What Does Win Probability Mean?
A win probability of 75% does not mean the favorite will win 75% of the time in this specific game. Baseball games are single events: there is no 75% outcome. Instead, win probability represents a calibrated estimate. If you could observe 100 games where both teams have a 75% vs. 25% projected probability, the 75% team should win approximately 75 times. Probability is about long-run frequency, not individual outcomes.
This distinction matters. A 75% favorite can lose tonight. Probability does not guarantee results; it measures confidence relative to a large sample. The higher the probability assigned to a team, the more statistical edge our model found. But edge does not guarantee victory.
The Nine Inputs: How StatScope Builds Confidence
StatScope's win-probability model (v2.2) blends nine factors, each weighted by its historical predictive power. First, Pythagorean expectation: a team's run differential, transformed into expected win percentage using the formula (Runs^1.83) / (Runs^1.83 + AllowedRuns^1.83). This provides the baseline.
Second, starter FIP (Fielding Independent Pitching). FIP predicts future ERA better than past ERA, so we weight the starting pitchers' FIP more heavily than their ERA. Third, bullpen ERA: a team's relief staff performance. Fourth, lineup wOBA (Weighted On-Base Average): how effectively the batting order reaches base and advance runners. Fifth, recent form: last-10-game record and run differential, blended with season average.
Sixth, Log5 method: Bill James' formula for head-to-head matchup probability, combining season win percentages in a way that corrects for quality. Seventh, home-field advantage: +6.6 percentage points for the home team, consistent with historical data. Eighth, park factor: whether the venue inflates or suppresses run scoring, weighted by the home team's run-production advantage. Ninth, regression to mean: a 22% pull toward 50%, reflecting baseball's inherent unpredictability.
Interpreting the Numbers: High, Medium, Low Confidence
The model outputs a win probability (say, 72%), but also a confidence badge. High confidence (probability ≥ 70%) typically means both starters have thrown ≥ 30 innings, both teams have ≥ 40 games played, and underlying stats (team ERA, wOBA) are stable. Moderate confidence (60–70%) suggests some stability but smaller sample sizes or pitcher workload concerns. Low confidence (< 60%) indicates thin data or one-sided starting pitcher advantage.
A 75% pick with high confidence should make you more comfortable than a 65% pick with low confidence. The probability is the model's best estimate, but the confidence badge reflects how much you should trust that estimate.
What the Model Cannot Do
The model operates on season-to-date data: ERA, wOBA, recent form, and pitcher workload. It does not account for weather, field conditions, or day-of-game roster scratches. It does not see lineup announcement; it uses the team's batting order as currently constructed. It cannot predict an unexpected bullpen meltdown or a surprise hot streak. These are inherent limitations of any pre-game model.
The model is best used as a starting point. Use it to identify mismatches (games where the favorite is undervalued or the underdog is overvalued according to betting markets), but combine it with your own research and the eye test. Model output + human judgment = better decisions than either alone.
How to Verify the Model Works: The /track Page
Every pick StatScope posts publicly is logged on the /track page, along with actual results, moneyline odds, and whether the pick won. You can see the cumulative record (wins-losses), win rate percentage, ROI (return on a flat $100-per-pick betting strategy), and a calibration curve.
The calibration curve plots our predicted win probability (x-axis) versus actual win rate (y-axis). If the model is calibrated, the curve should follow a diagonal line: picks we said were 60% likely should win ~60%, picks we said were 80% should win ~80%. Deviations indicate the model is over- or under-confident. This transparency allows you to judge for yourself whether the model is trustworthy.
Final Thoughts
Win probability models are tools, not crystal balls. They summarize complex information into a single number, but they cannot eliminate baseball's fundamental randomness. A 70% favorite loses roughly 30% of the time, and that is not a failure—it is expected. Use the model to find edges and make informed decisions, but always respect the uncertainty inherent in sports.