Can a pitcher with high variance still have a good ERA?

Yes. A 3.00 ERA with variance 6 is possible, alternating 0-run and 6-run games. The ERA is average. The variance tells you the games are extreme. 'Good' ERA does not mean consistent.

How many games do I need to trust variance?

At least 8 full games for a starter (72 innings). For relievers, 30 innings pitched minimum. Below those thresholds, variance is noisy and may regress toward the mean with more data.

Does inherited-runner scoring affect variance?

Yes. A reliever entering with two runners on base has a higher chance of a high-earned-run game. That inflates variance. Compare variance only between pitchers in similar entry situations.

Is variance a standard baseball statistic?

No. No official rulebook defines it. It is a derived statistic from game-level earned run data. Use it as a decision tool, not a published stat.

Game-to-Game Run Prevention and Variance Explained

Two Pitchers, One ERA, Two Different Realities

A 3.00 ERA hides a split personality. Pitcher A allows exactly 3 earned runs every complete game. Pitcher B allows 0 in one game, 6 in the next, then 0, then 6. Both end the season at 3.00 ERA, but they are not the same pitcher. The difference is game-to-game run prevention and variance, and it is the difference between a reliable starter and a rotation-wrecking gamble.

Variance in earned runs allowed per game reveals consistency that the season average hides. The formula is simple: σ² = Σ(ER_per_game − μ)² ÷ N, where μ is the mean earned runs per game. Units are runs² per game². Pitcher A above has variance 0. Pitcher B has variance 4. The ERA number gives you zero information about which one you just signed.

The primary authority for calculating earned runs themselves is MLB Rule 9.16 and MLB Rule 9.02. Those rules define what counts as an earned run versus an unearned run, the official scorer’s judgment, not a formula. Variance is not a baseball rulebook stat. It is derived from game-level earned run data, and no official rule defines it. But if you manage a bullpen, you need it.

Game-to-Game Run Prevention and Variance Example

Work one concrete example with real inputs. Take two pitchers, each with 36 innings pitched across 4 complete games (9 innings per game). Both allow exactly 8 earned runs total.

Pitcher A: earned runs per game are 2, 2, 2, 2. Mean = 2. Variance = ((2−2)² + (2−2)² + (2−2)² + (2−2)²) ÷ 4 = 0. ERA = (8 ÷ 36) × 9 = 2.00.

Pitcher B: earned runs per game are 0, 4, 0, 4. Mean = 2. Variance = ((0−2)² + (4−2)² + (0−2)² + (4−2)²) ÷ 4 = (4 + 4 + 4 + 4) ÷ 4 = 4. ERA = (8 ÷ 36) × 9 = 2.00.

Same ERA. Radically different predictability. Pitcher A is a baseline you can schedule around. Pitcher B produces extreme games, either a shutout or a meltdown, that force your defense to adjust on the fly. The season average is silent on the problem.

Calculation Steps

To calculate variance yourself, you need game-level earned run data, not season totals. Collect the earned runs allowed in each full game the pitcher started. Do not mix partial innings into the same bucket, a 5.2 inning outing (which is 5.667 innings, not 5.2) has a different earned run rate than a full game.

Compute the mean earned runs per game (μ). Then for each game, subtract the mean, square the result, add them all up, and divide by the number of games (N). That is the population variance. For sample variance, which matters when you have fewer than 30 games, divide by N−1 instead.

The variance calculator on the main ERA tools page automates this step once you enter per-game earned run totals. It also flags small sample sizes: variance from fewer than 8 games (roughly 72 innings for a starter) is unreliable. For relievers, who often enter with inherited runners, the minimum usable sample is 30 innings pitched, not 8 games.

One specific failure case: a pitcher with a 3.00 ERA but variance of 6.25 (alternating 0-run and 6-run games across 8 games) is considered inconsistent. A pitcher with the same 3.00 ERA and variance of 0 is highly reliable. The average alone misleads every coach who only looks at ERA.

Why the Method Works

Why does variance matter beyond consistency? Because a high-variance pitcher changes how your team plays. If you know Pitcher A will allow 2 runs every game, you allocate bullpen resources and defensive positioning accordingly. If you have Pitcher B, you cannot plan, the 0-run game is a gift, but the 6-run game may lose the match before the relief pitcher warms up.

The handoff context from the research is precise: variance in earned runs allowed exposes consistency that the season average hides. Two pitchers can share an ERA because ERA is a mean, not a spread. One produces much less predictable game results because their game-to-game earned run totals vary widely. The ERA formula, (Earned Runs ÷ Innings Pitched) × 9, has no term for spread. It never will.

This is not a theoretical edge case. In the 2023 MLB season, the league-average ERA was 4.33 (source: Baseball-Reference). A pitcher at 3.50 ERA with variance 0 is a proven commodity. A pitcher at 3.50 ERA with variance 8 is a lottery ticket. The ERA number alone cannot tell you which you are signing, and that is the point.

How Two Pitchers Share an ERA While One Is Less Predictable

ERA is a mean, not a distribution. Mean can be identical while the underlying values diverge. Pitcher A allows 1, 3, 2, 2, 2. Pitcher B allows 0, 5, 1, 3, 1. Both average 2 earned runs per game. Both have an ERA of 2.00 over 45 innings. But Pitcher B has a wider spread, a variance higher by a factor of roughly 2.5, meaning more extreme games.

The official scorer’s judgment on earned runs (per MLB Rule 9.16) does not change this picture. Even if both pitchers face identical defense and ballpark conditions, the variance difference remains a function of the pitcher’s own consistency, not of external scoring noise.

One limit: variance itself is unstable at small sample sizes. A reliever with 15 innings pitched and variance 3 may have a true variance of 1 or 5. Only after 30 innings for relievers and 50 for starters does variance become a reliable predictor (source: FanGraphs’ research on SIERA stability). Below those thresholds, treat variance as a rough indicator, not a decision metric.

Earned Run Variance for Two Identical-ERA Pitchers

Pitcher	Game 1 ER	Game 2 ER	Game 3 ER	Game 4 ER	Mean ER	Variance	ERA
A	2	2	2	2	2.00	0.00	2.00
B	0	4	0	4	2.00	4.00	2.00

A Worked Example

This concept suits scorekeepers and league officials who need to verify ERA consistency across a season. It suits coaches evaluating pitcher performance across different levels, high school to college to professional, who need to know that a 3.00 ERA from one pitcher may be more dependable than a 3.00 ERA from another. It suits fantasy baseball players projecting future ERA from components (K/9, BB/9, HR/9, BABIP) rather than from the ERA number itself. It suits parents and players trying to understand why a pitcher’s ERA differs from their feel of how well they pitched.

It does not suit anyone looking for a single 'good ERA' chart that applies to all levels, go to the league-average ERA table by season and level instead (Baseball-Reference provides season averages). It does not suit anyone wanting to calculate WHIP, K/9, or FIP, those are separate stats with their own calculators. It does not suit anyone seeking historical ERA leaders or career leaderboards, Baseball-Reference and FanGraphs already cover those comprehensively.

The single thing that most often goes wrong: using a low-variance pitcher’s ERA as a benchmark for a high-variance pitcher’s future performance. The ERA is the same. The gamble is not.