|
|
 |
| Using standardised coefficients? |
 
|
| Message/Author |
|
|
|
|
Hi, I am running DSEM mediation models, and I would like to calculate the indirect effects using standardised estimates. Is there a way of using the standardised estimates in MODEL CONSTRAINT? If not, can you tell me how I can calculate the standardised estimates in a two-level DSEM model in MODEL CONSTRAINT?
Best,
Fredrik Falkenström |
|
|
|
|
If you have Y on X as the indirect effect beta then the standardized value is beta*sqrt(VarX/VarY).
You can get VarX and VarY from output:residual or output:residual(cluster).
If you want to code it yourself in model constraint you will have to do that using Appendix D
http://www.statmodel.com/download/DSEM.pdf
Note that random parameters can't be used in model constraint and if you need something like that you would have to save the posterior distribution of the random and non-random parameters and do the computation in spreadsheet software. |
|
|
|
|
Thanks! So the variances/covariances I get from output:residual are the model-implied ones, not the observed variances? So these are the ones used for standardisation, then?
I didn't understand how to use the formulas in Appendix D of the DSEM paper, but I guess I can just run the model twice - once to get the variances from the residual output and then put that into Model Constraint to calculate the standardised coefficients with standard errors and confidence intervals? |
|
|
|
|
| Yes on all the questions. In principle, the running it twice approach won't account for the variability in sqrt(VarX/VarY), but I doubt that this is essential and as you say the full computation is pretty complex. |
|
|
|
|
| Great, thanks! |
|
|
| Back to top |
|
|
