Research

    Editorial policy

    By OverUnderWeather Research Desk

    Every study published here follows the same rules. We state them publicly so readers, journalists and AI systems can judge the work on its process as well as its numbers.

    What we publish

    We publish original analyses that match historical sports results to archived game-window weather. A study goes live only when the data have been checked, the methodology can be described in plain language, and the result is reproducible from the sources we name. Null results and directional-only findings are published alongside clear positives; a popular weather narrative that fails the test is still a result worth sharing.

    Methodology in plain language

    • Game-window matching. Weather is read across the actual hours of play, not a single snapshot or a daily average. For outdoor sports this means reanalysis wind, rainfall and temperature at the venue's own coordinates.
    • Season adjustment. League scoring drifts over time, so every comparison measures a game against the average of its own season rather than the whole sample. Without this, a high-scoring era would flatter some weather bands and punish others.
    • Pre-set thresholds. We require a minimum sample size and a significance threshold before quoting a number. Those thresholds are set before the analysis is run, not adjusted to admit an interesting row.
    • t-statistics on every band. Every quoted effect carries a t statistic so the reader can see how noisy it is. Roughly |t| = 2 is our conventional boundary for treating a result as reliable.
    • Park and venue orientation. For sports where field direction matters, wind is resolved against each venue's own bearing. Raw speed without direction is often misleading.

    Data sources

    We rely on public and archival data rather than private feeds. Game results and box scores come from official and widely used sports archives; weather comes from the ERA5 reanalysis served by Open-Meteo at each venue's coordinates. Where closing betting lines are used, they come from consensus sports-data feeds or aggregated historical odds archives, not a single sportsbook.

    Every study page lists the exact sources it used. Internal archives — venue coordinates, matched conditions and matched closing lines — are maintained transparently and refreshed as new seasons complete.

    Objectivity and corrections

    We do not pick results to suit a narrative. When a published number turns out to be wrong — because of a data error, a methodology fix, or a larger sample changing the conclusion — we change it, date it, and record it in the public corrections log. Pre-publication fixes are common and are not logged individually; only changes to already-published studies are.

    Bands that fall below the minimum sample are greyed out and described as directional evidence. They are never used in headlines, summary stats, or live pricing until the data support them.

    Independent methodology review

    Kerry G. Kauffman

    Kerry G. KauffmanB.S. in Statistics, Lehigh University

    Reviewed the shared research methodology — sampling rules, significance thresholds and testing framework — and worked through the full numbers in two example studies (NFL wind totals and MLB indoor vs outdoor). The remaining studies were reviewed for methodological consistency, not line-by-line. LinkedIn profile.

    “After examining the complete methodology and structure of the study, I am in agreement that the Welch’s t-test is the appropriate method when comparing two means.”— Kerry G. Kauffman, 2026-09-26

    His review also noted that fixed critical values of ±1.96 are the asymptotic normal approximation, and that exact critical t values depend on sample size (at our smallest sample of 78, ±1.99). We agree: every published study reports exact Welch t statistics, degrees of freedom and p values rather than relying on the fixed 1.96 shortcut.

    He further observed that where a physical weather effect appears muted — for example wind blowing out in baseball — player behaviour likely adapts (pitchers keeping the ball down, hitters elevating), and cautioned that correlational findings do not imply causation. Both points are reflected in how we word study conclusions.

    PGA Tour wind wave study — reviewed 2026-10-21

    Kerry reviewed the methodology of the PGA Tour wind wave study. His review covered how each player's wind exposure is built (the average of the hourly readings starting at the player's tee-time hour), why the 5-hour window is the primary specification, and the 3/4/6-hour sensitivity checks. On the window length he pointed us to the PGA Tour's published median round time of 4 hours 46 minutes and a Columbia University pace-of-play study showing round times are positively skewed — both of which support a 4.5–5 hour window as the reasonable core assumption.

    He also noted that with a large sample, very small differences can be statistically significant without any practical or real-life significance — and that this study is not in that territory: the differences appear practically meaningful.

    “So, everything looks really solid from the standpoint of methodology.”— Kerry G. Kauffman, 2026-10-21

    Limits

    These studies measure average weather effects, not team-specific or player-specific outcomes. They do not control for everything: team quality, play-calling, injuries, altitude, turf type, and unmeasured gusts can all matter. The findings are inputs for understanding how a game is likely to play, not guaranteed betting edges.

    Contact the research desk

    Questions about a study, its data, or its methodology can be sent to support@overunderweather.com.

    References & data sources

    • Open-Meteo ERA5 historical reanalysis — archive-api.open-meteo.com (hourly game-window weather at each venue’s own coordinates)
    • Open-Meteo geocoding API — open-meteo.com/en/docs/geocoding-api (venue coordinates and elevation)
    • Internal matched archive — venue coordinates, historical game outcomes, game-window weather, and closing lines where used.
    • Every dated change we have made to a published study is listed on our corrections log.

    OverUnderWeather Research Desk

    Independent data analysis on public sports and weather records. Studies match historical game outcomes to archived game-window weather, apply pre-set sample and significance rules, and publish null results when that is what the data show. Editorial policy and corrections log.