George Garrity
Sep 3, 2018 · 1 min read

Interesting piece. I’ve found the KS test to be highly useful in a number of settings when I need to make comparisons between two or mor non-parametric distributions.

I find you use of a KS-matrix interesting, although I’m not certain it would be useful in some settings (e.g. comparison of distributions that might change over time). I am, however, curious about your subsequent cluster analysis. In that, you show only one outlier whereas you have three in the filtered network analysis. This suggest that there may be some distortion of the data that each method imposes on the KS matrix. Have you compared the output of the two methods to determine if you obtain comparable results? If you use different clustering methods, does that change the result?

    George Garrity

    Written by