I'm reading Faraway's book (http://cran.r-project.org/doc/contrib/Faraway-PRA.pdf) to try to understand R's lm diagnostic plots. On page 72 of the book is this:


I have been trying to understand a few things about this for a few days reading online and some help would be very, very appreciated.

  1. In the derivation of $var(\hat\varepsilon)$, why did the term $(1-H)X\beta$ vanish? Where did it go?
  2. What is an intuitive explanation of what $h_i$ means in terms of my data? What does each $h_i$ in the $H$ vector (not matrix?) correspond to? How is $h_i = H_ii$?
  3. What is $p$?
  1. Try the obvious - expanding it out:

$$(I - H) X = X - HX = X - X(X'X)^{-1}X' X = X - X = \mathbf{0}$$

Oh, and $\mathbf{0}\beta = \underline{0}$.

2. Each element $h_i$ has several interpretations.

e.g. (i) It's the rate of change of $\hat{y}$ with respect to $y$.

e.g. (ii) It's the proportion by which the variance of a residual is reduced from the corresponding variance of the error it approximates - as was showed in the text in the image.

The reason $h_i = H_{ii}$ is because we defined it that way.

3. $p$ is the number of predictors (including the intercept in this case), the column dimension of $X$.


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.