I'm trying to calculate Sample size for a three-way interaction using easypower.
In order to do that I use the example provided in RStudio :
(Exercise 8.14, p.397 from Cohen (1988)) and try to adapt it to my own situation which can be resumed as : main.eff.1
Hello Rachid,
What you have, by way of output, are the lower bound estimates for N (and cell n) in order to have a power of at least .80 to detect effects as small as the eta-squared values you supplied, if they exist in the population. Using these recommendations does not guarantee that there will be a specific main effect observed as statistically significant in your data set, nor that a specific interaction will be found to be significant.
Based on the values you used for the example you posted, it would appear to me that the minimum size you'd want for your study is 144 cases (assuming equal cell n; n = 12 per cell in a 3 x 2 x 2 design), so that your power would be at least .80 for all of the testable effects.
If that doesn't make sense, I'd urge you to consult with a statistician at your institution so that you have a clear picture of how a priori power analysis should guide your data collection process, given your specific research questions.
Good luck with your work.
Hello David, thank you much for your answer.
Does it mean that I need 144 participants in total (12 participants x 12 cell)?
As an extension of this, to get a power of .80 for the interaction Groupe*frequence*Predic, I need 6 participants by cell, thus in total 12x6=72 participants?
Effectively, it is quite difficult for me to understand those aspects when I put them into perspective with variables used my experience.
I 've three group of participant (Adults, 3rd graders and 5th Grader) who read word that could be high- or low-frequency and also at each predictable or unpredictable. They read 18 word in each of those 4 conditions (2x2). I really don't understand how linked my design with results provided by easy power
Rachid Khelifi – why 80% power? That's a guaranteed 20% risk of not finding an effect that is actually present. Why would you bother running research with those odds of failure?
I would consider 90% power as a minimum ethically acceptable standard for research that involves living creatures, given that we have a moral responsibility not to waste lives or the time of others in the conduct of a piece of research that has a very significant chance of failing to find something important.
Of course, if your research is unimportant, will not bring any benefit to anyone except you, then it's unethical anyway.
I know that defenders of 80% power say "that's what we've always done". It's not an ethical defence.
Hello again Rachid,
To answer your follow-up question, yes; 144 it is (to cover _all_ of the effects with at least the target power).
Good luck with your work.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size