I am analyzing a multi-site data set (k = 14). One of the sites did not administer a subset of the items to be used in my analyses; and thus all participants from that site are coded as missing on that subset (resulting in 9% missing data on affected items).
My analyses will primarily consist of multi-group SEM with "site" as the grouping variable (e.g., testing factorial invariance across sites). The data have weight and stratum variables, and all items are dichotomous.
If possible, I would like to avoid deleting the site or deleting the affected items due to the abovementioned missingness. What would be the best way to handle this missingness scenario in Mplus? Would the standard missing command (which used to be TYPE = MISSING combined with ESTIMATOR = ML/MLR in previous versions), which I believe uses FIML, be sufficient for this scenario, or do I have to do something special?
I ran the model without specifying "variances = nocheck" and didn't receive an error using MLR (but did receive the error with WLSMV).
What I was really wondering is: as I am currently specifying the model, would the missing data on the items that were not used in one of the sites be considered MAR, or would I have to do something else to the model to make them MAR (i.e., incorporate the grouping variable somehow differently into the model)?