MLB temperature and scoring

Across 10,717 dry outdoor games at 30 parks from 2021 to 2026, temperature moves scoring in both directions and the cold side is the bigger move. Under 50°F a game gives up -1.14 runs and -0.72 home runs against mild conditions; at 90°F and above it gains +0.64 runs and +0.38 home runs.

The smaller figure in each cell is the season- and park-adjusted gap against the 70–79°F control, with its t statistic. Anything under roughly 2 is noise by our standard. Note the middle: 80–89°F adds home runs but its run-level gain (t = 1.60) is not distinguishable from no effect, so it carries no weight in our live model.

Raw scoring is one thing; whether the closing total already captured the temperature effect is another. Where we have closing totals (10,686 games so far, 2021–2026), cold games still go under the line relative to mild conditions, while heat has not cleared the significance threshold.

The “runs vs line vs mild control” column is the average (actual runs − closing total) in each band minus that same figure in the 70–79°F control band. Negative means games went under the closing total relative to mild conditions. Roughly |t| ≥ 2 is our usual noise boundary. The mild-cold 60–69°F band clears it; extreme cold and heat are directional only.

The main analysis is dry games only, so rain cannot carry the temperature result. The 864 games with measurable rain averaged 9.09 runs and 2.28 home runs — close to the dry-game average of 8.95 and 2.32. We have not measured an independent rain effect on MLB scoring and do not claim one here.

League scoring drifts year to year, so every figure above is a game measured against its own season's average — and its own park's average — rather than against the six-year pool.

• Cold is the cleanest signal. Every band below 70°F is a significant suppression, growing monotonically from −0.37 runs in the 60s to −1.14 runs under 50°F, and home runs follow the same ladder.
• Heat helps, less. The 90°F+ band adds about 0.64 runs and 0.38 home runs, both significant. The 80–89°F band's run gain is not significant — treat ordinary summer warmth as neutral for totals.
• Home runs carry the temperature story. Every non-control band is significant on home runs, including 80–89°F. As with wind, carry is the mechanism.
• Park adjustment was not optional. Hot games cluster at a handful of parks, so a raw cut would credit the thermometer with what the park itself does. The park-adjusted figures above are smaller and honest.
• The market does not fully price in the cold. Where closing totals are available, the mild-cold 60–69°F band still goes under the line relative to mild conditions (t ≈ −2.08). Extreme cold and heat are directional only vs the closing total, so the live model keeps using the raw scoring signal as its primary temperature input.

• Sample: 10,717 dry outdoor MLB games, 2021–2026, at 30 parks, each with a game-window temperature, a final score and a home-run count. Fully enclosed parks are excluded outright; retractable-roof parks contribute only roof-open games. Games with measurable rain (864) are held out of the main analysis and reported separately.
• Weather: hourly archive reanalysis at the park's own coordinates, read across the game window from first pitch.
• Every figure is season-adjusted and park-adjusted: each game is measured against its own season's average at its own park, so park mix and era drift cannot carry the result. Hot games otherwise cluster at parks like Coors Field and Chase Field.
• Minimum sample is 40 games per band, a threshold already in use across our earlier studies. Every band here clears it comfortably.
• Not controlled: humidity and air density beyond temperature (our archive does not carry humidity for these games), team quality, pitcher matchups, and extra-inning games.
• Closing totals are backfilled where available (10,686 games so far, 2021–2026). The “Did the market already price it in?” section reports runs-vs-line and Over rates against the mild control; remaining games are still being matched to historical lines.

Data generated 2026-09-18. Numbers refresh as the archive tops up; corrections are dated and kept visible.