I have several datasets with angular measurements, i.e. circular values from 0 to $2\pi$. These datasets tend to have peaks at 0 and/or $\pi$, and I need to tell if the peaks are detected/significant. There are several approaches to related problems I found, but they don't seem to answer the question of individual peak (mode) significance:

  • Fitting a mixture model with a uniform component + 2 gaussians at 0 and $\pi$, and getting the posterior for the parameters. The posterior means for weights of each component are basically always distinct from zero , even when fitting to a uniform distribution. So this cannot be used to assess peak significance.
  • Different multimodality tests: only test for the number of modes, and don't give their locations.

So I'm looking for help and suggestions - what are better ways to do this?

  • $\begingroup$ Since you want to perform a hypothesis test, can you say more specifically what your null hypothesis is? $\endgroup$ – user20160 Jul 18 '19 at 13:24
  • $\begingroup$ @user20160 there is not a single hypothesis to test here, but a combination of two: "is there a mode at 0" and "is there a mode at pi". Corresponding null hypotheses are that the distribution has no mode at those locations. $\endgroup$ – aplavin Jul 18 '19 at 13:38

Your Answer

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

Browse other questions tagged or ask your own question.