3
$\begingroup$

I have a sample of outcomes of my algorithm and would like to test to some degree of confidence if it differs from zero.

I know I can use a 1-sample t-test (as in https://docs.scipy.org/doc/scipy/reference/generated/scipy.stats.ttest_1samp.html), but I am wondering: does a non-parametric analogue exists?

I seem to understand that Mann Whitney U and Wilcoxon signed rank both work on 2 samples. Is this the case?

(questions are in bold)

Thanks in advance for any help :)

$\endgroup$
  • $\begingroup$ 1. Unless all your values are 0, the sample outcomes definitely differ from zero. 2. What are your outcomes? (what kind of variable is it?) 3. It's important to be clear about what your hypothesis is in order to be able to choose an appropriate test. (Note that hypotheses about about population quantities rather than samples) $\endgroup$ – Glen_b -Reinstate Monica Mar 31 '17 at 8:26
1
$\begingroup$

Wilcoxon test has two flavors: one sample test (known as Wilcoxon signed rank test, and can be applied either on one sample or on the difference between two paired samples) and two-sample test (known as Mann-Whitney test). So you can use one sample test version of it. In R you can check wilcox.test and letting only x parameter to have values.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.