Have anyone estimated the power of an analysis with Gpower from data of pilot studies?.
I used three different sampling techniques (underwater visual census, baited cameras and fish traps) to estimate fish abundance inside and outside a series of marinas, and would like to know whether the power of each technique to detect a change of 25% and 50% is similar from their mean and variance obtained from a pilot study, for three univariate variables.
I am using Non-central F probabilities with a simple one-way ANOVA ('inside' vs. 'outside'). The problem is that when I set the mean and variance estimates for each technique what changes in the Gpower is the effect size, and therefore the curves (y-axis: power (1-beta); x-axis: n (sample size)) are the same for the three techniques.
Does anyone knows how to fix the effect size in Gpower, and what change is the mean and variance estimates for each techniques so the curves may differ?
All the best
Hello Nestor,
The G*Power manual (http://www.gpower.hhu.de/fileadmin/redaktion/Fakultaeten/Mathematisch-Naturwissenschaftliche_Fakultaet/Psychologie/AAP/gpower/GPowerManual.pdf), on p. 24 (section 10.1) explains how to determine the ES based on either sample means (with presumed common SD) or variances (effect variance vs within group variance).
If you already have a target ES in mind, then just enter that directly into the a priori dialog for one-way anova, and solve for the requisite N.
Good luck with your work!
Hi David,
Many thanks for the reply. I found it very helpful, and manage to find a solution for my problem.
All the best
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics