I am working on analysis of an RCT where there are 4 groups: 1) A control group where they were given instructional material to read but which was not hypothesized to have an effect; 2) A comparison group given relevant instructional material hypothesized to have an effect; 3) Relevant instructional material + self-guided e-learning; 4) Relevant instructional material + self-guided e-learning + coaching.
The investigator hypothesizes that the 3 intervention conditions will show greater change on a set of DVs relative to the controls with the greatest degree of change in the fourth group that had the full package of intervention modalities. Participants were randomly assigned to condition and all were given pre- and post-tests on a number of DVs.
Standard OLS regression models for each DV using the pre-test as a covariate do not show any group effects so far. I suggested that the models take into account treatment dose or exposure as a mediator as participants completed varying amounts of the provided materials before taking the post-test at 6 months.
One question is how to calculate this treatment exposure? As the comparison group had no exposure to a treatment thought to effect outcome, how would their exposure be calculated? The investigator calculated percentage completion for each component (instructional material, e-learning, and coaching) and took an average. But I suggested that this is weighting each component as equally effective, which might or might not be true. It could be, for instance, that coaching is much more effective than anything else.
A second question is, is there a better way of modeling this kind of design than just regression post-test on pretest plus control condition?
And third, any thoughts on if I unbundled these treatments into dichotomous variables and calculated a variable for instructional material (yes/no), elearning (yes/no), and then coaching (yes, no). This would create different groups but would allow for testing individual "ingredients".
Any thoughts would be much appreciated from those of you with experience in analyzing treatment effectiveness data.