2
$\begingroup$

Is it all right to do an F test of equality of variances on error variances (RSS divided by degrees of freedom)? Thanks a lot.

$\endgroup$
2
$\begingroup$

There are numerous aspects to this; you'll also need to clarify some things.

Some things for you to clarify:

a) Are you referring to this two-sample variance test?

b) Why are you testing equality of variance?

Some points assuming that the answer to (a) is 'yes':

i) if the assumptions are true, the test is fine

ii) unfortunately the assumptions are almost never true, and this test is very sensitive to the assumption of normality (far more so than t-tests and anova are)

iii) there are choices of test that are not so sensitive to normality, such as Browne-Forsythe. Other choices can be found at the link above

iv) if the answer to (b) is that you're testing because equal variances are an assumption of say t-tests, it is better to assume the variances are unequal than to test and take different actions on the basis of rejection

$\endgroup$
2
  • 1
    $\begingroup$ Why are you doing that? $\endgroup$ – Glen_b Jul 16 '14 at 23:50
  • $\begingroup$ Okay. If the error variances are independent - or very nearly so - one of the alternative tests at the link I gave should work okay. $\endgroup$ – Glen_b Jul 17 '14 at 0:05

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.