adtest2.m

function [H, adstat, critvalue] = adtest2(sample1, sample2, alpha)
% ADTEST2: Two-sample Anderson-Darling test of significant difference.
% This test is implemented for sample sizes larger than 8. For smaller
% sample sizes please refer to A.N.Pettitt, 1976 (A two-sample
% Anderson-Darling rank statistic) for the critical values.
%
% CALL: adtest2 (sample1, sample2);
% [H,adstat,critvalue] = adtest2 (sample1, sample2, alpha);
% Sample1 and sample2 are the samples to be compared. They must
% be vectors of a size greater than 8. Alpha specifies the
% allowed error. If alpha is not specified, a default value of
% 0.05 for alpha is used. Alpha must be either 0.01, 0.05 or 0.1.
%
% RETURN: H gives the statistical decision. H = 0: samples are not
% significantly different. H = 1: sample1 and sample2 are
% significantly different (i.e. do not arise from the same
% underlying distribution).
% adstat returns the ADstatistic of the comparison of the two
% samples. If adstat is greater than the critical value,
% the two samples are significantly different.
% critvalue returns the critical value for the alpha used
%
% (c) Sonja Engmann 2007

if nargin < 2, error('Call adtest2 with at least two input arguments'); end
if nargin < 3, alpha = 0.05; end

% Assignment of critical value depending on alpha
if alpha == 0.01, critvalue = 3.857;
elseif alpha == 0.05, critvalue = 2.492;
elseif alpha == 0.1, critvalue = 1.933;
else error('Alpha must be either 0.01, 0.05 or 0.1.');
end
samplecomb = sort([sample1 sample2]);
ad = 0;
for i = 1:length(samplecomb)-1
 m = length(find(sample1(:)<=samplecomb(i)));
 ad = ad + (((m*length(samplecomb) - length(sample1)*i)^2)/(i*(length(samplecomb)-i)));
end
adstat = ad/(length(sample1)*length(sample2));
if adstat > critvalue, H = 1; else H = 0; end