We sometime face a situation of small sample size. If the assumptions of the pearson correlation are fulfilled, is it valid to calculate correlation?
Are bayesian tests not dependent or less dependent on sample size?
Sure, you can compute Pearson (or non-parametric) correlation provided that data is of a good quality. Since you have only small sample, I assume you expect to see a large effect size (coefficient). Otherwise, you will not have sufficient power to evaluate the relationship using statistical testing. Not trained in Bayesian but I suspect that this will not resolve the sample size issue :)
Indrajeet Indrajeet , Martin Marko , in fact Bayesian methods are suitable for small samples, because -in contrast to classical frequentist statistics- Bayesian estimators work well in small and large samples. I personally like this approach:
Article Estimation of the correlation coefficient using the Bayesian...
On the other hand, David (1938) suggests n >= 25 for a proper calculation of Pearson correlation:
David, F.N. (1938). Tables of the ordinates and probability integral of the distribution of the correlation coefficient in small samples. Cambridge: Cambridge University Press.
Rolando Gonzales thanks. r = 0.55, infact post hoc power analysis shows we require 24 participants. we had 20 but few were filtered as outliers. p is around 0.009 and BF is around 8. I am confident because I found same correlation in another data set N=30. Since the study was the first of its kind, i could not decide a priori sample size.
Could you please provide few references suggesting small sample size is relatively fine if bayesian approach is chosen ?
At current situation of corona, we can collect more data , so we are stuck.
- Badges
- Science topic
- Similar topics
- Biological Science
- Neuroscience