I am designing a study to examine the impact of manipulating instructions on memory performance. There are 5 different instructions and each subjects receives these instructions, making it a within subjects design and memory performance is the dependent variable. I would like to conduct a priori power analyses to determine a reasonable sample size. I have looked into programs like g power, but I have some questions. 1) My design differs from the designs in the prior literature. The prior papers I have been reading are all between-subjects design and often don't report effect sizes. If data is available for me to calculate these effect sizes, is it okay for me to use them in my power analysis given the different design? Say the study reports cohen's d, but the effect size g power asks for is f. Is there a way to convert these? 2) Would it be better to just base my sample size on the typical size per group used in prior studies and try to match that?
A priori power analyses to determine sample size for repeated measures design
- Thread starter staceofspades
- Start date
- Tags a priori power repeated measures anova
The difference between the effect size Gpower calculates for a between-subjects design and a repeated-measures one is that Gpower takes into account the correlation between the intervals of repeated measuring. So a sample similar to previous studies or a little larger might be approached, but knowing that it is not a correct way, although I agree it is a method very very used by authors to claim validity of their results and used by reviewers to criticize sample sizes smaller than previous ones.
I don't know about converting the effect size reported by Gpower to other measures. But I would use its own effect size calculated by the data you entered as your pilot study. Although F distribution does not fully apply to your case (since your data are apparently ordinal), but I think a repeated-measures ANOVA is the closest thing to the repeated-measures Friedman you need. So F might be acceptable with some precautions.
I don't know about converting the effect size reported by Gpower to other measures. But I would use its own effect size calculated by the data you entered as your pilot study. Although F distribution does not fully apply to your case (since your data are apparently ordinal), but I think a repeated-measures ANOVA is the closest thing to the repeated-measures Friedman you need. So F might be acceptable with some precautions.
