Dear SIr,
I had tried to analyse a 2 X2 factor experiment with PROC GLM. I reported the P-values for A, B and AB interaction.
My SAS syntax was of this fashion:
proc univariate plot normal;
title 'Trial4Digest';
var v1;
run;
Proc GLM Order = Data;
Class A B;
Model v1 = block A B A*B;
LSmeans A*B/TDIFF PDIFF P STDERR Adjust=tukey;
Run;
Quit;
1. Where interaction was NOT SIGNIFICANT, I had used the simple effect of each factor to compare my means and report in result and discussion.
2. Where interaction was SIGNIFICANT and either FActor A or B is Sig, or FActor A or B is NOT SIGNIFICANT, i had used tukey test to assign "abcd" to the means for comparisons.
|
|
A1
|
A2
|
|
P-values
|
|
3Parameter
|
B-
|
B+
|
B-
|
B+
|
SEM
|
A
|
B
|
A*B
|
|
Ac
|
58.4
|
51.5
|
54.2
|
55.7
|
0.71
|
0.86
|
0.02
|
˂0.01
|
|
Pro
|
20.5b
|
23.3a
|
22.5a
|
23.2a
|
0.39
|
0.07
|
<0.01
|
0.05
|
|
Bu
|
15.5
|
18.8
|
17.5
|
15.8
|
0.37
|
0.40
|
0.19
|
˂0.01
|
|
Bra
|
3.66
|
3.96
|
3.80
|
3.30
|
0.10
|
0.19
|
0.54
|
0.04
|
|
Val
|
1.95b
|
2.39a
|
1.98b
|
2.00b
|
0.06
|
0.02
|
<0.01
|
0.01
|
|
Ap
|
3.03a
|
2.25b
|
2.46b
|
2.43b
|
0.09
|
0.175
|
0.01
|
0.01
|
This is the response of the reviewer:
" The use of Tukey's test to separate means is incorrect. Actually, it is not acceptable to use two statistical tests to show a difference between treatments (treatments and interaction effects and mean separation). I greatly discourage the authors from using mean separation when the effect of treatment (factor A, factor B and interaction) is displayed in the tables. A slice test can be used when the interaction is observed. Therefore Tables and text (where appropriate) should be corrected".
I do not have a clue on the way forward on using this slice test and how to interpret the subsequent result. Any help to modify my syntax and report correctly will be appreciated.
Thank you
