Non-parametric test for trend suggestions needed


I am conducting trend analysis on rainfall time series using the Mann-Kendall rank correlation coefficient (tau). I would like to also use another non-parametric test to increase the reliability of the results, but I am having trouble finding a suitable test.

I don't have a dataset that can be used for comparison purposed with the Mann-Whitney-Wilcoxon test, and I have read that the power of Spearman's rho test is almost identical to that of the Mann-Kendall test. I read that here:

My rejection of the Spearman rho test is based on my understanding that an identical power makes the use of both tests invalid. Is this correct? If so, can anyone suggest another test I could use instead?



New Member
hi cohara, what do you mean by increase reliability of results? If you mean to see that the tests agree, then I'd advise against it in general; there are situations where I guess it's fine, like as a learning exercise (I've done that, I'm sure many people have), but often this might not be a good idea. Most analysts have seen situations where two essentially equally appropriate tests give results that disagree. This might happen when the significance is borderline. If you're in a real life situation, especially one involving human subjects, ethics would implore you to examine the discrepancy between the two tests and clearly expain why they disagree.


Super Moderator
As a learning excercise, you should try spearmans to comapre the results. Also, given that rainfall is highly seasonal, you should try seasonal Mann Kendall and perhaps a generalised addive model (if your data series is long enough). Also, why not try linear models and compare them to the non-parametric appraoches? Are the residuals from your rainfall data normally distributed (would a pearsons correlation give you the same answer)?