obtaining effect sizes for pairwise t-test in the one-way anova repeated measure context

[Nader Mehri]

I have a one-way repeated measure data fromat for which I am trying to compare the means for my outcome variable across three time points (0, 1, and -1). Using the below annova command as well as the subsequent postcommand, I have been able to obtain the effect size for the anova model. Also, the pwcomapre command returns the pair-wise t-tests for the three time points for my outcome variable of interest. I wonder how can I obtain the effect sizes for the pair-wise t-tests.


Here is my code:

anova outcome id time, repeated(time)
estat esize
pwcompare time, pveffects mcompare(bonferroni)

Here is a subsample of my data:

* Example generated by -dataex-. For more info, type help dataex
clear
input float(id time) byte outcome
 7 0 47
11 0 32
 5 0 35
 6 0  0
 4 0 40
 9 0  9
12 0 49
 3 0 12
 2 0 33
 8 0 16
13 0 19
14 0 36
 1 0 39
10 0 17
 6 1  9
12 1 31
 2 1 21
11 1 31
 1 1 33
 5 1 28
 8 1 19
14 1 16
 4 1  6
 7 1 26
13 1 10
 9 1 14
10 1 14
 3 1 23
11 2 26
 3 2 27
 8 2 30
 4 2 11
 6 2 12
 7 2 46
10 2 12
 5 2 35
 2 2 19
14 2 20
13 2 13
12 2 35
 1 2 35
 9 2 30
end
label values time time
label def time 0 "t-1", modify
label def time 1 "t", modify
label def time 2 "t+1", modify

[Joseph Coveney]

I wonder how can I obtain the effect sizes for the pair-wise t-tests.

This has come up on the list a couple of times before, for example, here and here.

I believe that the convention is to take the mean of the change scores and divide that by their standard deviation.

So, first reshape wide, then generate the pairwise differences, summarize them and display the quotient abs(r(mean)) / r(sd).