topic Re: LSMEANS in PROC GLM in Statistical Procedures

LSMEANS in PROC GLM

tka726 — Fri, 29 Mar 2019 19:16:27 GMT

Is there an option in LSMEANS of PROC GLM which will compute pairwise p-values?

In my program, the variable 'surgery' has 15 categories. I want to see pairwise differences & p-values between them all. I am trying to avoid writing many estimate statements.

proc glm data=ads;
class surgery ;
model event=surgery/solution ss3;;
lsmeans surgery / cl adjust=tukey stderr pdiff tdiff ;
means surgery / tukey alpha=0.05 cldiff;

I can get the differences but not the p-values I want.

I'm trying to see if I can get the p-value from such an estimate statement, without typing out each comparison. I'm interested in the p-value from each comparison, not accounting for others.

estimate 'surgery 1 vs 2' surgery 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 ;
estimate 'surgery 2 vs 1' surgery -1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 ;

run;
quit;

Re: LSMEANS in PROC GLM

data_null__ — Fri, 29 Mar 2019 20:45:00 GMT

Do you get something similar to this?

Re: LSMEANS in PROC GLM

sld — Fri, 29 Mar 2019 22:02:10 GMT

If you do not want to adjust for experiment-wise Type I error, then remove "adjust=tukey" from the LSMEANS statement (while keeping in mind that there are 105 pairwise comparisons among 15 means, and only 14 degrees of freedom).

Re: LSMEANS in PROC GLM

tka726 — Tue, 02 Apr 2019 18:03:50 GMT

Thanks! This does not give the p-value for each comparison though? Is there a way to get each p-value?

Re: LSMEANS in PROC GLM

PaigeMiller — Tue, 02 Apr 2019 18:11:33 GMT

The p-value is the bottom number in each cell of the table.

Re: LSMEANS in PROC GLM

tka726 — Tue, 02 Apr 2019 18:30:32 GMT

I do, but why do these p-values not match what I get if I write out contrast statments?

I.e. these are the comparison p-values I want.

estimate 'surgery 1 vs 2' surgery 1 -1 0 0 0 0 0 0 0 0 0 0 0 0 0 ;
estimate 'surgery 2 vs 1' surgery -1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 ;

Re: LSMEANS in PROC GLM

PaigeMiller — Tue, 02 Apr 2019 19:05:15 GMT

They don't match because the LSMEANS comparisons uses Tukey adjustments to the p-values (that's what your code asked for), but your estimate statements have no such adjustment.