|
Message/Author |
|
|
This is related to my previous post on testing group differences in slopes. I am working with a longitudinal data set (4 time points) collected in a randomized trial. The primary objective was to see if changes in parent child communication due to the intervention would lead to healthier behaviors. Thus it seems to me that this is a mediation model in which the slope would be the mediator. Are there any applications published which have used a model like this? thanks |
|
|
I don't know of any articles but you might want to contact Mike Stoolmiller or Terry Duncan. The following paper which can be downloaded from the website shows how to use growth shape in the form of the categorical latent variable rather than the slope as a mediator. This approach may be preferable. Muthén, B., Brown, C.H., Masyn, K., Jo, B., Khoo, S.T., Yang, C.C., Wang, C.P., Kellam, S., Carlin, J. & Liao, J. (2002). General growth mixture modeling for randomized preventive interventions. Biostatistics, 3, 459-475. |
|
|
Here's another paper that describes the differences between the slope as a mediator and the categorical latent variable as a mediator: Muthén, B., Khoo, S.T., Francis, D. & Kim Boscardin, C. (2003). Analysis of reading skills development from Kindergarten through first grade: An application of growth mixture modeling to sequential processes. In S.R. Reise & N. Duan (eds), Multilevel Modeling: Methodological Advances, Issues, and Applications (pp. 71-89). Mahaw, NJ: Lawrence Erlbaum Associates (#77). If you don't have access to it, you can request paper 77 from bmuthen@ucla.edu. |
|
|
We are attempting to conduct a similar analysis, and are having trouble getting parameter estimates when using a count outcome as the primary dependent variable, when slope is the mediator and treatment is the predictor. When using ALGORITHM=EM, the following error message comes up: "MODEL INDIRECT is not available for analysis with ALGORITHM=INTEGRATION." Below is the model set-up used: MODEL: i s | tbinge1@0 tbinge2@1 tbinge3@2 tbinge4@3 tbinge5@4 tbinge6@5; OBE4 ON I S; OBE4 ON IPT; MODEL INDIRECT: OBE4 VIA s IPT; OUTPUT: CINTERVAL(BCBOOTSTRAP); Any thoughts on estimating indirect effect in this type of model is greatly appreciated! Thank you! |
|
|
You can always create your own indirect effect by using Model Constraint, defining a New parameter that is the product of two Model parameters. A new paper will soon be posted with more on indirect effects with count and other non-normal outcomes and mediators. |
|
|
The paper sounds very interesting. I look forward to it. So the model would look like this: MODEL: i s | tbinge1@0 tbinge2@1 tbinge3@2 tbinge4@3 tbinge5@4 tbinge6@5; OBE4 ON I; OBE4 S (p1); OBE4 ON IPT (p2); MODEL CONSTRAINT: NEW (ab = p1*p2); OUTPUT: CINTERVAL(BCBOOTSTRAP); |
|
|
I figured out the MODEL CONSTRAINT language, but I still get the same error about integration. In this warning it references the use BOOTSTRAP as the reason for not calcluating the effect. If I drop the BOOTSTRAP request, the model terminates. MODEL: i s | tbinge1@0 tbinge2@1 tbinge3@2 tbinge4@3 tbinge5@4 tbinge6@5; OBE4 ON I; OBE4 ON S (p1); OBE4 ON IPT (p2); MODEL CONSTRAINT: NEW (ab); ab = p1*p2; OUTPUT: CINTERVAL(BCBOOTSTRAP); |
|
|
Yes, bootstrap is not allowed with integration because we figured it could lead to very slow runs. But I don't see a mediator in your setup. And even when you have mediation, bootstrapped SEs aren't always needed - I would say that's the exception rather than the rule. |
|
Back to top |
|
|