I've been putting some data through pfh's PyLevy, and I've been wondering how to see if the results are statistically significant.
To that end, at pfh's suggestion, I've run a hundred randomly-generated normally-distributed data sets through it, for data set sizes 500, 750 and 1000. They consist of two sections: the first section gives the return value of fit_levy() for each data set; the second section gives the alpha values sorted in ascending order.
Summary: for 1000 or more data points, an alpha of 1.98 or lower is significant. For 750 and 500, the threshold is about 1.96.
Of course, as pfh points out, this merely indicates that the data is non-normal. It doesn't say whether it is actually Lévy, or something altogether other...
⇦ I have seen the future... | ⇨ How will musicians live? |
