How to compare groups on multiple facets of a outcome and how that outcome relates to a subsequent outcome? I am interested in comparing levels of insight between a bipolar group and a schizophrenia / schizoaffective group. I am assuming I will have to use "current mood state" as a covariate because it could influence self-reported levels of insight in these psychiatric populations. I have 4 instruments that measured different facets of insight for each group. So although my DV is "insight", since I have multiple measures of insight  would the most appropriate test be a MANCOVA? 
Secondly, I want to see if these different facets of insight predicts quality of life scores in the individuals. I was told that a hierarchical multiple regression would be the best test to answer this question. But what if I wanted to split up the quality of life items into different domains to see if certain facets of insight predict certain domains within quality of life? 
 A: You will need to use Structural Equation Modeling (SEM) for this.  SEM is an analysis of the correlation matrix of your variables.  It works by allowing you to impose a pattern of relationships that lies behind the data you see.  This patter can be simple connections between variables, or it can even include latent (unobserved) variables.  A given patten will tend to produce certain types of correlations among your variables.  The correlations that would occur are compared statistically to the observed correlations to see how good the fit is.  That is, your hypothesized set of underlying relationships forms the null hypothesis.  As a result, the SEM modeling process is somewhat different from the more typical (e.g., regression) modeling process.  Of course, this can be learned, but if you aren't familiar with it and aren't very statistically sophisticated, you may want to work with a statistical consultant.  
"Pattern" strikes me as a bit vague, so here is an example I found via Google (it is copied from here--I have no knowledge of the article): 

I this picture we see nine 'manifest' variables (rectangles), so the observed correlation matrix was presumably $9\times 9$.  This specified pattern has implications for the correlations that show up, which may or may not fit well with the observed correlation matrix.  
If you'd like to read through some of the threads on CV pertaining to SEM, you can check out our sem tag.  
