I want to assess the association between several variables and the eating behavior in individuals with intellectual disability. Since I have missing data on some variables I want to calculate my regression using FIML in Mplus. In addition I have some clustered (some individuals live in institutions with several individuals from the same institution in my data set) data and some unclustered (some cases life with their families, only one individual out of each family in my dataset) in my data set.
Is it appropriate to use TYPE=complex in this situation? Any good reference?
THANK YOU!
Manuel,
Yes, that's the way to go. Some of your clusters have size 1, which doesn't present any problems.
I don't have a reference offhand, but group-delivered interventions often have this come up -- where the intervention group is clustered, but there's no analogous clustering for the comparison respondents.
Pat
I agree, it would be the more conservative approach and honor the nature of your data. You could also check the ICCs from the unconditional model and if they are low enough, ignore the clustering, although there is debate about that option. Best of luck, hope it goes well.
Commenting on Eric's "there is debate around that option."
I'm one of those who tries to avoid relying on low ICCs. In the case of a small number of large clusters (for uncertain definitions of "small" and "large"), even an ICC that appears very low on a correlation/R^2 scale can have substantial impact.
Added to that, including the clustering can not make your standard errors less valid. At worst, it will have no effect (that's the case with ICCs of exactly zero or the conceptual case of clusters all of size 1).
So unless there are computational difficulties from including the clustering (e.g., as can arise with bootstrapping), my advice is to always account for it if it is logically appropriate, regardless of observed ICC.
Pat and Eric, thank you very much for your helpful responses!
Best, Manuel
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling