统计量的方差为什么是除以m-1而不是m.

UTF8gbsn

本文就是一个简单的梳理方差的无偏估计.
首先,我们一般能直观的来定义一个方差的估计

$${\hat{\sigma }}_m^2=\frac{1}{m}\sum _{i=1}^m(x_i-{\hat{\mu }}_m{)}^2$$

但是很遗憾这个估计并不是无偏估计,
因为$E({\hat{\sigma }}_m^2)\ne {\sigma }^2$. 证明如下.

$$\begin{array}{rl}E({\hat{\sigma }}_m^2)& =E(\frac{1}{m}\sum _{i=1}^m(x_i-{\hat{\mu }}_m{)}^2)\\ & =E(\frac{1}{m}\sum _{i=1}^m(x_i-\mu -{\hat{\mu }}_m+\mu {)}^2)\\ & =E(\frac{1}{m}\sum _{i=1}^m((x_i-\mu )-({\hat{\mu }}_m-\mu ){)}^2)\\ & =E(\frac{1}{m}\sum _{i=1}^m((x_i-\mu {)}^2+({\hat{\mu }}_m-\mu {)}^2-2(x_i-\mu )({\hat{\mu }}_m-\mu )))\end{array}$$

我们单独展开三项来看

$$\begin{array}{rl}& E(\frac{1}{m}\sum _{i=1}^m(x_i-\mu {)}^2)={\sigma }^2\\ & E(\frac{1}{m}\sum _{i=1}^m({\hat{\mu }}_m-\mu {)}^2)=E(({\hat{\mu }}_m-\mu {)}^2)\\ & E(-({\hat{\mu }}_m-\mu )\frac{2}{m}\sum _{i=1}^m(x_i-\mu ))=-2E(({\hat{\mu }}_m-\mu {)}^2)\end{array}$$

合并三项可得, $E({\hat{\sigma }}_m^2)={\sigma }^2-E(({\hat{\mu }}_m-\mu {)}^2)$,
又因为${\hat{\mu }}_m=\frac{1}{m}{\sum }_{i=1}^m{x}_i$带前式可得.

$$\begin{array}{rl}E[({\hat{\mu }}_m-\mu {)}^2]& =E[(\frac{1}{m}\sum _{i=1}^m{x}_i-\mu {)}^2]\\ & =E[(\frac{1}{m}\sum _{i=1}^m{x}_i-\frac{1}{m}\sum _{i=1}^m\mu {)}^2]\\ & =\frac{1}{m}E[\frac{1}{m}\sum _{i=1}^m(x_i-\mu {)}^2]=\frac{1}{m}{\sigma }^2\end{array}$$

由此可见, $E({\hat{\sigma }}_m^2)=\frac{m-1}{m}{\sigma }^2\ne {\sigma }^2$.
而且我们可以马上得出只要加上一个修正因子.即可得到${\sigma }^2$的无偏估计.也就是

$${\hat{\sigma }}_m^2=\frac{1}{m-1}\sum _{i=1}^m(x_i-{\hat{\mu }}_m{)}^2$$