fof-visually-guided/elutils/+stats/bootmedian.m

function varargout=bootmedian(varargin)
% [p,ci]=bootmedian(A,B,['boots'])
% When passed in a single vector, tests whether the median is different from 0
% When passed in two vectors does a permutation test to see whether the medians
%	are significantly different
% Optional inputs:
%   'boots'  the number of shuffles to perform (2000)


if nargin==0
    help('stats.bootmedian')
    return
end

if nargin>1 && isnumeric(varargin{2}) 
    A=varargin{1};
    B=varargin{2};
    one_dist=false;
    if nargin>2
        varargin=varargin(3:end);
    else
        varargin={};
    end
else
    A=varargin{1};
    one_dist=true;
    if nargin>1
        varargin=varargin(2:end);
    else
        varargin={};
    end
end


boots=2000;


utils.overridedefaults(who,varargin);

if one_dist
    % assume test whether mean of the population is differenct from zero.
    dist=A;
    if isvector(dist)
        dist=dist(:);
    end
    n=size(dist,1);
    
    
    [B]=bootstrp(boots, @nanmedian, dist);
    
    ps=[0:0.01:100];
    sd_ps=prctile(B,ps);
    
    sd_p=stats.get_p(0,B);
    
    varargout{1}=sd_p;
    varargout{2}=prctile(B,50);
    varargout{3}=B;
    
elseif nargin==2
    sA=size(A,1);
    sB=size(B,1);
    sd=nanmedian(A)-nanmedian(B);
    
    if min(sA,sB)<=6
        warning('Not meaningful to compute 2000 bootstraps when n<7');
        boots=factorial(min(sA,sB));
    end
    
    
    ALL_DATA=[A;B];
    boot_score=zeros(boots,size(ALL_DATA,2));
    for bx=1:boots
        
        shuff_d=ALL_DATA(randperm(sA+sB),:);
        A=shuff_d(1:sA,:);
        B=shuff_d(sA+1:end,:);
        
        boot_score(bx,:)=nanmedian(A)-nanmedian(B);
    end
    
    sd_p=stats.get_p(sd, boot_score);
end
varargout{1}=sd_p;
varargout{2}=prctile(B,[2.5 97.5]);
varargout{3}=prctile(B,[50]);