Given a linear model $$ y = \beta_0 + \beta_1x_1 + \beta_2x_2+\beta_3x_3 + \beta_4x_4 $$ we can perform an $F$-test for the null hypothesis $$ H_0: \beta_1 = \beta_2 = \beta_3 = \beta_4 = 0 $$ However, what test is appropriate for a subset of these coefficients being zero in the same model? $$ H_0': \beta_1 = \beta_4 = 0 $$


1 Answer 1


You can use still an F-test. The test statistic is: $$F_0 = \dfrac{SS_R/2}{MS_E},$$ where $SS_R$ is the increase in the residual sum of squares of the reduced model (setting $\beta_1=\beta_4=0)$ with respect to the full model with all the parameters. $MS_E$ is the residual sum of squares of the full model divided by $n-5$, where $n$ is the number of observations.

Under $H_0$, $F_0$ is distributed as a $F_0$ of Fisher-Snedecor with $d_1=2$ (i.e. number of restrictions) and $d_2=n-5$.

Remark: clearly the result is valid under the usual assumptions of the Linear regression (e.g. Gaussian residuals).

See here the details of what they call partial F Test.

  • 2
    $\begingroup$ Perhaps you could expand your post a bit to summarize the main ideas. It's common practice on this site when linking to an external source. This makes the answer self-contained, and preserves it for the benefit of future readers in case the link ever goes dead. $\endgroup$
    – user20160
    Feb 28, 2018 at 12:54
  • $\begingroup$ @user20160, you are right! I edited as suggested. $\endgroup$
    – Jim
    Feb 28, 2018 at 18:17

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.