This may be a follow up on these questions: Pearson's or Spearman's correlation with non-normal data
How robust is Pearson's correlation coefficient to violations of normality?
What method is preferred, a bootstrapping test or a nonparametric rank-based test?
Since I am new here and cannot comment so I have to open a new question.
The discussion suggests Pearsons r does not assume normality of the data, yet I think the significance test required normal data right?
So I run bootstrapping to get a better estimate of r for my right-skewed data. At the same time I also run Kendall trying to see if there are any difference.
It turns out the significance test of both are pretty similar (i.e. those correlated sig. in bootstrap Pearson test usually also correlated sig. in Kendall), I am just thinking which values would be a better one for report.
Any thought? I am wondering if my data are nonnormal, is the bootstrap estimates of Pearson r reliable enough so maybe I can report that as seems the p value is more significant --- or would Kendall still be better?
(Or should I bootstrap Kendall?)
