起源故事 · 一个德国天文学家的细账
Origin Story · A German Astronomer's Careful Bookkeeping
这条校正以德国天文学家、数学家 Friedrich Bessel 命名。19 世纪测量天体位置时,人们只有少量观测样本,
却想估出测量误差到底有多大。Bessel 注意到:用样本自己的平均当作"真中心"去算偏差,会让偏差总是偏小 ——
因为样本均值就是那个让平方和最小的点。少了 1 个自由度(n 个数据,1 个已经被均值"用掉"),
所以分母该是 n − 1 而非 n。这一改,估计才不偏不倚。
The correction is named after the German astronomer and mathematician Friedrich Bessel. In the 19th century,
astronomers had only a handful of observations of a star's position but still wanted to know how large the measurement
error really was. Bessel noticed something subtle: if you use the sample's own mean as the "true center" when computing
deviations, those deviations are guaranteed to be too small — because the sample mean is, by definition, the point that
minimises the sum of squared deviations. One degree of freedom has been "spent" estimating the mean (n data points,
but 1 has been used up), so the denominator must be n − 1, not n. With that one tweak, the estimator becomes
unbiased.
1 ÷n vs ÷(n−1) · 谁对准了真值 ÷n vs ÷(n−1) · Which One Hits the Truth
n = 52 现实里的 n−1 n − 1 in the Real World
实验小样本:研发只跑 5、8 个样品估波动,÷n 会让你以为过程比实际更稳,÷(n−1) 才不冒进。
Tiny R&D samples: when engineering only builds 5 or 8 prototypes, ÷n makes the process look steadier than it actually is. ÷(n−1) keeps you honest.
Excel / 软件默认:VAR.S、STDEV.S 用 n−1(样本);VAR.P、STDEV.P 用 N(总体)。选错分母结果就偏。
Excel and software defaults: VAR.S and STDEV.S use n − 1 (sample); VAR.P and STDEV.P use N (population). Pick the wrong one and your estimate is biased.
自由度:t 检验、ANOVA、回归里到处是 n−1、n−k —— 同一个逻辑:每估一个参数,就"花掉"一个自由度。
Degrees of freedom: t-tests, ANOVA and regression are full of n − 1 and n − k — same logic: every parameter you estimate "spends" one degree of freedom.
总体已知时:若真知道全体每个值(普查 / 全数检验),分母就该用 N,不必校正。
When the whole population is known: in a census or 100% inspection where every value is observed, divide by N — no correction needed.
一句话In One Line
n − 1 不是凑数,是为"我们不知道真均值"付的税。真方差是相对真均值 μ 算的,
可现实只有样本均值 x̄ —— 而 x̄ 恰恰是让 Σ(xᵢ − x̄)² 最小的点,所以这堆平方和天生偏小。
除以更小的 n − 1,正好把缺口补平,使估计的期望等于真值,这就是无偏。
n 越小,校正越关键;n 越大,n−1 和 n 几乎没差别 —— 这也是为什么大样本里你常常看不出区别。
n − 1 isn't a fudge factor — it's the tax you pay for not knowing the true mean. The true variance
is measured around μ, but in practice you only have x̄ — and x̄ is exactly the point that minimises Σ(xᵢ − x̄)², so the sum of
squares is born too small. Dividing by the slightly smaller n − 1 closes the gap and makes the
estimator's expectation equal the truth — that's the definition of unbiasedness.
The smaller n is, the more the correction matters; the larger n gets, the less n − 1 and n differ —
which is why on big samples you barely notice the choice.
常见误用Common Mistakes
样本数据却除以 n(误用总体公式)。手上是样本就用 n−1,否则系统性低估波动与风险。
Applying the population formula (÷ n) to sample data. If your data is a sample, always use n − 1 — otherwise you will systematically understate both variability and risk.
以为 s 也是无偏的。无偏的是方差 s²;开根号后的标准差 s 略有偏(但仍是常用估计)。
Assuming s is unbiased too. The unbiased quantity is the variance s²; the square root, the sample standard deviation s, is slightly biased — though it is still the workhorse estimator in practice.
Excel 随手抓 VAR.P / STDEV.P 算样本。样本统计一律用带 .S 的函数(n−1)。
Reaching for VAR.P / STDEV.P in Excel on sample data. For sample statistics, always pick the .S functions — they use n − 1.