Running anovan on a Large (21504Rx90C) matrix

I am currently attempting to run a full anovan model on a 21504Rx90C matrix, which results in an out-of-memory error. I am doing this in order to extract the sum of squares contribution of each factor used in the anova model.

The anova model itself has 7 factors with 3,4,4,4,7,4,4 levels respectively. I was wondering if there was perhaps a way of breaking down the code to avoid running into this error and staying within the memory constraints?

Dummy code below:

Data = randn(21504,90)
SumSq = zeros(190,size(Data,2));

for iii = 1:size(Data,2)
[~,tbl,~,~] = anovan(Data(:,iii),{var1 var2 var3 var4 var5 var6 var7},...
    'model','full',...
    'varnames',{'var1','var2','var3','var4','var5','var6','var7'});
SumSq(1:end,iii) = cell2mat(tbl(2:end,2));
iii
end

[Jeff Miller]

If you can't get anovan to give you the SS's, you could get them for the dataset you describe with one of the short-cut SS methods that only work for balanced designs. I don't know if there are canned routines for any of those, though, so you might have to do the programming yourself.

[John Smith]

I am not aware of any such short cut methods. Where could I find information on these?

[Jeff Miller]

See chapter 9 of this book

Watch out for numerical problems, though.

[John Smith]

Thank you. I will try and implement this method and see if I have any luck.