Maxim K posted on Monday, October 07, 2013 - 8:40 am
I am struggling with the choice of a proper estimation method for the following model. The effects of job mobility pattern (nominal) have to be established on the retirement age (continuous). Both direct and indirect effects have to be estimated, the latter via a number of factors, e.g. pension benefit, health etc (either continuous or dichotomous). Consequently, this is a path model with manifest variables only.
The difficulty is that the data are essentially multilevel, but the 2-nd level is comprised of 12 countries. This is too small a number of clusters for the multilevel approach. On the other hand, the clustering is expected to have a profound effect, so one cannot disregard it.
What does MPLUS offer in terms of estimating models with this setup? Thanks in advance for any information, references and links to additional resources are most certainly welcome.
You should control for non-independence of observations by creating 11 dummy variables and using those as covariates in your model. For multilevel analysis you need a minimum of 30-50 clusters in most cases.
Maxim K posted on Tuesday, October 08, 2013 - 6:30 am