Mixed ANOVA with covariate - interpretation help!!!

#1
Hi,

I conducted the following mixed ANOVA with a covariate:

2 x groups (BS factor)
3 x Switching as repeated measures (WS factor)
IQ as covariate.

My output gave me the following results:

in the BS-table, the effect of IQ and group x IQ was not significant. Which is good, I guess. not so nice is that the main effect of groups is also non-sign.

WS table:

- Main effect of switching not sign.
- switching x covar. not sign. (which is good I guess!)
-switching x group significant (apparently the result of your switching depends on which group you are in)

- SIGN group x switching x IQ interaction!!!!
What do I do with this? I could not find any help online on this matter. Does that mean that IQ confounds the interaction of switching x group???

I am really stuck on this problem, any help is appreciated!!
 

Karabiner

TS Contributor
#2
The many exclamation marks and question marks seem to
indicate that something very uncommon or unexpected
was happening, but I am not sure why.

By the way, you didn't describe whether allocation to groups
was randomized, what you mean by switch, or how large the
group sizes are.

So, without sufficient knowledge of the research question and
the study design, at least one could say that the groups effect
depends on / varies between levels of "switch", and that this
interaction effect is moderated by IQ.

With kind regards

K.
 
#3
Thanks for the reply! To answer your questions:

- Allocation to groups was not random
- Switch means: subjects had to answer certain items. in order to answer these items, they had to perform a certain switching operation, for example answering an item about the size of geometrical figures (Question1) and then answering a question if a word would make sense (question2). the switch from answering question 1 to question 2 would be a categorical switch. there are 2 more of these switching operations in the experiment.
- group sizes are:
1. N=12
2. N=8
3. N=7

I found that the groups seem to differ sign. on IQ. I wanted to use IQ as a covariate, but I can't do that if groups differ on it sign. However, only group 3 differed sign. from the other groups, so I excluded the group when including the covariate. Therefore I only included group 1 and 2 in the following analysis.
What I found was the following: (NOTE: Cattell = IQ = covariate)
http://postimg.org/image/kl60qio3n/
http://postimg.org/image/rcwftdd37/

I read that it might be reasonable to exclude the interaction with the covariate if you have a 3-way interaction. However, I could not find out how to determine if this is true or not, or how to test this.
 

Karabiner

TS Contributor
#4
I wanted to use IQ as a covariate, but I can't do that if groups differ on it sign. However, only group 3 differed sign. from the other groups, so I excluded the group when including the covariate.
First of all, with groups as small as yours, non-significance
is hardly an indicator that groups don't differ on IQ. Moreover,
you include a covariate if you have reason to assume that
groups differ on it, and that it can be responsible for apparent
group effects. If groups don't differ regarding a covariate,
it is no longer a covariate.
I read that it might be reasonable to exclude the interaction with the covariate if you have a 3-way interaction. However, I could not find out how to determine if this is true or not, or how to test this.
If you consider IQ purely as a covariate, you should have
defined a model without those interactions where IQ is involved
(and used all 3 groups, of course). After you have found a moderation
effect, I don't know if it is justifiable to ignore it.

With kind regards

K.