样本均值和样本方差
首先对于样本$x_1...x_n$来说,他们的均值为与方差分别为:
$displaystylear{x} = frac{1}{n}sumlimits_{i=1}^{n}x_i$
$displaystyle s^2 = frac{sumlimits_{i=1}^{n} (x_i - ar{x})^2}{n-1}$
要证明样本方差的无偏性,首先要计算样本均值的方差。
样本均值的方差
$displaystyle D(ar{x}) = D(frac{sumlimits_{i=1}^{n}x_i}{n}) = frac{1}{n^2}sumlimits_{i=1}^{n}D(x_i) = frac{1}{n^2}sumlimits_{i=1}^{n}sigma^2 = frac{sigma^2}{n}$
样本均值和样本方差的无偏性证明
$displaystyle E(ar{x}) = E(frac{1}{n}sumlimits_{i=1}^{n}x_i) = frac{1}{n}sumlimits_{i=1}^{n}E(x_i) = frac{1}{n}sumlimits_{i=1}^{n}mu = mu$
$displaystyle E(s^2)$
$displaystyle= E(frac{sumlimits_{i=1}^{n} (x_i - ar{x})^2}{n-1})$
$displaystyle= frac{1}{n-1}sumlimits_{i=1}^{n}E(x_i^2 + ar{x}^2 - 2x_iar{x}) $
$displaystyle= frac{1}{n-1}sumlimits_{i=1}^{n}[D(x) + E(x)^2 + D(ar{x}) + E(ar{x})^2 - 2E((x_i^2 + x_ix_1 + ... + x_ix_{i-1} + x_ix_{i+1} + ... + x_ix_n)/n)]$
$displaystylexlongequal[]{E(x_ix_j) = E(x_i)E(x_j)} frac{1}{n-1}sumlimits_{i=1}^{n}[sigma^2 + mu^2 + sigma^2/n + mu^2 - 2(sigma^2 + mu^2 + (n-1)mu^2)/n]$
$displaystyle= frac{1}{n-1}sumlimits_{i=1}^{n}(frac{n - 1}{n}sigma^2)$
$displaystyle= sigma^2$
样本方差的方差
如果总体$X sim N(mu, sigma^2)$,那它的样本方差有:
$displaystylefrac{(n-1)s^2}{sigma^2} sim chi^2(n - 1)$.
由于$chi^2$分布的方差为两倍的自由度,得:
$displaystyle D(frac{(n-1)s^2}{sigma^2}) = 2(n - 1)$
$displaystyle D(s^2) = frac{2sigma^4}{n - 1}$