Pearsons correlation using corrcoef not working

I am trying to calculate the pearsons correlation between two variables in a timetable. I am utalising the corrcoef() function. However the output it gives me is always NaN NaN NaN NaN. Why is this? How can I fix this?

[Daniel Pollard]

Have you looked at the documentation? There are many reasons it could output NaN.

[Elizabeth Lees]

Hi thank you for the link, just looking over the document now I think it could be an issues of some of the columns having NaN values within them I have now used the corrcoef( data.datacolumn1, data.datacolumn2, 'Rows', 'Pairwise') and it has worked. Thanks for the help!

[William Rose]

@Elizabeth Lees

Are the variables (vectors) which you are trying to correlate both numbers (not strings or times or dates), with the same length?

If you provide the timetable, and the code that does not work, as attached files, it will help others help you. An example of corrcoeff() operating on two columns from a table is below.

>> height=1.7+0.2*randn(10,1);
>> weight=70+10*randn(10,1);
>> T=table(height,weight);
>> corrcoef(T.height,T.weight)
ans =
    1.0000    0.2262
    0.2262    1.0000

[Elizabeth Lees]

@William Rose Thank you for your answer. I have discovered it was because many rows of the numerical data contained NaN values, I have now used 'Rows', 'Pairwise' to discard these in the statistics and am now recieving results. However when just the NSE() function to calculate the Nash-Sutcliffe value between the variables I am still getting an NaN result I believe for the same problem however 'Rows', 'Pairwise' doesn't seem to work within the NSE() function, do you know of another way to discard these empty rows to calculate the Nash-Sutcliffe?

Example code:

NSE(combined_data.Observed_Rio_Branco,combined_data.Simulated_Rio_Branco, 'Rows', 'pairwise')

Error using NSE

Too many input arguments.

Answer 1

[William Rose]

@Elizabeth Lees

This code eliminates each row with one or more NaNs. Thank you to @Jan for this.

X = rand(10, 4);
X(X < 0.1) = NaN;
disp(X);                        %array containing NaNs
X(any(isnan(X), 2), :) = [];    %delete rows with NaN
disp(X)                         %array with NaN rows removed

[William Rose]

Assumng your input data are column vectors, you would do

%make a Nx2 array 
A=[combined_data.Observed_Rio_Branco,combined_data.Simulated_Rio_Branco];
A(any(isnan(A), 2), :) = [];    %delete rows with NaN
NSE(A(:,1),A(:,2));

I do not have the NSE() function so I cannot test the code above.