故事 · 淹死在「平均一米」的河里 Story · Drowned in a River With "Mean Depth 1 m"
统计学里有句老话：一个人若把双脚分别泡在冰水和开水里，平均温度很舒适，但他已经痛得跳起来了。 有个工程师听说要过的河「平均水深 1 米」，自己身高 1 米 8，便放心下水，结果在 3 米深的主槽里没了顶。 均值告诉你中心在哪，却对波动只字不提 —— 而真正决定风险、决定良率、决定客户体验的，恰恰是波动。 这就是为什么六西格玛把名字押在希腊字母 σ（标准差）上：管理质量，本质是管理波动。 平均达标只是及格线，把 σ 压到足够小、让结果稳定可预测，才是真功夫。 There's an old line in statistics: put one foot in ice water and the other in boiling water, and the average feels just fine — except you're already screaming. A field engineer once heard the river he had to cross had a "mean depth of 1 m". He was 1.8 m tall, so in he went — and went under in the 3 m central channel. The mean tells you where the center is and says nothing about spread — yet spread is what drives risk, yield, and customer experience. That's why Six Sigma pinned its name on the Greek letter σ: quality management is really variation management. Hitting the average is just the passing grade. Driving σ down until results become tight and predictable — that's the real craft.

1 同一个均值，σ 决定波动带的宽窄 Same Mean, σ Decides How Wide the Bands Open

σ 适中

蓝线是均值 x̄（中心）。绿、黄、红三条带分别是 ±1σ / ±2σ / ±3σ。每个点到均值的竖直距离平方平均后开根号，就是 σ。把散布拉大，三条带一起撑开；拉小，它们收拢到均值附近。 The blue line is the mean x̄ (the center). The green, yellow, and red bands are ±1σ / ±2σ / ±3σ. For each point, square the vertical distance to the mean, average those, take the square root — that's σ. Pull the spread up and all three bands fan out together; pull it down and they collapse toward the mean.

2 两条河：均值相同，σ 不同 Two Rivers: Same Mean, Different σ

两条河平均水深都是 1 米。左边 σ 小（处处一米，能蹚）；右边 σ 大（忽浅忽深，淹人）。平均一样，σ 不同，风险天壤之别 —— 这就是为什么只看平均会害死人。 Both rivers have a mean depth of 1 m. On the left, σ is small (1 m everywhere — wadeable); on the right, σ is large (knee-deep then over your head — drownable). Same mean, different σ, opposite risk — which is why a mean alone can get someone killed.

3 现实里的标准差 σ in the Real World

波动的统一度量：σ 和数据同单位（毫米、秒、元），直观可比；方差（σ²）单位是平方，常用于计算而非直观解读。 A unified yardstick for spread: σ carries the same unit as the data (mm, s, $), so it reads naturally. Variance (σ²) lives in squared units — handy in formulas, awkward for intuition.
经验法则：近似正态时，约 68% 数据落在 ±1σ、95% 落在 ±2σ、99.7% 落在 ±3σ。一眼判断「离谱程度」。 The empirical rule: when data is roughly normal, about 68% falls within ±1σ, 95% within ±2σ, 99.7% within ±3σ. A quick "how unusual is this?" check.
过程波动：制程能力 Cp、Cpk、控制图上下限，全建立在 σ 上。σ 越小，过程越稳，良率越高。 Process variation: Cp, Cpk, and control-chart limits all sit on top of σ. Smaller σ → steadier process → higher yield.
风险 = 波动：投资波动率、交期不确定性、尺寸离散度 —— 凡是「不确定」都用 σ 度量。降 σ 就是降风险。 Risk = variation: portfolio volatility, lead-time uncertainty, dimensional scatter — anything labeled "uncertain" is measured in σ. Shrinking σ shrinks risk.
一句话In One Line
标准差是「数据离均值平均有多远」的一把尺。算法很朴素：每个点到均值的距离， 先平方（消掉正负、放大远点），求平均得方差，再开根号让单位变回原样 —— 这就是 σ。 它之所以是六西格玛的灵魂，是因为质量的敌人不是「偏离目标」，而是「不稳定」。 一个均值刚好达标却忽高忽低的过程，远比一个略有固定偏差但极其稳定的过程更危险。 先把波动（σ）压下来，再谈把均值调到目标 —— 这是改善的正确顺序，也是「平均一米照样淹死人」的真正教训。 Standard deviation is the ruler for "how far points sit from the mean on average". The recipe is plain: take each point's distance from the mean, square it (kills the sign, punishes outliers), average them to get the variance, then take the square root to land back in the original unit — that's σ. The reason it's the soul of Six Sigma: the enemy of quality isn't "off-target", it's "unstable". A process that hits the mean but jumps around is far more dangerous than one with a small fixed bias but rock-solid consistency. Crush variation (σ) first, then dial the mean to target — that's the right order of improvement, and the real lesson behind "mean depth 1 m can still drown you".
常见误用Common Mistakes
只看平均，不看波动（以为均值达标就万事大吉）。均值配 σ 一起看，否则平均一米的河照样淹死人。 Reading the mean and ignoring the spread (assuming "on-target = all good"). Always pair the mean with σ — otherwise the "mean depth 1 m" river still drowns you.
对偏态/有极端值的数据硬套 68-95-99.7 法则。经验法则只对近似正态成立，偏态数据要先看分布形状。 Forcing the 68-95-99.7 rule onto skewed or heavy-tailed data. The empirical rule only holds when data is roughly normal — for skewed data, inspect the shape first.
把方差（σ²）当标准差直接解读单位。σ 才与原数据同单位；方差用于计算，不便直观比较。 Reading variance (σ²) as if it shared the data's unit. Only σ shares units with the data; variance is for computation, not intuitive comparison.

标准差