topic Re: proc surveyreg - class statement in Statistical Procedures

proc surveyreg - class statement

nuitamment — Tue, 12 Jul 2016 12:23:47 GMT

Hello,

I am using the class statement in proc surveyreg in order to construct industry dummy variables. However, sometimes some industries don't contain any observations or too few. Then I cannot use the lsmeans statement for instance. Is it possible to automatically exclude one group within the class statement when the group is empty?

Re: proc surveyreg - class statement

PGStats — Tue, 12 Jul 2016 17:35:58 GMT

You must exclude industries before running proc surveyreg. You can do that by 1) creating an intermediate dataset, 2) by creating an intermediate data view, or 3) by adding a WHERE clause, either as a dataset option or as a WHERE statement, when calling proc surveyreg.

Re: proc surveyreg - class statement

nuitamment — Tue, 12 Jul 2016 17:39:09 GMT

I was hoping that there would be a more elegant solution. But thank you.

Re: proc surveyreg - class statement

SteveDenham — Wed, 13 Jul 2016 17:31:39 GMT

Because of the weighting, it becomes difficult to implement what Milliken and Johnson refer to as a "means model" in PROC SURVEYREG. However, you could give it a try--fitting what is essentially a saturated one-way model--and then use LSMESTIMATE statements to construct tests and comparisons of interest.

Steve Denham