Overview
JeffSelfSports ratings are designed to estimate true team and player quality while minimizing noise from randomness, sequencing, and schedule imbalance.
Depending on the sport, different statistical signals are more reliable. As a result, JeffSelfSports uses two complementary modeling approaches:
- Elo-style iterative ratings for game-driven sports
- Run- and production-based models for baseball
Both approaches share the same goal:
isolate underlying strength rather than relying solely on wins, losses, or raw totals.
Elo-Style Power Ratings (Football, Basketball, Hockey)
For football, basketball, and hockey, JeffSelfSports uses an iterative Elo-style rating system adapted for team sports with varying scoring patterns.
Rather than treating games as simple win/loss events, the model evaluates relative performance, accounting for score margin, opponent strength, and sport-specific scoring dynamics.
Sport-Specific Scoring Adjustment
Each sport applies a scoring adjustment function to reduce the impact of extreme results while still rewarding strong performance.
Raw scores are transformed using:
- A quadratic dampening function to limit runaway margins
- A non-linear scaling step to normalize scoring ranges
This ensures that dominant performances matter, but teams are not rewarded for running up the score against weaker opponents.
Different sports use different parameters:
- Football: Touchdown-based scoring with moderate totals
- Basketball: High-frequency scoring with large point totals
- Hockey: Low-scoring, goal-based outcomes
These sport-specific factors prevent score inflation and allow results to be compared consistently across leagues.
Game-Level Performance Ratio
Each game produces a performance ratio representing how well one team performed relative to the other.
The ratio incorporates:
- Adjusted scores for both teams
- Expected performance based on current ratings
- Game outcome (win, loss, or tie)
Hockey-Specific Handling
Hockey includes additional logic:
- Overtime and shootout losses receive partial credit
- Regulation wins are valued more than OT/SO outcomes
This mirrors the structure of hockey competition without directly copying standings rules.
Iterative Rating Convergence
All teams begin with the same baseline rating.
The system then:
- Processes all games
- Compares expected vs actual results
- Adjusts team ratings incrementally
- Repeats until changes fall below a convergence threshold
This allows ratings to stabilize naturally rather than being forced by fixed weights or arbitrary adjustments.
Strength of Schedule (SoS)
Strength of Schedule (SoS) is calculated as the average rating of all opponents faced.
SoS provides context when comparing teams with similar records or ratings.
A strong rating against difficult competition is more meaningful than the same rating achieved against weaker opponents.
Baseball Rankings Methodology (MLB)
Baseball differs fundamentally from other sports. Single-game outcomes are highly variable, and win–loss records can be heavily influenced by sequencing, bullpen usage, and one-run variance.
For this reason, JeffSelfSports uses run-based and production-based models for baseball rather than an Elo-style game iteration.
MLB Team Rankings (Pythagorean / Bill James Model)
MLB team rankings are based on the long-established principle that runs scored and runs allowed are the most stable indicators of team quality.
Inputs
For each team:
- Runs Scored
- Runs Allowed
- Wins and Losses
From these, total games played are calculated.
Run Environment Adjustment
An exponent is computed based on the season’s scoring environment:
Exponent = ((Runs + Runs Allowed) / Games) ^ 0.287
This adjusts the relationship between run differential and winning percentage to reflect league-wide scoring conditions.
Expected Wins (Bill James / Pythagorean Expectation)
Expected win percentage is calculated as:
Runs^Exponent / (Runs^Exponent + RunsAllowed^Exponent)
Expected wins are then derived by multiplying expected win percentage by total games played.
This produces a measure of how many games a team should have won based on run differential alone.
Final Team Rating
To balance actual results with underlying performance, the final team rating blends:
- Actual win percentage
- Expected win percentage
Final Rating = (Actual Win% + Expected Win%) × 100
This approach:
- Rewards teams for winning games
- Anchors rankings in run-based performance
- Reduces noise from close-game variance
MLB Player Rankings (TOP – Total Offensive Performance)
Individual batting rankings are based on Total Offensive Performance (TOP), a composite rate statistic.
Total Bases Accumulated (TBA)
TBA incorporates:
- Hits and extra-base hits
- Walks and hit-by-pitches
- Baserunning value (stolen bases and caught stealing)
TOP Calculation
TOP is computed as:
TOP = Total Bases Accumulated / Plate Appearances
This rewards players who generate offensive value efficiently per opportunity rather than through raw volume alone.
Qualification Rules
Different reports apply different thresholds:
- Unfiltered: All players with at least one plate appearance
- Filtered (in-season): Minimum PA = 3.1 × team games played
- Final: Minimum PA ≥ 502 (official MLB qualification)
MLB Pitching Rankings (Batting Against TOP)
Pitcher rankings evaluate offensive production allowed rather than traditional pitching stats.
The same TOP framework is applied using opponents’ batting lines.
Key interpretation:
- Lower TOP allowed = better pitching performance
Final-season reports are split into workload tiers to ensure fair comparisons across different usage levels.
Limitations
No model captures everything.
These ratings do not directly account for:
- Injuries or lineup changes
- Home-field or rest advantages
- Matchup-specific dynamics
They are intended as structured analytical tools rather than prediction engines.
Closing Notes
JeffSelfSports ratings are designed to be:
- Transparent
- Statistically grounded
- Consistent across sports
Different sports require different approaches, but all rankings aim to answer the same question:
How strong is this team or player, independent of short-term noise?