Cluster-Robust Standard Errors or MLR Estimator?

[Psych Student]

Hello Everyone,

I'm currently trying to analyze a path model in lavaan and my dataset has a few interesting aspects. 

• The data are clustered, such that individuals (N = 110) are distributed across 21 teams (ranging in size from 3 to 11). 
• My mediator has strong positive skew.
• The dataset has missing data

My research questions and hypotheses reside at the individual level, so I'm simply trying to account for the clustered nature of the data in my analyses. 

It is my understanding that I can use the "cluster = " argument in conjunction with the "missing - 'FIML'" to simultaneously account for missing data and also calculate cluster-robust standard errors. However, I'm not sure if cluster-robust standard errors correct for univariate non-normality. I also have read that I can use the "estimator = 'MLR'", which both accounts for non-normality and also missing data, but I don't see any option to specify my clustering variable when using the MLR estimator.

Additionally, when using either approach listed above, I am not able conduct bootstrapped standard errors with my defined parameters (e.g., indirect and direct effects).

My Questions:

• Do cluster-robust standard errors correct for univariate non-normality in addition to non-independence?
• Does the MLR estimator not only correct for non-normality, but also clustering? If so, how do I specify my clustering variable?
• How do I decide on which approach to take (e.g., using cluster-robust standard errors vs. MLR estimator)?
• Do both cluster-robust standard errors and MLR eliminate the need for bootstrapping for my defined parameters?

Thank you,

Tim