I'm currently reading Introduction to statistical learning. When trying to prove the collective significance of a regression linear model, we use the F-test with the following formula.


Whre $TSS=\sum(y_i-/y)²$, and $RSS=\sum(y_i-\hat y_i)²$

What I understood from this formula is that we would want that $F$ be larger than $1$ in order to reject the null hypothesis, because that means the amount of variability in $y$ explained by our model ($TSS-RSS$) is larger than the amount that the model couldn't remove ($RSS$).

Anyway, I am stuck on proving the result in the following section:

If the linear model assumptions are correct, one can show that: $E{RSS/(n − p − 1)} = σ²$ and that, provided $H0$ is true, $E{(TSS − RSS)/p} = σ²$.

with: $H_0$: $b_1=b_2=...=b_p=0$

$b_i$ is the coefficient for the ith regressor.

So, what I want is a proof or a hint for the quote above. And additional explanations would be appreciated.

  • $\begingroup$ σ² is the variance of the error term. $\endgroup$ – Youssef Esseddiq Feb 8 at 15:32
  • $\begingroup$ Do you understand how the F-test works in an ANOVA? $\endgroup$ – gung Feb 8 at 15:36
  • $\begingroup$ No not very much, I know F-test is used to compare models fitted to a certain dataset..it would be very helpful if you could explain it. $\endgroup$ – Youssef Esseddiq Feb 8 at 15:42
  • $\begingroup$ You may glean some insight from my answer here: How does the standard error work? $\endgroup$ – gung Feb 8 at 15:45
  • $\begingroup$ It looks like you mis-stated the null hypothesis. $\endgroup$ – Isabella Ghement Feb 8 at 16:58

Your Answer

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

Browse other questions tagged or ask your own question.