I have two independent variables, one independent variable has 2 mediators and the other independent variable has 3 mediators and there are three outcome variables. I'm confused whether I should treat this model as two independent models in structural equation modeling as in the whole model, there are different set of mediators for each exogenous variable and only the outcome variables are same. Moreover, when I run two separate models comprising of each of the exogenous variable with its set of mediators and outcome variables, the fit indices are fine but when I combine it with other exogenous variable with its corresponding set of mediators, then the model fit are poor.Thorough guidance required please
Razia, I guess that your independent variables are correlated as well as your mediator variables. You should have strong theoretical reasons for fixing all these relations to zero. Just have a look that the correlation matrix.
Such a large model is of course quite complex and you have to think about the relations carefully: Are the variables correlated or not and does there exist some further direct and indirect effects that should be included in the model?
It would be better to include all variables in one model.
Regards, Karin
Thanks Madam Karin for your guidance, my independent variable are negatively correlated, please explain how I can test the mediation in structural equation modeling using AMOS, should i run the structural model simultaneously with exogenous variables, their corresponding sets of mediators and outcome variables
Razia, I think that not only your independent variables are correlated, but also the independent variables with all mediator variables, and the mediator variables among each other. Have a closer look at the correlations between all variables, this will give you an idea about the relationships that you did not include.
And yes, you should analyze a path model with all variables included in the model. Unfortunately, I am not familiar with AMOS. First, you would need to know all paths to be included in the complete model and you should also decide on the paths that should be fixed to zero.
Hope this helps, Karin
Thank you very much Madam Karin for your guidance, it is quite helpful
Razia, when you developed theoretical framework (which include mediator constructs), I guess you have already supported it with literature, right? Otherwise you are not allowed to proceed. Thus, no question should arise about literature again.
And when you conduct CFA (Confirmatory Factor Analysis) to validate the latent constructs involved in the model, you already developed the Discriminant Validity Index Summary, Right? Thus, the issues of correlation among constructs in the model should not arise (because if the constructs are not discriminant among them you would have modified the measurement model).
Now, you assemble the model and draw Amos Graphic as your theoretical framework says and execute. The mediator should be analysed one at a time even though you may have multiple mediators in your model. In the path analysis, look for indirect effect and direct effect. You can download the attached chapter in my RG.
I agree with Karin, may your independent variables are correlated! or Type I or Type II error may occurred! However, before testing the model, underlying assumptions should be satisfied! These assumptions includes (Correlation Test: “if the data trustworthy”; Multicollinearity Test: “may it helpful to avoid a Type I or Type II error, or over- or under-estimation of significance or effect size(s)”; Normality Test: “if the data normally distributed”). Failing to satisfy the assumptions does not mean that your answer is wrong. It means that your solution may under-report the strength of the relationships. Outliers can distort the results. Once the foregoing assumptions satisfied! You can analyze a path model with all variables included in the model.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling