Hello Rosa,

your initial model assumes/implies that you have got all common causes of your 5 mediators (the two IVs) in the model and that there no left common causes that are not in the model. In addition, your model assumes that the mediators do not have causal effects on each other. Quite many assumptions, right?

You cannot blindly draw arrows as in any other case each arrow represent a cause in which you really believe. The most straightforward approach would be to let the error terms covary to reflect unobserved common causes. This will lead to a fitting model and unbiased direct effects (if I don't overlook something) but you probably miss indirect in case the mediators (or some) have effects on each other.

In my personal opinion, such a situation should rather lead to running an exploratory SEM (e.g., for example using the TETRAD software that is free and easy to use) but I know that this is beyond the horizon of typical researchers and their hypothesis-testing fetish.

https://www.youtube.com/watch?v=7gktywp6j1Y&t=1484s

HTH

--Holger

I think the below video will help you to solve the problem. You can watch it till the end.

https://www.youtube.com/watch?v=xAVHnSMxW0c

You're using path analysis (using computed observed variables) instead of SEM where latent variables are formed through the relevant observed variables hence leading to more degree of freedom (df) as comparative fit indices and goodness of fit indices are based on degree of freedom.

Thanks Muhammad Anwar and Imran Anwar for helping out!