1. Kolmogorov-Smirnov vs. Kuiper test

Hello,

I would like to compare two 2D distributions with quantitative variables, illustrated here:

For each "x", they are several measures "y".
I can't assume these distributions are parametric.

A Kolgomorov-Smirnov test (p=0) and a Kuiper test (p=0.5) give rise to very different results.

Could you please tell me in which cases a Kolgomorov-Smirnov test is more appropriate than a Kuiper test, and vice versa?

Thank you very much

2. Re: Kolmogorov-Smirnov vs. Kuiper test

See http://cxc.harvard.edu/csc/why/ks_test.html for a good explanation. The K-S test is sensitive to shifts in the distribution (particularly around the median) while the Kuiper test is sensitive to changes in the tail areas.

3. The Following 2 Users Say Thank You to Miner For This Useful Post:

hlsmith (08-05-2015), recherche888 (08-05-2015)

4. Re: Kolmogorov-Smirnov vs. Kuiper test

How can I know what test should be done with my two distributions as I have no assumption about them?

5. Re: Kolmogorov-Smirnov vs. Kuiper test

That's hard to say. The two tests are essentially answering different questions. Which question is relevant to your analysis? Or, are both/neither relevant?

6. The Following 2 Users Say Thank You to Miner For This Useful Post:

hlsmith (08-05-2015), recherche888 (08-05-2015)

7. Re: Kolmogorov-Smirnov vs. Kuiper test

I'd like to test the null hypothesis according to which these two distributions are "the same".

I'd say both questions are relevant (or I wouldn't have reason for considering one of those more or less relevant than the other one).

What do you think?

8. Re: Kolmogorov-Smirnov vs. Kuiper test

Then your two tests appear to say that there are differences in the central region, but not near the tail areas. This corresponds to what a visual examination of the data appears to show.

9. Re: Kolmogorov-Smirnov vs. Kuiper test

Wikipedia indicates:
The trick with Kuiper's test is to use the quantity D+ + D− as the test statistic. This small change makes Kuiper's test as sensitive in the tails as at the median and also makes it invariant under cyclic transformations of the independent variable.
That seems to me different from what you wrote:
The K-S test is sensitive to shifts in the distribution (particularly around the median) while the Kuiper test is sensitive to changes in the tail areas.

10. Re: Kolmogorov-Smirnov vs. Kuiper test

My comment was based on the Harvard article for which I provided a link.

 Tweet

Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts