effect size for paired samples

[Iryna Lobach]

This command is very useful in the independent samples setting http://www.stata.com/stata13/effect-sizes/.
And I am looking for an equivalent in the paired samples, what might be viewed as one sample setting - what is important to me is both the estimate of an effect size and 95% confidence interval. Please recommend what to use.

[Roman Mostazir]

A little example of the data would have been more helpful. If you have paired data on two variables i.e. case id=1 and values are paired on variable 'x1' and var 'x2', you still could use the 'unpaired' option but include a condition that cases with missing in either variable should be dropped. That way you are actually using a paired comparison and using the same amount of data for both variables to calculate the standard deviation and the mean, thus the 'd' family effect size (assuming that is what you want). Elaborating the example:

esize unpaired var1==var2 if !missing(var1) & !missing(var2),all

[Iryna Lobach]

Thanks!

I was wondering about denominator of the effect size that should probably incorporate dependence between paired samples (e.g. baseline and follow-up measurements on the same set of individuals), and the corresponding standard errors. Just found an answer and references here: http://analytics.ncsu.edu/sesug/2012/SD-06.pdf, page 3