LSMEANS in PROC GLM

[tka726]

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;

[data_null__]

Do you get something similar to this?

[tka726]

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 ;

[PaigeMiller]

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.

[sld]

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).

[tka726]

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

[PaigeMiller]

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