First, let

\begin{equation} SSE = \overrightarrow{y}'(I - H) \overrightarrow{y} \end{equation}


\begin{equation} \overrightarrow{y} \sim MN(\textbf{X} \overrightarrow{\beta}, \sigma^{2}I) \\ H = X(X'X)^{-1}X' \end{equation}


\begin{equation} E(SSE) = E(\overrightarrow{y}'(I - H) \overrightarrow{y}) = k_{1} \end{equation}

\begin{equation} => k_{1} = tr[(I-H) \sigma^{2}I] + (\textbf{X}\overrightarrow{\beta})'(I-H)\textbf{X}\overrightarrow{\beta} \\ = \sigma^{2}(tr(I)-tr(H)) + 0 \\ = \sigma^{2}[n-(k+1)] \end{equation}

\begin{equation} Var(SSE) = Var(\overrightarrow{y}'(I - H) \overrightarrow{y}) = k_{2} \end{equation}

\begin{equation} => k_{2} = 2tr[((I-H) \sigma^{2}I)^{2}] + 2(\textbf{X}\overrightarrow{\beta})'(I-H)(\sigma^{2}I(I-H))\textbf{X}\overrightarrow{\beta} \end{equation} and using a similar argument as the expectation: \begin{equation} => k_{2} = \sigma^{4}[n-(k+1)] \end{equation}

I know my mean is correct but I'm thinking my variance is wrong. Any thoughts?


1 Answer 1


Looks good to me except you forgot to multiply by $2$ for the first term in the expression for $\mathrm{Var}(SSE)$. So the final answer for the variance should be double what you wrote. By the way, the formula for $\mathrm{Var}(\mathbf{y}'(I-H)\mathbf{y})$ should have a coefficient of $4$ rather than $2$ for the second term (see this), but it doesn't matter since that part becomes $0$ anyway.

Also, another way to check your answer would be to recall that $$\frac{(n-(k+1))\widehat{\sigma}^2}{\sigma^2}\sim \chi^2_{n-(k+1)}$$ in this model. You can then take the mean and variance of both sides (using the formula for mean and variance of a chi-squared random variable).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.