Thanks people, this toolbox is really helpful and easy to use.
I have one stats question -- forgive me that it is not a direct question about this toolbox but perhaps someone could help nonetheless.
I have repeated measures of circular data for multiple participants, lets say 15 participants where each participant contributes four angles. I have reason to believe that the angular distributions are going to be multipolar and not von Mises distributed. This would in essence require some kind of non-parametric repeated measures test which I am not sure has been developed for circular data. Is there a way to test for circular uniformity in this data set by using the procedures in the circstat toolbox, perhaps using some kind of p-value correction?
Mear - could you be more specific regarding your doubts about the wwtest?
The negative values for circ_mean are a result of the way circ_mean is implemented. If you prefer them to be between 0 and 2pi, just edit the function to provide the data in that format.
I'll update the von Mises function in a future release.
Just a quick question concerning the circ_mean function: I usually get the results in negative values, despite all the input angles being in positive degrees (conv. to rads). It's hardly a big deal to translate this to [0,360] degrees, but it is a bit annoying and seems unneccesary. Is this how it should be? I'm also getting some results for the wwtest which seem very wrong to me (but make sense in light of the negative mean values), and it's making me question the accuracy of this toolbox.
Marc, thanks, you are right.
I generated a von mises distribution with the mu and kappa estimated from my angles, say x, i.e.:
[mu kappa] = circ_vmpar(x)
vonmis = circ_randvm(mu,kappa,length(x))
Then I use the kuiper test to see whether the two distribution x and vonmis differ significantly (the difference can be in any property, such as mean, location and dispersion):
[H,pValue] = circ_kuipertest(x, vonmis)
However I was wondering if it is possible to have more accurate p-value estimates in the Kuiper test, as already asked by another user before.
thanks for your tip, however I'm not really convinced.
Both the circ_ktest and the circ_kuipertest are not described in the pdf:
Anyway, circ_ktest is a parametric two-sample test to determine whether two concentration parameters are different.
The circ_kuipertest is a two-sample test which allow to test whether two input samples differ significantly. The difference can be in any property, such as mean location and dispersion. It is a circular analogue of the Kolmogorov-Smirnov test.
I do not understand how these tests could help me with a goodness-of-fit test for the Von Mises-Fisher distribution, but probably is my limit.
Could anyone being of any help?