Bettina R posted on Monday, March 07, 2016 - 5:51 am
Dear Profs. Muthén,
I want to fit two multilevel mediation models with type=crossclassified.
In the first model, the mediation is on level 1 with a categorical mediator variable. As cross-classified models use Bayes estimation, I’m not sure whether I have to define the mediator variable as categorical like in a linear regression model. Or can I just use it as a continuous variable with Bayes?
In the second model, the mediation is on level 2. Can I calculate the indirect effect with ind = a*b when using cross-classified with Bayes estimation?
You can use both continuous and categorical mediator and categorical should be preferred since it reflects your data more accurately. Also take a look at the mediator=observed/latent option, see page 643 in the user's guide.
To calculate the indirect effect you need this command
In my data, pupils have responded to the same questionnaire multiple times during different school lessons. In other words, the responses are cross-classified by lessons and pupils.
I'm trying to conduct a multilevel mediation analysis with multiple mediators. The causal variable (X) is an intervention, which is assumed to increase students' use of imagination (Y). Variable X is dichotomous, and variable Y is measured on a 4-point Likert scale.
The M-variables are also dichotomous: the pupils were asked whether they were building models, conducting research, reading textbooks etc. when prompted to respond to the questionnaire.
When it comes to pupils, there is a 1 -> 1 -> 1 type of mediation, and all paths could/should be random. When it comes to lessons, the mediation type is 2 -> 1 -> 1 and only the paths between the M and Y variables could/should be random.
I tried to estimate the model using TYPE=CROSSCLASSIFIED RANDOM, but the variables being categorical is a problem. I get the following message:
*** FATAL ERROR MODELS WITH RANDOM SLOPES FOR CATEGORICAL VARIABLES CAN NOT BE ESTIMATED WITH THE BAYES ESTIMATOR.