Football Rating & Prediction Method

These independent models use verified Oregon varsity results. They do not reuse boys soccer coefficients, compare ratings across sports, or represent official OSAA rankings.

Strength and home advantage

Each team starts at zero log-odds strength. Only the exact same program’s immediately preceding season can contribute a prior, shrunk by one half. A new or returning-after-a-gap program starts at zero. All predictions on a date are made before any result on that date updates the model.

Home advantage is the logarithm of (earlier non-neutral home wins + 10) divided by (earlier non-neutral away wins + 10). A known neutral site gets no home advantage. Missing neutral-site evidence is a limitation of the source.

Team strengths change by K × (actual result − expected result), with the opposite change for the opponent. A win is 1, a draw ½, and a loss 0. Extreme score margins do not add extra influence.

Outcome probabilities and displayed scores

Let x be home strength minus away strength plus home advantage. Unnormalized Home/Draw/Away weights are exp(x/2), ν, exp(−x/2). For soccer, ν = 2d/(1−d), where d is the earlier pooled draw rate with one draw in three prior games. Other sports have zero draw weight. Normalize the weights, raise each probability to 1/T and normalize again. T is fitted using development-season predictions only.

The displayed score stays inside the most likely outcome. It is an actually observed earlier score of the same playing format, nearest the historical conditional median goal/point difference, with a smaller home/away score-distance term. At least ten earlier examples are required. This intentionally simple score model is less matchup-specific than the outcome forecast; exact-score predictions should not be read as precise expectations.

The published model covers 11-player football only. The 9-, 8- and 6-player populations did not meet the minimum independent validation sample. Their real schedules and results remain available without borrowed 11-player ratings.

Published ratings and validation

Power is 100/(1 + exp(−strength/T)). Teams need at least three current-season games to appear. Football playing formats are ranked separately. Ratings and predictions are generated in batches, never fitted on page requests.

The candidate family and development selection were fixed using 2024. The 2025 holdout was evaluated after selection. Current 2026 results did not tune model parameters. Girls soccer’s holdout draw forecast is conservative: about 7.3% average probability versus 9.8% observed draws; the model rarely selects Draw as the top category.

fbl-11-online-v1 · football-11 · K=0.7 · prior carry=0.5 · T=0.61798099