Mixed ANCOVA- significant covariate, insignificant between subj interaction!

#1
Hi,

Hopefully someone can help me!

The problem:

I had participants do a test of knowledge, then use an educational intervention, take the same knowledge test straight afterwards, and then again 3 months later to test retention of knowledge. These participants also did a test of dexterity as that might affect how they learn from the intervention. I'm looking at whether the condition (type of intervention) affects the change in knowledge scores over time, and whether dexterity is a covariate.
So I've run a mixed ANCOVA. The intervention was one of two methods, so I have 2 conditions: A and B. I also have 3 (repeated measures) knowledge test scores from 3 time points: pre intervention, post intervention and retention (3 months later). I also have a dexterity test score as a possible covariate.

Results:

There is a significant main effect (knowledge scores increased from pre to post, but did not differ significantly post to retention) and no significant interaction effect, so conditon didn't significantly effect the change in knowledge scores.

However, dexterity is a significant covariate! On an insignificant interaction. If I run the ANCOVA again with just pre and post scores, dexterity is no longer significant a covariate.

So what does this output mean? If dexterity is a covariate in an insignificant interaction, does this mean that condition doesn't effect knowledge gain, but fine dexterity does? Does it mean that those with higher dexterity sores remembered better (had better retention scores) regardless of condition? Are there any follow up tests I need to do? Any opinions on this result? Any help would be great!
 
#2
These participants also did a test of dexterity as that might affect how they learn from the intervention.
This describes a supposed moderation effect and should be tested by dexterity*group*time.
However, dexterity is a significant covariate! On an insignificant interaction
What do you find surprising - that dexterity affects the scores? If you assume
dexterity as a covariate (although you described it as a moderator, see above),
then naturally you should expect its association with test scores. Why bother?

With kind regards

Karabiner
 
#3
Hi, thanks for replying.

>This describes a supposed moderation effect and should be tested by dexterity*group*time.

When I run the ANCOVA, the interactions that it shows me are time*dexterity and time*condition. It's time*dexterity that is signficiant.

I ran a custom model to get a time*condition*dexterity interaction and it came out as signficant in the within-subjects output too.

I'm not entirely sure what this means.

This is how I understand it at the moment :

If the interaction between conditon and time is insignificant then condition doesn't effect the change in scores over the time points.
The dexterity*time interaction means that regardless of the condition, dexterity is having a signfiicant effect on the change in knowledge score across the 3 time points. Does this look correct to you?

My follow up question is, why does the significant time*dexterity interaction disapear when I run the ANCOVA with just pre and post scores as within subjects. What does it mean that it's only significant when I add retention scores? Is it that those with better dexterity retain their knowledge better?

Thanks again.
 
#5
Many thanks for you help. After some more work in SPSS, I found that the difference in score from post to retention is correlated signficiantly with the dexterity covariate. No correlation with pre or post scores though. This suggests that it's the retention of knowledge that's effected by fine dexterity score.

Best Wishes,
S_P