How can I center a dataset with NaN values in it?

I know this sounds like an easy question. Really, it is. I've even asked some form of this question before. But, I have run into a new problem today. I have a 736x1 vector, called 'ARRAY4' which I am trying to center. The vector contains NaN values. The code I am using is:

ARRAY4c=bsxfun(@minus,ARRAY4,nansum(ARRAY4)/size(ARRAY4,1));

This code works fine with many other datasets, but for some reason it does not with this one. The output, rather than being centered around the average, is all positive values and NaNs. I have checked the original data several times manually to look for any anomalous values, but to no avail. What could possibly be happening here?

[Azzi Abdelmalek]

You said: but for some reason it does not with this one.

Which one?

[Patrick]

This one, as in this dataset, "ARRAY4". It has worked with my other datasets, "ARRAY1", "ARRAY2", "ARRAY3", which also contain NaN values.

Answer 1

[Patrick]

As a different approach, I used:

nanmean(ARRAY4c)

to find the mean of the positive data, and centered it around that.