When conducting a t-test we can always use a Welch approximation of the df, and let go of the assumption of equal variances.
Is there something similar for (even one way) anova? (any R implementation of that?)
$\begingroup$
$\endgroup$
asked Feb 12, 2011 at 16:46
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
The latest version of ez lets you pass a white.adjust argument to car::Anova(), which implements a correction for heteroscedasticity. See ?car::Anova() for details.
answered Feb 12, 2011 at 17:26
$\begingroup$
$\endgroup$
There is a function named oneway.test() in the base stats package, which implements Welch correction for a one-way ANOVA. Its use is similar to the standard t.test() function. It is also referred to as O'Brien transformation (Biometrics 40 (1984), 1079--1087) and might be applied with two or more independent samples.
