I ran an experiment (psychology) with two groups: experimental and control. I want to see whether the intervention changed the way in which participants habitually use some emotion regulation (ER) strategies, measured with Likert scales. One of my hypotheses is that the experimental group increased in their use of one certain ER and that these changes consequently translate into a different relationship with another outcome variable, which is emotional reactivity. So I basically induced distress in pre-test and measured emotional reactivity and ER use, delivered the intervention, and then did the same thing again.
So what I'm trying to do now is the following:
> lm(postReactivity ~ preReactivity + GROUPS + gainER + GROUPS:gainER,
data = dataset)
As you can see, I am basically using pre-test measurements of emotional reactivity to the distressing stimulus as covariates, but at the same time I am interested in how the changes in ER relate to the post-test emotional reactivity. I hypothesize a different relationship between the changes in emotional reactivity and changes in use of emotion regulation, so I am testing for moderation. Ergo, I am running an ANCOVA testing for moderation on the outcome variable. Now, this model gives me the expected results, but I want to make sure that it's not bad practice to do so. Are there better alternatives to using gain scores (post-test - pre-test) in order to test for this interaction? Also, is it an issue to test for interaction on a post-test measure while controlling for the pre-test measure?
*This post has been edited to improve clarity on the matter.