I am working on a very complicated analysis and am quite lost!
I am analysing data from a randomised controlled trial. I have 3 groups (treatment 1, treatment 2, waitlist), and 3 repeated measures (baseline, time 1, time 2). In addition, I have 7 possible continuous moderators that I want to test, so I want to know whether the relationship between treatment group and time point changes at different values of the potential moderating variable.
I am running separate analyses for each moderator.
I created two dummy variables for both time (time 1, time 2) and treatment (treatment 1, treatment 2).
I entered those in a mixed model on SPSS:
MIXED Y with time1 time2 treatment1 treatment2 moderator
/PRINT R SOLUTION TESTCOV
/METHOD = ML
/REPEATED Time | SUBJECT(ID) COVTYPE(UN) .
What I am interested in are the last 4 three-way interactions, because they would indicate that the treatments have a differential effect on the post-tests along different levels of the moderating variables (right?)
I adopted a model building strategy whereby I first enter the "main effectiveness" predictors (time1 time2 treatment1 treatment2 time1*treatment1 time1*treatment2
time2*treatment1 time2*treatment2), then the moderator main effects (moderator
moderator*time1 moderator*time2), and finally the rest of the interactions.
I only proceed if there is a significant improvement in model fit according to the -2LL value.
Unfortunately, the last step never seems to be significant...
I am wondering whether I am correct in using all these dummy variables, and whether I actually need them or not. I am very confused.
For instance, do I need the main effects of time 1 and time 2 dummies? If I exclude them I get better results...
I would very much appreciate your thoughts on how best to analyse my data! Please please please help me!!!!