Hello everyone, I'm hoping you can assist me. As an infrequent AMOS user, I'm encountering a persistent error while trying to model the attached diagram:
'The model is probably unidentified. In order to achieve identifiability, it will probably be necessary to impose 4 additional constraints.'
Could anyone guide me on how to identify where these 4 necessary constraints should be applied within my model?
I am studying wellbeing based on latent variables, which consist of other measurable factors. My goal is to analyze how these variables influence overall wellbeing.
I don't see an obvious identification problem with your model. One issue that could lead to empirical underidentification in your model is the fact that the latent variable "Physical environment" has only two indicators (manifest variables: PE-clean and PE-comfort). If (1) the two indicators of this latent variable are not substantially correlated with one another or (2) the latent variable "Physical environment" itself is not substantially correlated with at least one other latent variable in the model, then the model may be underidentified. You can check on that by (1) looking at the observed correlations of PE-clean and PE-comfort (they need to be substantially greater than zero) and (2) by fitting separate CFA models in which "Physical environment" is correlated with only one of the other factors at a time. (2) should reveal any potential issues with (near zero) latent variable correlations.
Hello Ahmad,
without having more details about your measures and just looking at their names, i got the question wether conceptualizing your indicators reflecting an underlying common cause (which a factor is) is the most appropriate way. For instance the service delivery items look like different facets of quality and the overall quality is rather a composite of them instead of their underlying cause. The fact that you had to incoporate several error covariances supports my skepticism. Perhaps simply add them together.
Best,
Holger
- Badges
- Science topic