z-scores in qmtraffic may not be appropriate for measures that aren’t normally distributed

[dkp]

Thanks to Shashank Yadav for pointing out this issue! You can use qmviolin to get a sense of which measures are normally distributed…but we need a better solution.

[hickst]

Please provide better explanation of the problem in the issue comment.

[dkp]

Thanks to Shashank Yadav for pointing out this issue!

Z-scores are not a great solution for measures that are not evenly distributed around the mean (Chalk the choice of z-scores up to my lack of statistical chops). There are statistical tests of normality that can help assess the normality of the distribution and hence the appropriateness of z-scores in any particular case (Shashank suggests the Kolmogorov-Smirnov Test or Shapiro-Wilk Test).

In the mean time, one can use qmviolin to get a sense of which measures are normally distributed…but we should consider a better solution for qmtraffic.