topic Re: LSmeans multiple control levels in SAS Procedures

LSmeans multiple control levels

DDK — Thu, 23 Apr 2015 13:06:52 GMT

Ok. This might be a ridiculously stupid question easily solved by experienced users.

SAS PROC mixed documentation states the following.

For the life of me I can't seem to get that statement to work in proc mixed. My variables have a similar lay-out. I am able to specify only A*B and the appropriate controls in the LSmeans statement (without B*C in there). When I add the B*C and specify control levels as in the SAS documentation, I get the error "Wrong number of control levels for A*B" and "Wrong number of control levels for B*C". Am I missing something???

Re: LSmeans multiple control levels

SteveDenham — Fri, 24 Apr 2015 17:14:30 GMT

Can you share your PROC MIXED code?

This does make a really good case for the use of the LSMESTIMATE statement, though...

Steve Denham

Re: LSmeans multiple control levels

DDK — Mon, 27 Apr 2015 12:09:20 GMT

Hello SteveDenham,

The code is really simplified because I wanted to try it out:

proc mixed data=test.testdata;

class a b c;

model y = a b c a*b a*c;

lsmeans a b / pdiff=control('1' '1');

run;

The same problem occurs with the lsmeans statement:

lsmeans a*b b*c/ pdiff=control('1' '1' '1' '1');

I get output from (each separately):

lsmeans a*b / pdiff=control('1' '1');

lsmeans a / pdiff=control('1');

lsmeans b*c / pdiff=control('1' '1');

lsmeans b / pdiff=control('1');

Once I try to add a factor in the same line as SAS suggests (i.e. lsmeans a*b b*c/ pdiff=control('1' '1' '1' '1');) , it doesn't work.

If I use :

lsmeans a*b b*c/ pdiff=control('1' '1');

lsmeans a b / pdiff=control('1');

I only get output for a*b or a but not for b*c or b.

It's not a major problem as I of course can specify it separately, but as SAS indicates it is a possibility I wondered why it is not working for me.

Re: LSmeans multiple control levels

SteveDenham — Mon, 27 Apr 2015 18:24:09 GMT

I was able to get the error, and I came up with a solution.

First, the levels in the quotes have to be the formatted values in the dataset. For instance, in my test case, when I had control('1' '1') it failed, as my levels for a were 'Female' and 'Male', so control('Female' '1') works.

Second, when multiple factors are present, the separate parts have to be separated by a comma:

This works for me:

lsmeans sex*group group*week/diff=control('Female' '1' , '1' '1');

This does not:

lsmeans sex*group group*week/diff=control('Female' '1' '1' '1');

That comma does not appear in the documentation, and is clearly needed.

Steve Denham

Re: LSmeans multiple control levels

DDK — Tue, 28 Apr 2015 06:38:40 GMT

Aha! Thanks Steve!!!! With the comma in there I now can get it to work.