Bug with computation statistical significance in CANDID #144
Description
There was a bug with the statistical significance computation in CANDID. If I remember correctly my offline discussion with @DrSoulain, the bug is not yet fixed in AMICAL. I just wanted to open an issue to track this. The bug has been fixed in CANDID, so the solution can probably be found somewhere in their code and copied in AMICAL relatively easily. The warning about this in CANDID's README is quoted below.
WARNING: before version 1.0.5 (April 2021), there is a bug in the computation of the statistical significance: the computation are correct below ~4 sigmas, but get progressively overestimated, to double by 14 sigma. In practice, this does not change results, as the significance is only important around or below 4 sigmas. Also, note that the number of sigma cannot go above ~8 with the new (and correct) formula.