Linear SEM with Mediation/Moderation PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 Rieke Heeringa posted on Wednesday, August 01, 2018 - 3:15 am
Dear readers,

In order to test for predictors (14) of suicidal ideation (Y), I have conducted a multivariate linear regression and found some significant predictors (A, B and C, all continuous).

The aim of the research was to see there is mediation/moderation in the relationship between a significant predicting concept (X) and suicidal ideation (Y).
Therefore, I also did a MLR with my predictors on X and found a significant effect of D (dichotomous), E, A and F (continuous).

Based upon the literature, I expected a mediation effect of predictors G, B and C (all continuous) and a moderating effect of predictors H, I and J (all dichotomous).

Therefore I did the following:

1. A model with all predicted mediators (G, B and C);
Model: Y on G B C
G B C on X
Model indirect: Y ind TB
According to RMSEA, CFI and TLI the model fit is suboptimal. Based upon bootstrapping results, I did find a significant effect of predictors B and C.
- Can I interpret these even though the model fit is not significant, and if so, how do I interpret them (what does the outcome mean)?

(1/2)
 Rieke Heeringa posted on Wednesday, August 01, 2018 - 3:16 am
2. Three separate models with the predicted moderators (H, I and J);
Model: Y on
X (b1)
H (b2)
XH (b3) (where XH = X*H)
None of the interaction terms were significant.
- Is this due to H, I and J not showing up as significant on the MLR?
- Can I put these together in the same model? Or will this not affect anything?

3. I was planning on making one model with all of the hypotheses together, but I'm not sure if that's possible/helpful when most of them turn out to be non-significant. Could you help me out on this point?

4. Finally, I was planning to do an explorative analysis and check whether my significant predictors from the MLR (A, B and C and maybe even D, A and F) could be a mediator/moderator in the relation between X and Y.
- Is it possible to check this when there is no prediction about them being a moderator/mediator?
- Can I do this with all these variables when only A loads on both X and Y?
- Is this possible using SEM?
- If so, how do I do this.

I absolutely realize I have a lot of questions. I have been studying the various manuals, topics, online information and youtube, but still remain a bit clueless. I would like to thank you with the most sincerity for taking the time to read and answer my questions.

Thank you

(2/2)
 Bengt O. Muthen posted on Wednesday, August 01, 2018 - 1:50 pm
These general analysis strategy questions are better suited for SEMNET. When you post it there, I recommend that you make sure to describe what "TB" is and why you talk about "A, B, C"; "E, A, and F"; as well as G, B, and C" - that's confusing to me at least.

Also, we ask that postings be limited to one window. For longer Mplus-related questions, send to Support along with your license nmber.
 Anna Lee posted on Thursday, August 02, 2018 - 5:12 am
Thank you, and my apologies for the inconvenience. Maybe it's best to delete this topic then? I'm not sure how to do so though.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: