM) weights playing time against opponent quality; Strength of Impact (SoI) weights goal/assist/save contributions against the opposition quality faced; and Credibility (Cred) provides a Bayesian confidence stabilizer for small sample sizes.
Player Score = 0.30·SoM + 0.45·SoI + 0.25·Cred
SoM = Σ(min_i × opp_elo_i) / total_min
SoI = Σ(contribution_i × opponent_quality_i) / n
Cred = Bayesian shrinkage toward positional mean (prior N=8)
SoM = Σ(min_i × opp_elo_i) / total_min
SoI = Σ(contribution_i × opponent_quality_i) / n
Cred = Bayesian shrinkage toward positional mean (prior N=8)
Composite Blending
Each pillar is z-score normalized across all teams, then combined with weights of 50% Elo, 30% Massey, and 20% Performance Index. Weights were determined by backtesting against historical NCAA tournament outcomes to maximize predictive accuracy. Preseason ratings carry forward from prior years with regression to mean, modified by the Roster Capital Index.
FAQ
How often are rankings updated?
Rankings are recalculated nightly during the season after all results are entered. The "Live" badge reflects data current as of last night's update.
Why doesn't my team's record match their rank?
The Kolesky Index is schedule-adjusted. A 10–5 team that beat three ranked opponents may rank higher than a 13–2 team in a weak conference.
How are GKs rated differently?
Goalkeeper SoI uses saves-per-90 and clean-sheet percentage weighted against shot quality faced (xG against), rather than the goal+assist model used for field players.
What does Credibility mean?
A sample-size confidence indicator ranging from 0 to 1 based on minutes played. Higher values mean the player has a larger body of work, making their production stats more trustworthy.
References
- Kovalchik (2020) — Extension of the Elo rating system to margin of victory
- Tang, Wang & Jin (2025) — Is Elo Rating Reliable?
- Massey (1997) — Statistical Models Applied to the Rating of Sports Teams
- Hvattum & Arntzen (2010) — Using ELO ratings for match result prediction
- Andrzejewski et al. (2022) — Analysis of team success based on match performance
- Lago-Ballesteros & Lago-Peñas (2010) — Performance in Team Sports
- Ball, Huynh & Varley (2025) — Comparing player rating systems