If you are truly interested in using mathematics to convery an observation with some accuracy, please run a monte carlo simulation and see how the results turn out. What YesYNOT is suggesting is you are using unequal population samples. You need to account for the real population size of each cohert and the event frequency (murder rate) for each cohert. If there is a random distribution (no bias) of whites killing blacks and blacks killing whites, (cohert combinations) the actual kill rate combinations will be influence by the cohert's relative size. If whites kill more often than blacks, or vice versa, then the draw rate (for the combination) is also different.
Sounds like a good excercise for a curious mind.