Re: Question regarding Effect Sizes

On Sat, 24 Feb 2007 20:47:18 EST, quebecstat <probameasure@xxxxxxx>
wrote:

A researcher recently asked me to do a piece of work for him.

He has data (one variable about pain) on subjects from three groups: two treatment groups and one control group. There are two timepoints.

He said he is interested in computing "within group and between group effect sizes". That does not make much sense to me as an effect size is for me the effect of a treatment.
I am thinking I can compute
1/The effect size of the first treatment group with the control group.
2/The effect size of the second treatment group with the control group.

I read that an effect size is computed as a standardized difference between two means. But here, I have two timepoints. So, how should I compute those effect sizes?

The Researcher has a valid question. If there are two time
periods, then there *is* a within-group effect size for each group.

In addition, there is a between-group effect size. There is little
problem if you are expressing effect sizes in the natural units
of the analysis -- seconds, miles, etc.

Depending on what-all is being said, it might be desirable to
express standardized effects in two different ways.

On the one hand, there are "changes" in each group. Those can
be quantified in terms of the common standard deviation of changes.
That yields the effect size that is important for the statistical
significance of the test, or for a future a power analysis.

On the other hand, the magnitude of an effect may be more
familiar or more descriptive when it is described in terms of
the standard deviations of the pooled scores, or sometimes,
if variances are different Pre and Post, in terms of the
pooled Pre-scores.

You can do it either way, or both ways, but be sure to be clear
about what you are doing.

Should I first compute the improvements in pain (an improvement can be negative) from time 1 to time 2, then compute the mean improvements for each group, then compute the standardized difference in mean improvement and call that the effect size?

many thanks for your help.


--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.

Relevant Pages