← Home
Methodology
How Ref Watch computes referee analytics, ranks findings, and labels data confidence. Descriptive historical tendencies, not predictions or betting advice.
- We measure historical tendencies, not outcomes.
- Officials influence game environment, not guaranteed results.
- Some signals are meaningful; some are not.
- We show uncertainty and sample size.
- We do not provide betting advice.
What we measure
- Ref profiles and rankings default to officials with 30+ games. Below-gate refs are hidden or dimmed; treat as directional only. Minimum games required before we show a confident stat. Below the threshold, the value may still appear but is marked as below gate.Minimum games required before we show a confident stat. Below the threshold, the value may still appear but is marked as below gate. badges flag thin samples on slate cards and profiles.
- Team–ref splits need 8+ games per pairing; crew anomalies need 12+. ATS and O/U splits need 30+ decisive games with closing lines.
- Average fouls on one team minus the opponent per game. Positive means that team is whistled less (or the opponent more).Average fouls on one team minus the opponent per game. Positive means that team is whistled less (or the opponent more).: team whistle volume (fouls, flags, minors) in games a ref worked (crew-level correlation, not fouls charged to that ref alone). W-L splits use the same per-ref game sample.
- How many points above or below league average this crew’s games tend to score, a pace signal, not a betting pick.How many points above or below league average this crew’s games tend to score, a pace signal, not a betting pick.: crew average combined score minus league baseline (222 NBA / 6.3 NHL goals). Home/road bias uses win and foul splits, not ATS.
- Tonight's edges require qualified refs (2+ at 30+ games) before alerting.
- Tendency index covers referees; linesmen appear on crews but are excluded from penalty/scoring analytics.
Data & confidence
- NBA: 2016-17, 2017-18, 2018-19, 2019-20, 2020-21, 2021-22, 2022-23, 2023-24, 2024-25, 2025-26 (2016-10-25 – 2026-04-12). NHL: (). League trends from up to ten seasons of game-log aggregates in
data/baselines.json. - When The spread and total sportsbooks posted near tip-off. ATS and O/U tables here use that number, not our fixed 225 benchmark.The spread and total sportsbooks posted near tip-off. ATS and O/U tables here use that number, not our fixed 225 benchmark. are unavailable, we use league averages from game logs (222 NBA / 6.3 NHL) as over-rate proxies.
- Some ATS/O/U splits use league-average benchmarks where sportsbook data is missing. This number is not computed from real game logs yet; it uses a fallback constant (e.g. league baseline or odds benchmark) until we have enough real game data.This number is not computed from real game logs yet; it uses a fallback constant (e.g. league baseline or odds benchmark) until we have enough real game data. markers flag partial samples only.
- Historical line data is unavailable for some games; league-average benchmarks and provenance markers apply where noted.
Findings & limits
- Findings are scored as effect size × √sample size, weighted by sample depth. Category deduplication keeps the hub diverse.
- Language stays descriptive: “historical tendency,” “over rate,” “foul edge”, never picks or locks. Browse the full index on the insights hub.
- Patterns from past games do not predict future results. Ref Watch is independent research, not affiliated with leagues or sportsbooks, and not betting advice.