topic Re: Proc glimmix estimate statement - one tailed p value not displaying in Statistical Procedures

Proc glimmix estimate statement - one tailed p value not displaying

didymo — Fri, 28 Apr 2023 19:49:13 GMT

Below is the code I am using. I am specifying a one tailed test for the contrasts in the estimate statement but it is not displaying the upper one tailed p value in the Estimates table. Looks like it is the default two-tailed p value for the contrasts.

Any ideas as to why? Thanks, Brad

ods output lsmeans=larvae_order_MEANS_lognormal;
proc print data=larvae_order_MEANS_lognormal;
proc sort data=library.larvae_order_MEANS; by order;
title 'GLIMMIX LOGNORMAL REML - LARVAL ORDER DENSITIES - BLOCKED';
proc glimmix data=library.larvae_order_MEANS method=LAPLACE; by order;
class type subsite pair year;
model MEAN_density_m2=type year year*type / solution dist=lognormal link=identity;
random pair pair*type;
lsmeans year*type / adjust=tukey CL ilink diff=all;
lsmeans type / adjust=tukey CL ilink diff=all;
estimate 'treated restored change - untreated restored change' year*type 0 0 1 -1 -1 1 / upper;
estimate 'treated restored change - reference change' year*type 1 -1 0 0 -1 1 / upper;
estimate 'untreated restored change - reference change' year*type 1 -1 -1 1 0 0;
run;
ods output / reset;

Re: Proc glimmix estimate statement - one tailed p value not displaying

sbxkoenk — Sun, 30 Apr 2023 13:08:39 GMT

Instead of halving the p-value of the two-tailed test (halving method) , the ESTIMATE statement in many procedures can provide the one-sided p-value directly by specifying the UPPER or LOWER option as appropriate.

What makes you think you are looking at a two-sided p-value?

Usage Note 24094: Testing the null hypothesis that a parameter (or parameters) is equal to, greater than, or less than a constant
https://support.sas.com/kb/24/094.html

Thanks,
Koen

Re: Proc glimmix estimate statement - one tailed p value not displaying

StatDave — Sun, 30 Apr 2023 15:22:02 GMT

For the 2 of 3 of your ESTIMATE statements that have the UPPER option, what makes you think you are not getting a one-sided p-value? Is the p-value label "Pr>z" (for one-sided) or "Pr>|z|" (for two-sided)?

Re: Proc glimmix estimate statement - one tailed p value not displaying

didymo — Sun, 30 Apr 2023 16:58:27 GMT

Thanks. Good point but "Pr t" is all that is provided in the Estimates table, which I assume is two tailed. Using proc mixed for other response variables, the same estimate statements produce an Estimates table has a column with Upper, Upper, Both, which seems correct, but I can not get glimmix to produce a similar table. Perhaps it is the method or link I using? I need the power of the one tailed test for other response variables, not the example below.

Unless I am wrong, the contrast statements (e.g., 0 0 1 -1 -1 1) are not testing a directional difference, but a non-directional difference, right?

Thanks

Re: Proc glimmix estimate statement - one tailed p value not displaying

StatDave — Sun, 30 Apr 2023 17:16:44 GMT

That label reflects the fact that you have some ESTIMATE statements with UPPER and some without. As a result the label cannot indicate one- or two-tailed because you have both. However, the p-values from the statements with UPPER are one-tailed, and the p-value from the statement without UPPER is two-tailed. You can verify by simply rerunning omitting UPPER to see the change in the p-values.

Re: Proc glimmix estimate statement - one tailed p value not displaying

didymo — Sun, 30 Apr 2023 23:48:28 GMT

Got it. Thanks, However, this is not the problem. GLIMMIX with ESTIMATES statement and '/ upper' does not produce a one tailed p value, which I am trying to resolve.

Re: Proc glimmix estimate statement - one tailed p value not displaying

StatDave — Mon, 01 May 2023 01:23:01 GMT

As I mentioned in my previous reply, the ESTIMATE statements with the UPPER option, ARE one-sided. What evidence do you have that they are not?

Re: Proc glimmix estimate statement - one tailed p value not displaying

didymo — Mon, 01 May 2023 02:44:02 GMT

Oops, they are, but the output could be clearer. I removed the 'upper' and the p value label is: Pr > |t| with a larger p value! The one sided is denated 'Pr t' in glimmix I guess versus the output having a label column with upper and both in mixed. Cheers!