removing paths from the model after bootstrapping

[aoaguiar]

After bootstrapping, some paths resulted not significant (t<1.96). I should then remove these paths from the model, right? Should I remove all paths in one single step or should I remove one and run the model again? How to choose the path to be removed?

[cringle]

Depends! Oftentimes showing that an anticipated path is not significant in a certain context may also be a valuable result.

Best
Christian

[BenLien]

Depends! Oftentimes showing that an anticipated path is not significant in a certain context may also be a valuable result.

Best
Christian

I have a hard time figuring out, when not to delete insignificant paths from the model, and then reestimating it.
Is the following statements true:
- Because the PLS-SEM algorithm uses adjecent LV to approximate a given latent variable, then including a insignifcant path in the final model must affect the entire model. Hence, if the other relationships in the model is of interest, should we not remove insignificant relations?
- If the goal is only to test if a given relationship is significant, then maybe CB-SEM could be a better alternative (depending on complexity, distribution assumptions and so on)?
- After identifying an insignificant effect and removing this, I should think that we still had shown that an anticipated relationship in a given context is not significant?

Best
Benjamin

[cringle]

In PLS, you usually look for strong effects that hold in general or in a specific context (hence, you may decide to skip the other ones; the do not contribute to the explanation of the target construct).

For model testing, which can include very low coefficients or even null relationships, you may want to use CB-SEM.

Best
CR