Understanding the model behind the predictions
StatIQ started as a personal project to apply rigorous quantitative analysis to college football. The goal is to go beyond traditional box score statistics and dig into what actually drives performance: efficiency metrics (PPA), situational splits, playcalling tendencies, halftime adjustments, homefield advantage, and how teams perform in close games.
The platform breaks down team and player data across meaningful efficiency contexts - not just raw season stats. Understanding how a team performs in specific situations, against certain opponent styles, or in critical moments is what separates deeper analysis from surface-level takes.
The prediction models synthesize this efficiency and situational data into quantitative forecasts, updated weekly throughout the season. They're designed to identify expected value discrepancies in the betting markets where quantitative model predictions may diverge from consensus.
Questions, feedback, or ideas? I'm always open to discussion and collaboration. Feel free to reach out: anthonycbrooks1@gmail.com.
The Quant models use Predicted Points Added (PPA) metrics—capturing offensive and defensive efficiency along with situational variables such as close game analysis and halftime adjustments. The model helps identify raw efficiency patterns that traditional rankings may miss.
The Quant Adj model expands on the same PPA efficiency foundation but incorporates strength-of-schedule context and season-long performance trends. The model provides a more stabilized view of team performance, helping distinguish overall team efficiency from week to week PPA volatility.
ESPN's comprehensive college football rating system that measures overall team quality independent of strength of schedule. Accounts for offensive efficiency, defensive efficiency, special teams, and margin of victory adjustments to produce a single rating for each team.
Measures how many points an offense adds or subtracts on each play relative to expectation, adjusted for down, distance, and field position. Positive values indicate the offense exceeded expected performance; negative values indicate underperformance.
Inverse of offensive PPA—measures points prevented per play. Negative defensive PPA indicates strong defense; positive defensive PPA indicates defensive weakness.
Splits offensive PPA by play type—rushing plays versus passing plays—to identify offensive strengths or weaknesses in specific situations.
Defensive PPA split by rushing and passing plays, revealing whether a defense struggles more against the run or pass.
Percentage of plays that achieve "success"—defined as gaining 50%+ of necessary yards on early downs (1st/2nd down) or 100%+ of necessary yards on late downs (3rd/4th down). Measures offensive consistency.
Frequency of big plays—defined as plays gaining 15%+ of total yards needed for a first down on early downs or 25%+ on late downs. Captures offensive ability to create sudden gains.
The mathematical advantage or disadvantage of a bet. Calculated as: (Probability of Win × Payout) − (Probability of Loss × Stake). Positive EV indicates a profitable bet over time; negative EV indicates an unprofitable bet.
Model's probability estimate for a predicted outcome, expressed as a percentage. Higher confidence indicates the model is more certain in its prediction based on feature values.