Which SD to use when calculating standardised mean differences?

I have a question regarding the calculation of effect sizes of continuous outcomes (cohen’s d or hedges g) in clinical trials. This is for both reporting effect sizes in my own publications, or for the calculation of effect sizes for meta-analysis

Everything i have read explains that the calculation of these effect sizes is the differences in means between the two groups, divided by the pooled standard deviation. However, there is no clarification as to which difference (post values, Δint - Δcon), and which standard deviation (baseline SD, SD of the change score, post-SD).

Typically, we have used the mean difference (Δint - Δcon)/Pooled baseline SD. The assumption here is that we are presenting the effect of the intervention on an outcome based on the observed distribution of that outcome in the sample. However, I commonly see other authors using the SD of change scores. This produces a very different effect size.

I have searched and searched, and have yet to find an answer on this. Perhaps there isn’t one? Can someone provide some clarity on this for me if possible?


you likely want an estimate that corresponds to your analysis, and you would likely use ancova for the analysis, thus you probably want (what SAS calls) ‘least squates means’. This guy describes it well using a basic example: http://onbiostatistics.blogspot.com/2009/04/least-squares-means-marginal-means-vs.html