anova2w.m

偏最小二乘算法在MATLAB中的实现

function anova2w(dat,alpha)
%ANOVA2W Two way analysis of variance
% Calculates two way ANOVA table and tests significance of
% between factors variation (it is assumed that each column
% of the data represents a different factor) and between
% blocks variation (it is assumed that each row represents
% a block). The inputs are the data table dat and the 
% desired confidence level alpha, expressed as a fraction 
% (e.g. .95, .99, etc.). The output is an ANOVA table.
% The I/O format is: anova2w(dat,alpha)

%  Copyright
%  Barry M. Wise
%  1994
[n,k] = size(dat);
xbar = mean(mean(dat));
xbari = mean(dat);
xbarj = mean(dat');
sst = sum(sum(dat.^2)) - (1/(n*k))*sum(sum(dat))^2;
sstr = n*sum((xbari - xbar*ones(size(xbari))).^2);
ssb = k*sum((xbarj - xbar*ones(size(xbarj))).^2);
sse = sst - sstr - ssb;
mssb = ssb/(n-1);
msstr = sstr/(k-1);
msse = sse/((n-1)*(k-1));
ff = msstr/msse;
fb = mssb/msse;
disp('  ')
disp('___________________________________________________________')
disp('  Source of           Sum  of        Degrees          Mean')
disp('  Variation           Squares      of  Freedom       Square')
disp('___________________________________________________________')
s = sprintf('Between factors    %11.4f       %4.0f        %10.4f',...
sstr,k-1,msstr);
disp(s)
disp('(columns)')
s = sprintf('Between blocks     %11.4f       %4.0f        %10.4f',...
ssb,n-1,mssb);
disp(s)
disp('(rows)')
s = sprintf('Residual           %11.4f       %4.0f        %10.4f',...
sse,(n-1)*(k-1),msse);
disp(s), disp('  ')
s = sprintf('Total              %11.4f       %4.0f',...
sst,(k-1)+(n-1)+((n-1)*(k-1)));
disp(s), disp('  ')
s = sprintf('Effect of factor = %g/%g = %g',msstr,msse,ff);
disp(s)
fstatf = ftest(1-alpha,k-1,(n-1)*(k-1));
pc = alpha*100;
s = sprintf('F at %g percent confidence = %g',pc,fstatf); 
disp(s)
if fstatf < ff
  disp('Effect of factors IS significant')
else
  disp('Effect of factors IS NOT significant')
end
disp('  ')
s = sprintf('Effect of block  = %g/%g = %g',mssb,msse,fb);
disp(s)
fstatb = ftest(1-alpha,n-1,(n-1)*(k-1));
s = sprintf('F at %g percent confidence = %g',pc,fstatb); 
disp(s)
if fstatb < fb
  disp('Effect of blocks IS significant')
else
  disp('Effect of blocks IS NOT significant')
end
disp('  ')