Matching residuals from multiple membership models

[MFalster]

Hi there,

I am running a two-level multiple membership model, and am trying to map the higher-level residuals to their corresponding classification ID.

I know that in a standard two-level model the residuals are stored according to the order of the sorted IDs for that classification. My trouble is my multiple membership model includes higher level groupings which are not identified in the first of the classification weight/ID variables, but in the second, or third, or fourth etc. My weights are derived from a frequency distribution of services supplied, so this happens for higher-level groupings that are not the most common provider of services for any of the participants, but still provide some level of service provision.

The number of residuals produced matches the total number of groupings across all variables, but I'm not sure what order the residuals are sorted in and how to correctly match them, particularly the groupings who are not identified in the first weight/ID variable. Do you have any advice on how this is done/how I could identify them?

Many thanks,
Michael

[ChrisCharlton]

I believe that the residuals should be sorted in the order that you would get if you sorted the combined multiple-membership ID columns. If you find an instance where this isn't the case could your provide an example and I will look into it further?

[MFalster]

Thank you Chris - this does seem to be the case. I had tried sorting by the order they first appeared if you sequentially moved down along each variable, or across along each row, but hadn't considered the sorted order of all possible values of the classification. Many thanks for your help.

All the best, Michael