Hi,

I think reporting effect sizes are more informative than computing post-hoc power.

But to answer your original question, rather than trying to determine power for an entire model, focus on the contrasts of interest, which can often be conceptualized as standardized mean differences.

Hope this helps.

Scot

Dear Scot,

thank you for your answer!

Of course, I intend to present effect size measures as well. The reason for why I would like to do a power analysis is, that for some outcome measures no effect of the treatment could be shown. So I'd like to compute post-hoc power to see, if this no-effect is possibly based on a too weak design (e.g., too small n).

Standardized mean differences: Maybe I'm mislead, but is this term not a synonym for effect size like Cohen's d?

(see: Article Interpreting estimates of treatment effects: Implications fo...

Best wishes,

Daniel

Try to use SPSS, it has options for computing effect size.

Dear Mustafa,

thank you for your answer! Computing effect sizes was not the problem, though. But I found an option for computing "observed power" as well in the GLM-Menu of SPSS. I just oversaw it before because I'm running the program in German language...

Related question: How can I compute a a-priori power analysis for the scenario described above? Because the post-hoc is actually not very informative or useful (as I found out and several authors have mentioned, see, e.g., http://daniellakens.blogspot.com/2014/12/observed-power-and-what-to-do-if-your.html), I now try to estimate the power at which my design and sample size is/was able to detect a small, but still interesting effect (defined, e.g., as eta squared = .01).

Unfortunately I couldn't find any procedure for that calculation yet. Again, G*Power is of no help here, but also, e.g., D'Amico et al. (2001)'s paper does not cover this design.

Maybe anyone has a better overview in this issue than me?

Hi Daniel Spitzenstätter ,

Were you able to solve the problem? I am also interested in performing a post-hoc power analysis. My design is a 2x2 design (1 within factor, 1 between factor) with 1 covariate.

Cheers

Hi Daniel Spitzenstätter

Did you find a proper software for a-priori power analysis for mixed design ANCOVA? I'm also interested in power analysis for this design ( 1 wth subject factor (e measurements, 1 btw subject factor (2 group) and 1 covariant).

Earlier you mentioned "observed power" command in SPSS which I also saw and run it. What do you think, is observed power proper for this design? Also, is eta square is proper for effect size (for each F test in the design)?

I used G*Power 3 app for Windows in which there is only ANCOVA option (both sensitivity and a priori power analysis). I'm not sure if it is the right power analysis.

I hope you answer.

Best wishes,

Vahide.