Does anybody know and can anyone help with the following?
In a partially restricted actor-partner interdependence model (on a dataset of distinguishable dyads) estimated using lavaan in R, the function parameterEstimates returns DIFFERENT standardized coefficients for effects that were restricted to be equal across partners (while the unstandardized coefficients are indeed equal). Is this correct and supposed to be so? Also, how to handle this in a paper? Is it unadvisable to use standardized coefficients in APIM? (any reference to back the claim up, please?) Or: is there a better way to obtain (more? :) correct estimates?
Hello Anna,
A quick answer: When you impose restrictions in lavaan, these apply to the unstandardized values. Conversion to standardized coefficients can (as you report!) result in different values. Nothing more complex.
What I can't advise you on is what the prevalent reporting of values for APIM models might be.
Good luck with your work.
Hello Anna,
A quick answer: When you impose restrictions in lavaan, these apply to the unstandardized values. Conversion to standardized coefficients can (as you report!) result in different values. Nothing more complex.
What I can't advise you on is what the prevalent reporting of values for APIM models might be.
Good luck with your work.
Just to explain beta = b x (sd_x / sd_y). So if you constrain b to be equal then beta will depend on whether sd_x (the sd of the predictor) is equal for the two predictors (presumably sd_y being common to both).
I don't think it makes much sense to constrain the standardised coefficients to be equal except by also constraining the two sds to be equal (if you just constrained the betas to be equal then that would risk forcing b to be different). I've not run APIM models so I don't offhand know if you can constrain both b and the variances of the the predictors. That would be reasonable if the sds are very similar and its a prior plausible that the variances are equal.
Generally one should be cautious about interpreting standardised coefficients and effect sizes in multilevel and SEM models as weird things can happen. It also depends how they are computed - especially for things like interactions.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics