Dear research community

(bold: key element)

To analyze the question of moderated/mediated regression in clinical psychology:

whether score in questionnaire A at baseline

predicts the later score in questionnaire B (metric, different time intervals)

AND whether the predictive effect is different depending on the type of intervention administered after baseline (6 different types, categorial variables), I wonder which statistical techniques would be most suitable, (i.e. with possible moderation or even mediation by treatment type).

A basically found two options: Linear mixed models with t-tests, and moderated regression:

LMM: suitable for the question, whether there is a predictors effect of A on later B (with baseline score of A as independent variable (between subject), the time as independent variable (within subject), and the dependent variable the subsequent score of B), however, I did not find a proper solution for for the moderation/mediation aspect. Splitting the sample in the 6 categories, calculate effect size and compare them with t-tests? I don’t think so.

However, the alternative, a moderated regression analysis doesn’t seem to be more suitable for this, neither, as there are six treatment types (Type1, 2, 3, 4, 5, 6), and if I understand right, this would need 5 coding variables which I guess, would be too complex to be interpreted. But maybe I am wrong.

Do you have any experience and/or suggestions about best practices of the analysis of such questions?

I would appreciate any statement

Thank you, Fabienne