Tim Wind posted on Tuesday, June 05, 2012 - 7:04 am
My research question is related to multilevel growth modeling. We did research in Rwanda and appied an intervention that fosters community social capital which in turns influences mental health.
We measured mental health on three time points (within level), measured social capital on three time points (between level), and have an intervention group and a control group (u-variable which is independent).
The idea is: Intervention (intervention versus control) --> community social capital (t1, t2, t3) --> mental health (m1, m2, m3).
How do I analyse this hypothesis in MPlus (with multilevel growth modeling?)? Further, how can I use a categorical variable as an independent variable? In the output, it says a categorical 'u'-variable can only be used as a dependent variable.
The scale of independent variables is not an issue in model estimation. In regression, independent variables can be binary or continuous. In both cases, they are treated as continuous in model estimation.
Is the intervention on the within or between levels?
Tim Wind posted on Thursday, June 07, 2012 - 12:06 am
Thank you for the explanation on independent variables. The intervention is on the between level.
Variables on the between level can be used only on the between level. A variable on the within level can be used on both levels. See Examples 9.1 and 9.2 for the building blocks you can use to develop your model.
1. Above, you specified that variables on the within level can be used on both levels. However, if I attempt to include the regression of between-level slope and intercept factors on within-level variables, I get errors stating that "within-level variables cannot be used on the between level." Is this something that has changed in Mplus?
2. In the Mplus User's Guide Example 9.12 it states that at the within-level the residual variances of indicators are fixed to equality across times "in line with conventional multilevel growth modeling." This is the same for the residual variances of the indicators being fixed at 0 in the between level of the model. Can these constraints be relaxed? In my models, relaxing these constraints improves model fit, but I am not sure if this causes problems for interpreting the results of the model.
1. A variable measured on the within level can be used on both levels in most cases if it is not put on the WITHIN list. If it is put on the WITHIN list, it can only be used on within. See Examples 9.1 and 9.2 in the user's guide for more information about this.
2. The residual variances are fixed at zero as stated in the example. This is not holding them equal. This could also be done or they could all be free.