topic Re: how to get p-value from pairwise comparisons in Statistical Procedures

how to get p-value from pairwise comparisons

echoli — Wed, 24 May 2017 16:42:20 GMT

Hi All,

I am trying to do pairwise comparisons to see the difference of means within each two groups (I have four groups totally). But in the SAS output below, the last column shows which group comparison is significant or not. But I want to get the p-value of the last column, not just star signs. Any idea?

Here is my output Table:

Comparisons significant at the 0.05 level are indicated by .*
group Comparison	Difference Between Means	Simultaneous 95% Confidence Limits
(1) Toddler - (2) Young	8.862	0.592	17.131	***
(1) Toddler - (3) Child	16.756	8.674	24.838	***
(1) Toddler - (4) Teen	20.183	11.744	28.621	***
(2) Young - (1) Toddler	-8.862	-17.131	-0.592	***
(2) Young - (3) Child	7.894	0.574	15.214	***
(2) Young - (4) Teen	11.321	3.608	19.033	***
(3) Child - (1) Toddler	-16.756	-24.838	-8.674	***
(3) Child - (2) Young	-7.894	-15.214	-0.574	***
(3) Child - (4) Teen	3.427	-4.084	10.938
(4) Teen - (1) Toddler	-20.183	-28.621	-11.744	***
(4) Teen - (2) Young	-11.321	-19.033	-3.608	***
(4) Teen - (3) Child	-3.427	-10.938	4.084

Here is my code:

proc anova data=combine;
class group;
model variable = group;
means group / BON;
run;

proc glm data=combine;
class group;
model variable = group;
means group / bon;
run;

Thanks,

Re: how to get p-value from pairwise comparisons

Rick_SAS — Wed, 24 May 2017 17:06:38 GMT

use

lsmeans group / pdiff adjust=bon; /* or another adjust= option */

Complete example:


proc glm data=sashelp.cars;
class type;
model mpg_city = type;
lsmeans type / pdiff adjust=bon;
run;

The doc shows many ways to adjust the p-values for multiplicity. I like Bonferonni and SIMULATE.

Re: how to get p-value from pairwise comparisons

echoli — Wed, 24 May 2017 18:47:18 GMT

Hi Rick,

Yes, I got this way before, but in this way, SAS outputs P-value table in as a matrix (as shown below). Then I have to write a long code to reorganize the p-values along with group comparison so they can display as one column (as the last column in the table shown in my question). I think that's a complex way....

Least Squares Means for effect group Pr > \|t\| for H0: LSMean(i)=LSMean(j) Dependent Variable: Variable
i/j	1	2	3	4
1		0.0047	<.0001	<.0001
2	0.0047		0.0045	0.0001
3	<.0001	0.0045		0.2270
4	<.0001	0.0001	0.2270

Any way to show the p_values in the last column instead of star signs like in the below table directly?

Comparisons significant at the 0.05 level are indicated by .*
group Comparison	Difference Between Means	95% Confidence Limits
(1) Toddler - (2) Young	8.862	2.732	14.992	***
(1) Toddler - (3) Child	16.756	10.764	22.747	***
(1) Toddler - (4) Teen	20.183	13.927	26.438	***
(2) Young - (1) Toddler	-8.862	-14.992	-2.732	***
(2) Young - (3) Child	7.894	2.467	13.320	***
(2) Young - (4) Teen	11.321	5.604	17.038	***
(3) Child - (1) Toddler	-16.756	-22.747	-10.764	***
(3) Child - (2) Young	-7.894	-13.320	-2.467	***
(3) Child - (4) Teen	3.427	-2.141	8.995
(4) Teen - (1) Toddler	-20.183	-26.438	-13.927	***
(4) Teen - (2) Young	-11.321	-17.038	-5.604	***
(4) Teen - (3) Child	-3.427	-8.995	2.141

Thanks,