I've 3 points on statistics about interracial violence. The first two points are examples for general missunderstandings. In the third point I use the first two points at USA-Crime-Statistics from 2012-2013.
(1)
If the victum decision is done by RNG. The race-distribution of the victums will be equal to the race-distribution of the population. With 10 white and 1 black people, the RNG will attack 10:1 (whites:blacks) whites. It does not follow that RNG hates whites (… or is biased against whites).
(2)
In a 50% white and 50% black population:
You have a White who kills 5 Whites and 1 Black;
and you have a Black who kills 3 Whites and 15 Blacks.
The Black killed 3 times more Whites than vice versa.
Is the Black more violent? Yes.
Is the Black biased against Whites? I say no.
(3)
The image (NCFS-Table.jpg) shows the statistic. To calculate the racial bias I use:
C * pRNG * bias = ConX,
for a single race Y on race X case,
C - is the number of all crimes Y does,
pRNG - is the RNG propability to hit race X,
ConX - is the number of crimes that Y does against X.
For example, if whites do 100 crimes and 10 of them are against blacks, and blacks are 1/6 of the population, then the bias is 0.6. If in this case the RNG would attack blacks 100 times, the whites attack blacks only 60 times. The resulting matrix for all race combinations are ([W]hite, [B]lack, [H]ispanic, [O]ther):
ATTACKER
W B H O
W 1.33 0.62 0.82 0.89
B 0.27 3.10 0.35 1.38
H 0.44 0.83 2.28 0.84
O 0.89 0.82 0.68 1.59
The matrix shows that blacks attack whites 62 times, when RNG attacks whites 100 times; and whites attack blacks 27 times, when RNG attacks blacks 100 times. Additionally blacks attack blacks 310 times, when RNG attacks blacks 100 times. This represents point (2). Next, I get the interracial biases. Unbiased is bias=1, and innerrace is unbiased. Therefore I normalise the attacker bias by the innerrace bias:
W B H O
W 1.00 0.20 0.36 0.56
B 0.20 1.00 0.16 0.87
H 0.33 0.27 1.00 0.53
O 0.67 0.26 0.30 1.00
The symmetry shows that there is no bias between black and white. Interrace biases are eerie close.
To sum it up, my thesis is: The high frequency of black attacks on whites are due to the race-distribution of the population (1) and because of the more violent behavior of blacks (2); but there is no bias.
no guarantees
Statistics:
http://www.amren.com/news/2015/07/new-doj-statistics-on-race-and-violent-crime/
http://quickfacts.census.gov/qfd/states/00000.html
Matlab Code:
clear, clc
% http://quickfacts.census.gov/qfd/states/00000.html
% Pt = 318857056; % total population 2014 estimate
Rr = [0.621 0.132 0.174 0.073]; % race rate [White, Black, Hispanic, Other]
% http://www.amren.com/news/2015/07/new-doj-statistics-on-race-and-violent-crime/
Vt = 6484507; % total violence
VV = [4091971; 955800; 995996; 440741]; % number of violence victims
Var = [42.9 22.4 14.8 12.1]; % attacker race ratio for total violence
Varu = Var/sum(Var); % Var with unknown compensation
M = [ 56.0 13.7 11.9 10.6 % Matrix, attacker race ratio per victum race
10.4 62.2 4.7 15.0
21.7 21.2 38.6 11.6
40.3 19.3 10.6 20.3 ];
Mu = bsxfun(@rdivide, M, sum(M,2)); % unknown distribution same as known
p = bsxfun(@times,VV,Mu)./(ones(4,1)*Varu*Vt); % propability
B = bsxfun(@rdivide, p, Rr'); % Bias
Bir = bsxfun(@rdivide, B, diag(B)'); % Bias interrace (norm with innerrace)
