|
|
 |
| Multilevel mediation SEM. |
 
|
| Message/Author |
|
|
|
|
Hello,
A colleague of mine and I are estimating a multilevel mediation SEM using TWOLEVEL RANDOM. Our IV is a latent variable.
%WITHIN%
T1MMQcop ON ders_5 (aw);
T9CUPITc ON T1MMQcop(bw);
T9CUPITc ON ders_5
%BETWEEN%
T1MMQcop ON ders_5 (ab);
T9CUPITc ON T1MMQcop (bb);
T9CUPITc ON ders_5 ;
We get this error:
*** ERROR in MODEL command
A latent variable declared on the within level cannot be used on the
between level. Problem with variable: (IV)
Does mplus support this kind of model The error does not occur when we comment out the between level effect.
Thank you, |
|
|
|
|
| I don't see a BY statement that defines the latent variable so we need to see the full output - send to Mplus Support along with your license number. |
|
|
| Sara Geven posted on Tuesday, August 27, 2019 - 7:52 am
|
|
|
Dear all,
I have data with students nested in schools. I estimated a mediation model with random slopes. There are statistically significant random slopes and I have calculated 95% Confidence Intervals for the separate paths of the indirect effects with a random slope (i.e., (slope mean +/- 1.97*(slope var^0.5)). However, I would also like to get an indication of the variation in the total indirect effect (not just the separate paths). I find this especially interesting for indirect effects that are on average not statistically significant, but do have considerable slope variation. Is there maybe a way to calculate 95%CI for the total indirect effect, or can I get a better grasp about the variation in the total indirect effect in another way? Thank you!
All the best,
Sara Geven |
|
|
| Sara Geven posted on Wednesday, August 28, 2019 - 7:09 am
|
|
|
Dear all,
Excuse me for my previous question, as of course the 95% CI of the indirect effect can be obtained from the estimates of the indirect effect in model constraint.
Sorry, I was just being confused.
All the best,
Sara |
|
|
| Sara Geven posted on Wednesday, August 28, 2019 - 9:17 am
|
|
|
Dear all,
Sorry my question actually still holds, because if I'm correct this CI says something about the average indirect effect, but does not indicate to what extent the total indirect effect may vary across schools....
Sara |
|
|
|
|
| Our Short Course Topic 7, slides 81-83 describes the indirect effect when both a and b are random and like you say, this is done via Model Constraint and it shows the expected value and its SE. I don't know the formula for the variance; maybe the original article by Bauer, Preacher and Gil (2006) has it. You can, however, also get factor scores (see the FSCORES option in the UG index) for a_j and b_j, where j denotes cluster j. Then you can see what their product is in different clusters. |
|
|
| Sara Geven posted on Thursday, August 29, 2019 - 2:28 am
|
|
|
Dear Bengt,
Thank you for your quick reply.
Yes, I am interested in the variance of the indirect effects (or the predictive intervals of these effects - I mistakenly used CI in my first message to refer to this).
Unfortunately the FSCORES option does not work, because I am using multiply imputed data.
It does seem that Bauer, Preacher and Gil (2006) do provide an equation to calculate the variance of the indirect effect. I am tempted to use this, yet I was wondering whether applying this equation to the obtained estimates based on multiply imputed data may lead to biases. I guess that officially one should calculate the variance of the indirect effect for each imputed dataset separately, and then combine these estimates. What is your take on this?
And how could I do this?
Thank you!
Sara Geven |
|
|
|
|
| Yes, with multiple imputations your parameter estimate is the mean of the estimates from analyzing each imputed data set. The SE for the estimate is obtained by the usual Within-between formula of imputation. |
|
|
| Back to top |
|
|
