effect size (Cohen's d) for two group repeated measures model

[Andy Grogan-Kaylor]

Hello, is there a way to use Stata to calculate effect sizes for a two group repeated measures model: e.g. pre-test and post-test for a treatment and comparison group? esize (https://www.stata.com/manuals/resize.pdf) appears to do this for a cross sectional model comparing treatment and control, but not for a longitudinal repeated measures case. In case it's helpful, I have in mind the formulas shown at https://www.campbellcollaboration.or...-with-pretests. I have no problem hand coding these calculations, just wondering if there is a Stata command that can do this already. Many thanks.

[Tiago Pereira]

Hi, Andy.

I don't think we have any Stata program that implements those formulas (i.e., effect size calculations, such as d or g, for dependent data).

All the best,

Tiago

[Andy Grogan-Kaylor]

Thanks Tiago!

[Dirk Enzmann]

I am also not aware of any Stata program for Cohen's d or Hedges' g for paired (dependent) data.

However, I am suspicious whenever I read "Cohen's d" because often it is confused with Hedges' g (see here). It would be better to denote them as biased or unbiased effect size measures, instead. Additionally, the choice of the standardizer (sd of differences, pretest sd, or average of pretest and posttest sd) has to be considered carefully. Cumming (2012, p. 291-292) argues to use the latter.

Reference:
Cumming, G. (2012). Understanding the New Statistics: Effect Sizes, Confidence Intervals, and Meta-Analysis. New York, NY: Taylor and Francis.

[Andy Grogan-Kaylor]

Thanks Dirk for your thoughts. I agree that choosing the standardizer seems to be a key issue. I'll make sure to look at the Cumming reference.

[Tiago Pereira]

Thank you, Dirk. I have put your paper on my to "read list". I like the term "Babylonian Confusion".