For my research I have two sample sizes; one consists of 195 cases and the other of 31. The rho of the 195 cases is much bigger then the rho for the 31 cases. I tried to search for the minimum sample size for Spearman, but almost everything relates to Pearsons R.
Does anyone has an idea of how big the sample size needs to be for Spearman, to get good results.
Hi,
You might find useful the paper of Douglas Bonett & T.A. Wright. It 's available for free download at ResearchGate:
Article Sample size requirements for estimating Pearson, Kendall and...
Good luck with everything.
Hi,
Pearson's product-moment coefficient is computationally identical to Spearman's rank correlation coefficient, therefore you can use criteria for Pearson correlation to determine sample size for Spearman's rank correlation using G Power software.
If you do not have G Power, you can look up the online tutorial available on YouTube or the online manual for G Power to conduct a priori power analysis.
Moreover, you might find this link useful :
https://www.statisticssolutions.com/spearman-correlation-2-tailed/
Any amplification in Pearson due to small sample size will be propagated with Spearman.
I mean this as a serious question: What does it mean "to get good results" ? That is, are you thinking about the rho value is needed to get a certain p value given a certain sample size ? Or is the question something different?
Salvatore S. Mangiafico The question is if this small sample size (n=35) is reprecentable for my research. The outcome "good" or "bad" doesn't really matter. It's just about how to get a better rho and what is needed to change for future research to get this. As the article of Rohali Keesari states that there needs to be a samle size of at least 29 for a large effect size. This number only increases for medium and effect size. This could mean that there are enough cases or there should be more cases to conduct the study on. This depends on how great of and effect you want.
Thanks for the reactions!
To get good results with Spearman correlations, the sample size must be less than 30. With your samples size, the Pearson correlations is more adequate
One thing that may help here is to calculate the confidence interval for rho. There may be a direct calculation for this , but otherwise it can be determined by bootstrap.
In theory, if you have a good sample of your population, an effect size statistic like rho shouldn't change with increasing sample size. But of course, you need a sample large enough to get a "stable" or representive statistic. The confidence interval should give a sense of how representative your statistic is.
- Badges
- Science topic