Hi there,   I'm automating a 2-stage probability-based sample selection method in SAS that will be used on 20 or so different input data sets, each with its own specifications. In stage 1, I'm selecting PSUs proportional to size and in stage 2, I'm selecting SSUs (individuals here) SRS within PSUs. The number of SSUs in each PSU is not equal, but that's not a big deal.   In my implementation, I use PROC SURVEYSELECT with METHOD=PPS. I have once gotten an error reading "For METHOD=PPS, the relative size of each sampling unit must not exceed (1/SAMPSIZE)" which I understand. It was well explained here: http://support.sas.com/kb/23/759.html    To fix it, I've manually gone into the input sample list, removed the PSU causing trouble, selected that PSU with certainty for my sample (which seems logical to me since the probability of selection is greater than 1), then selected the remaining n-1 PSUs via PPS using PROC SURVEYSELECT.   To me, this seems clunky and annoying, but I haven't been able to find another way to work  around this issue. Does anyone know of a way to have SAS treat these PSUs as being selected with certainty rather than having it spit out an error that the probability of selection is greater than 1? I've looked at the other variations on the PPS method in SAS and they all seem to have the same way of calculating probability of selection.    Thank you!!!! ... View more

Hi all,   I was wondering where I could find information on how both PROC MIXED and PROC NLMIXED deal with missing data. Do they assume MAR? Do they have the same approach to dealing with missing data?   When I run mixed models in both procs, the output includes "Number of Observations Read" and "Number of Observations Used," so I want to understand what is going on under the hood with both of these procs.   Any help woud be greatly appreciated! Thanks!! ... View more