Separate question on linear mixed models - I'm not exactly sure how to interpret the fixed effects results. In this case, my dependent variable is Beck Depression Inventory scores (log-transformed), and my depression variable is 0=No and 1=Yes. I'm using log-transformed values, because the raw scores are highly skewed. Here are the values, all are significant (p<0.001):
Intercept: 1.14
Depressed=0: 1.065
SessionNum (continuous): -0.072
Depressed=0*SessionNum: -0.058
I know these are log-values, so I presume I need to exponentiate them. My best interpretation is that Beck scores decrease by an average of 0.93 (e^-0.072) points per session, and being depressed decreases Beck scores by an average of 2.90 points (e^1.065). Finally, the interaction of time and depression accounts for another decrease of 0.94 points each session (e^-0.058). The tests indicate that depression does affect the rate of decline of Beck scores over treatment sessions.
Is this basic interpretation on the right track? I'm confused about the beta for "Depressed=0" - If it's for the effect of not being depressed, I expect the value to be negative. However, negative log-values do not mean the same thing as negative raw values. When I run the LMM on the raw values, the beta is negative (intercept=10.4 and beta(Depressed)=-8.12). Should I just model the LMM on the raw values of Beck?
I'm starting out using a very simple model with relationships that are obvious (of course being depressed is going to affect depression measures, and of course, the depressed are going to have a steeper decline of depression measures than non-depressed). This is so that I can get a good understanding before I look at the outcomes of real interest (related to cigarette smoking).
Thanks for any help you provide.
Karen Harker
Locked
