|
|
 |
| Bifactor Model Monte Carlo Power Anal... |
 
|
| Message/Author |
|
|
|
|
Hello,
We are trying to run a bi-factor power analysis in M+, but are having some issues. We have continuous dependent measures a1-a7 and d1-d8. We want all but three of the items (a1-a6, and d1-d6) to load on to the common latent factor 'Common_Int', and a1-a7 to load on to 'Anxiety', and d1-d8 to load on the 'Depression'.
Our effect sizes are based on standardized effect sizes we found for depression and anxiety in our previous study, and should be large enough to have adequate power with our sample size. But when we run the model, the %sig coefficient is low for Depression and Anxiety, and the estimate averages are also low which is unexpected considering our sample size (N=550).
In our analysis, we are using the latent variables as outcomes, so we thought this may be our problem. Is it possible we may want to edit how we standardize the variance? We were standardizing it to 1, but with the latent variables as outcomes is this incorrect? When we set this to 1 are we actually setting the residual variance, and do we need to set this to the specific residual variance for each of the latent variable outcomes instead?
Thank you!
Aly |
|
|
|
|
| Yes, variances for DVs are residual variances not the full variance (explained var. + residual var.). Because you build on standardized estimates, you should choose a residual variance that makes the full variance = 1. |
|
|
| Back to top |
|
|
