|
|
Three level path analysis |
|
Message/Author |
|
|
Hi there, I'm having some issues constructing a multilevel path analysis, probably mainly because I have no experience in multilevel path analysis. All my variables are recoded into categorical, dichotomous, variables. As I want to predict a child outcome by a couple of parent and grandparent variables and the grandparent variables (4) are used to predict the parent outcomes (2, of only the father ór the mother... separate analysis for both parents) which in turn predict the child outcome (1), I assumed I would have to be using a three level path model (all family members are included in the data, so I usually have multiple children in a family). Now I already tried that by partially using example 9.21 from the manual. As I’m not familiar with multilevel modeling in this context, I’m probably doing multiple things wrong, but at least one, since the model gives the error I'm using variables on too many levels. Do you perhaps have an idea what type of model I do have to construct to control for nesting? |
|
|
I would try this as a single-level model as a first step. Use one child per parent and put the data in a wide format with child variables, parent variables, and grandparent variables. Multivariate analysis takes care af the non-independence of observations. |
|
|
Thanks for your suggestions! I already ran a single level path model, which worked fine. Using one child per parent is a good suggestion, I will do that as well. Though, I'd still like to control for clustering in a full model. However, I still have issues setting up the multilevel model. It doesn't seem to be possible to indicate variables as 'categorical' and include them into the model as both dependent and independent variables. Is this even possible in the multilevel model? Moreover, if I want a variable to be modeled on both levels 2 and 3 - and I thus do not label it as a level 2 or 3 variable - I get an error, while I use this variable in a comparable way as variable y2 in example 9.21. |
|
|
Send the outputs that show errors and your license number to support@statmodel.com. |
|
Back to top |
|
|