在有些时候,直接计算随机变量的方差非常麻烦,此时可以用方差分解公式,将方差分解为条件期望的方差加条件方差的期望:

Var(X)=Var[E(X|Y)]+E[Var(X|Y)]

证明非常简单,注意到

\begin{aligned} \text{Var}[\text{E}(X|Y)] =& \text{E}\left\{\left[\text{E}(X|Y)\right]^2\right\} - \left\{\text{E}\left[\text{E}(X|Y)\right]\right\}^2\\ =& \text{E}\left\{\left[\text{E}(X|Y)\right]^2\right\} - \left[\text{E}(X)\right]^2 \end{aligned}

\begin{aligned} \text{E}[\text{Var}(X|Y)] =& \text{E}\left\{\text{E}(X^2|Y) - [\text{E}(X|Y)]^2\right\}\\ =& \text{E}(X^2) - \text{E}\left\{\left[\text{E}(X|Y)\right]^2\right\} \end{aligned}

将上面两式相加,即得证。