# Multiple Mediation Model

3 posts / 0 new
Offline
Joined: 06/30/2021 - 23:08
Multiple Mediation Model

Hello,

I am using metaSEM and random-effects TSSEM to estimate a multiple mediation model. There is a single (exogenous) IV, there are six (endogenous) Mediators, and there is a single (endogenous) DV -- thus, there are six indirect effects. Although the model seems to run correctly there are a couple issues I am running into. I would really appreciate guidance on this!

First, occasionally one or two of the UB/LB confidence intervals print as NA. This doesn't seem to be a problem because it only happens sometimes, but I am wondering whether there is an issue that is causing this. Second, although the RMSEA indicates good fit, other indices show very poor fit. I am unsure why the fit is so poor because the parameters and estimates seem to make sense.

I am not exactly sure how to troubleshoot this and thus don't know exactly which questions to ask, but this has led me to wondering whether I’ve specified the model correctly using the lavaan syntax and the matrix algebra in the tssem2 function. Should the mediators be allowed to covary in the model syntax, for example, or should the indirect effects be specified to take into account the other paths in the model, or am I not setting the correct arguments in tssem1 or tssem2? Perhaps the model is too complex and there is insufficient data available to estimate the whole thing, or perhaps there is a single path without enough data? These are basic questions, but I would very much appreciate an expert opinion on this to make sure that I've done this analysis correctly!

Thank you very much!
John

Offline
Joined: 10/08/2009 - 22:37
Hi John,

Hi John,

Different fit indices may not agree well with each other sometimes.

For mediation models, the common practice is to allow correlations among the mediators. If you free these residuals, the NAs disappear.

ts ~~ cp + sc + ss + ea + fs
cp ~~ sc + ss + ea + fs
sc ~~ ss + ea + fs
ss ~~ ea + fs
ea ~~ fs

Mike

Offline
Joined: 06/30/2021 - 23:08
Hi Mike

Hi Mike

Thank you very much -- I really appreciate your help!

Best,
John