Dear reader,
I am working on a paper where I want to report effect sizes for the
association between nominal/ordinal variables and interval variables. I
only report bivariate associations.
One of the reasons that I report effect sizes is so that I can compare
association strengths of variables of different measurement levels (e.g.
ordinal and ordinal, ordinal and interval, and dichotomous and
interval).
I am in doubt as to whether to use eta-squared (or partial eta-squared;
since I only report bivariate associations, these are the same anyway)
or omega-squared.
As I understand it, omega squared represents the proportion explained
variation in the population, whereas eta squared represents the
proportion explained variation in the sample.
Although I would prefer to report the population estimate, I think that
the other effect size measures I report (Cramer's V and Cohen's D)
estimate sample parameters rather than population parameters. This would
obstruct one of my main goals (i.e. comparison across measurement
levels).
Does anybody know whether Cramer's V and Cohen's D estimate sample or
population parameters?
Thank you in advance, kind regards,
Gjalt-Jorn Peters
Work & Social Psychology, Faculty of Psychology and Neuroscience
Maastricht University, The Netherlands
|