Currently working through Introduction to Statistical learning, and on page 76 I am struggling with the following:

They state the equation for the F statistic as:

$F = \frac{\frac{TSS − RSS}{p}} {\frac{RSS}{n−p−1}}$

where, TSS = $\sum(y_i − y)^2$ and RSS = $\sum(y_i −yˆi)^2$

Then they say:

If the linear model assumptions are correct, one can show that $E \frac{RSS}{n − p − 1}$ = $\sigma^2$ and that, provided $H_0$ is true, $E\frac{TSS − RSS}{p}$ = $\sigma^2$.

Can someone please explain this?

  • $\begingroup$ Are you looking for proofs? or just plain English explanation? Proofs available all over the Internet. $\endgroup$ – SmallChess Oct 9 '17 at 22:07

Your model is $y_i = \beta_0 + \beta X_i + e$ with the $e \sim_{IID}$ normal$(0, \sigma^2)$. $\beta$ is a vec of length $p$, $X$ a model matrix of dim $n \times p$.

The correct expression for TSS is $\sum(y_i - \bar{y})$ with $\bar{y}$ the sample mean.

The first expectation is a direct consequence of the model assumption, using a degrees of freedom correction.

If the null hypothesis is true, what do we know of $\hat{y}$'s value?

| 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.