The GLM hat matrix is $$\mathbf{H}_{GLM} = \mathbf{W^{1/2}X(X^{T}WX)^{-1}X^{T}W^{1/2}}$$

where $\mathbf{W}$ is a diagonal matrix with elements $w_i = (\delta\mu_i/\delta\eta_i)^2/\text{var}(y_i).$ For OLS, the link function is identity so $\mu_i = \eta_i$ and $\delta\mu_i/\delta\eta_i = 1.$ Also, the random component is Gaussian, so $\text{var}(y_i) = \sigma^2$.

From this I can express $\mathbf{W}$ as $\frac{1}{\sigma^2}\mathbf{I}.$ Plugging this into the GLM hat matrix, I get

\begin{align}\mathbf{H}_{OLS} &= \left(\frac{1}{\sigma^2}\mathbf{I}\right)^{1/2}\mathbf{X(X^{T}}\left(\frac{1}{\sigma^2}\mathbf{I}\right)\mathbf{X)^{-1}X^{T}}\left(\frac{1}{\sigma^2}\mathbf{I}\right)^{1/2} \\ &=\frac{1}{\sigma^4}\mathbf{X(X^{T}X)^{-1}X^{T}} \end{align}

But thats not right! That $\frac{1}{\sigma^4}$ term does not belong there. The correct answer should be $$\mathbf{H}_{OLS} = \mathbf{X(X^{T}X)^{-1}X^{T}}$$

What did I do wrong?


In \begin{align}\mathbf{H}_{OLS} &= \left(\frac{1}{\sigma^2}\mathbf{I}\right)^{1/2}\mathbf{X(X^{T}}\left(\frac{1}{\sigma^2}\mathbf{I}\right)\mathbf{X)^{-1}X^{T}}\left(\frac{1}{\sigma^2}\mathbf{I}\right)^{1/2}, \end{align} we have (twice, once left, once right) $\left(\frac{1}{\sigma^2}\right)^{1/2}=\frac{1}{\sigma}$ as well as $$ \left(\frac{1}{\sigma^2}\right)^{-1}=\sigma^2 $$ in the middle term.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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