Hi,
I am testing a path model and Modification Indices given by MPlus suggest that I specify a correlation between A and AB (interaction).
Is this valid? This affects some coefficients in the model, such as those in of the regression of A on another variable. It improves the fit but does feel weird statistically. I guess treating this variable as endogenic (predicted by A&B) is even worse.
Can you please tell how trustworthy is this modification index, and to what extent the influence of an interaction (and suggested correlation) on model fit is valid?
Thank you,
Pier-Eric
Creating the AB interaction leads to a new variable that is going to be highly correlated with A (since A is part of AB).I would rather recommend setting up direct and indirect relationships from A and B to your other variables in the model (unless AB is highly necessary in the model).
It is not statistically(classical) valid & hence I would not recommend to specify correlation in this case.
However few are doing such things to justify their needs in MR industry (but that depends on their research questions).
Modification Indices are trustworthy but need a review or statistical point of view for interpretations.
In this situation, the AB interaction explains significantly more variance, which I cannot achieve with an indirect effect (no really part of the specification)
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Materials Science
- Materials
- Inorganic Materials
- Ceramics