nanstd2

 FORMAT: Y = NANSTD(X,DIM,FLAG)
 
    Standard deviation ignoring NaNs

    This function enhances the functionality of NANSTD as distributed in
    the MATLAB Statistics Toolbox and is meant as a replacement (hence the
    identical name).  

    NANSTD(X,DIM) calculates the standard deviation along any dimension of
    the N-D array X ignoring NaNs.  

    NANSTD(X,DIM,0) normalizes by (N-1) where N is SIZE(X,DIM).  This make
    NANSTD(X,DIM).^2 the best unbiased estimate of the variance if X is
    a sample of a normal distribution. If omitted FLAG is set to zero.
    
    NANSTD(X,DIM,1) normalizes by N and produces the square root of the
    second moment of the sample about the mean.

    If DIM is omitted NANSTD calculates the standard deviation along first
    non-singleton dimension of X.

    Similar replacements exist for NANMEAN, NANMEDIAN, NANMIN, NANMAX, and
    NANSUM which are all part of the NaN-suite.

    See also STD

0001 function y = nanstd2(x,dim,flag)
0002 % FORMAT: Y = NANSTD(X,DIM,FLAG)
0003 %
0004 %    Standard deviation ignoring NaNs
0005 %
0006 %    This function enhances the functionality of NANSTD as distributed in
0007 %    the MATLAB Statistics Toolbox and is meant as a replacement (hence the
0008 %    identical name).
0009 %
0010 %    NANSTD(X,DIM) calculates the standard deviation along any dimension of
0011 %    the N-D array X ignoring NaNs.
0012 %
0013 %    NANSTD(X,DIM,0) normalizes by (N-1) where N is SIZE(X,DIM).  This make
0014 %    NANSTD(X,DIM).^2 the best unbiased estimate of the variance if X is
0015 %    a sample of a normal distribution. If omitted FLAG is set to zero.
0016 %
0017 %    NANSTD(X,DIM,1) normalizes by N and produces the square root of the
0018 %    second moment of the sample about the mean.
0019 %
0020 %    If DIM is omitted NANSTD calculates the standard deviation along first
0021 %    non-singleton dimension of X.
0022 %
0023 %    Similar replacements exist for NANMEAN, NANMEDIAN, NANMIN, NANMAX, and
0024 %    NANSUM which are all part of the NaN-suite.
0025 %
0026 %    See also STD
0027 
0028 % -------------------------------------------------------------------------
0029 %    author:      Jan Gl�scher
0030 %    affiliation: Neuroimage Nord, University of Hamburg, Germany
0031 %    email:       glaescher@uke.uni-hamburg.de
0032 %
0033 %    $Revision: 1.1 $ $Date: 2014/06/10 20:18:11 $
0034 
0035 if isempty(x)
0036     y = NaN;
0037     return
0038 end
0039 
0040 if nargin < 3
0041     flag = 0;
0042 end
0043 
0044 if nargin < 2
0045     dim = min(find(size(x)~=1));
0046     if isempty(dim)
0047         dim = 1; 
0048     end      
0049 end
0050 
0051 
0052 % Find NaNs in x and nanmean(x)
0053 nans = isnan(x);
0054 avg = nanmean(x,dim);
0055 
0056 % create array indicating number of element
0057 % of x in dimension DIM (needed for subtraction of mean)
0058 tile = ones(1,max(ndims(x),dim));
0059 tile(dim) = size(x,dim);
0060 
0061 % remove mean
0062 x = x - repmat(avg,tile);
0063 
0064 count = size(x,dim) - sum(nans,dim);
0065 
0066 % Replace NaNs with zeros.
0067 x(isnan(x)) = 0; 
0068 
0069 
0070 % Protect against a  all NaNs in one dimension
0071 i = find(count==0);
0072 
0073 if flag == 0
0074     y = sqrt(sum(x.*x,dim)./max(count-1,1));
0075 else
0076     y = sqrt(sum(x.*x,dim)./max(count,1));
0077 end
0078 y(i) = i + NaN;
0079 
0080 % $Id: nanstd2.m,v 1.1 2014/06/10 20:18:11 cbassuser Exp $