Re: Covariate in ANCOVA question

Bruce Weaver <bweaver@xxxxxxxxxxxx> wrote in
news:1193226596.924562.47100@xxxxxxxxxxxxxxxxxxxxxxxxxxx:

On Oct 23, 11:35 pm, David Winsemius <doe_s...@xxxxxxxxxxx> wrote:
snipped OP posting


ANCOVA packages should also let you specify appropriate post-hoc
pairwise comaprisons. The first step would be fine but your second
would be incorrect. You would not get the desired type I error levels
in the pairwise comparisons. I suggest you search with these terms:
SPSS "analysis of covariance" "pairwise comparisons"


In discussing Fisher's LSD, Dave Howell argues (in "Statistical
Methods for Psychology") that *when there are 3 groups*, the family-
wise error rate is controlled at alpha.

* If the complete null hypothesis is true (mu1 = mu2 = mu3), the
requirement of a significant omnibus test before proceeding to pair-
wise contrasts provides the needed protection against Type I error.

* If none of the mu's are equal, Type I error is not possible for the
omnibus test, nor for any of the pair-wise tests. (Type I error can
only occur when the null hypothesis is true.)

So because there are 3 groups we should not think about type I erors in
post-hoc testing? Why, then, do the texts I have talk about controlling for
alpha and use three group examples?

--
David Winsemius
.

Loading