What you've done is shown that Serbs are convicted more often than you'd expect in random experiments.

No he hasn't. That's a statistical mirage, because he arbitrarily classifies 7 of the Serbian suspects as being related to Kosova (rather than, say, split them evenly among the three wars, or split them among the three wars in proportion to the number of civilian casualties, both of which would have been more appropriate).

To understand why this is misleading, consider the outcome if I placed those seven indictees in Bosnia instead of Kosova: Then the vanilla average (which remains an inappropriate measure), goes from 3 per mille to 2.4 per mille. Purely from arbitrarily designating the 7 Serbian indictees for which there is too little information in the indictments to tell which wars they were indicted over.

And then he proceeds to happily make a vanilla average out of ratios that are up to a full order of magnitude different in both numerator and denominator. You're not allowed to do that. When you divide a very small number by another very small number, you get a higher uncertainty on your ratio than when you divide a very big number by another very big number. One or two wrongfully indicted Serbs from the Croatian war - or one or two hundred civilians miscounted - would count far more towards the ratio than a similar number for Bosnia.

So the three ratios cannot be given equal weight. You avoid all these problems by using a weighted average. But of course, the "signal" goes away when you do that...

- Jake

Friends come and go. Enemies accumulate.