I am conducting a simulation study on the power of randomization tests
for two-independent groups. I evaluate also the power of
the Student-t test, for comparison purposes.
In some conditions, the data for the two groups are simulated from
distributions with the same standard-deviation. For these
conditions, I used an effect-size defined as the difference between
the two population means, in population standard-deviation
units.This is Cohen’s d measure of effect-size: (μ1 - μ2)/σ.
In other conditions, the variability of the two simulated
distributions is different. For example, the data for the first group
is simulated from a distribution with σ = 1 and the data for the
second group is simulated from a distribution with σ = 3.Now, how
is the effect-size defined? It is not an estimation problem, because
the standard-deviation of each distribution is known.
Supposing I have one experimental and one control group, one possible
solution is to define it as the difference between the two
population means, in control group standard-deviation units. This is
Is this acceptable? Are there better solutions? What is the usual
measure of an effect-size in simulations studies - I couldn’t find
an answer in the literature I reviewed.
Can anybody advise me?
Thanking in advance,
[log in to unmask]