MATLAB's crosscorr function and R Studio ccf show different results

Hi, I'm using MATLAB's crosscorr function and R Studio's ccf. For the same data, the results differ. It appears that the lag axis is flipped in one of them.

I've reproduced the crosscorr documentation example in both and this is what I see. The data is in the attached .txt file.

vs R Studio

Why is this happening? Thanks.

[Bjorn Gustavsson]

They "simply" seems to be flipped. That most likely is because they define the lags in opposite direction. If that's the case you should find the information somewhere in the documentation.

[Lizzy Cedeno]

Hi, I've checked the documentation for both but couldn't find any information there that implies the lag is treated differently. Thanks for your reply!

Answer 1

[Bjorn Gustavsson]

If you cannot find the information in the documentation, you'll have to go to the source. This is what I find in the xcorr function:

    X = fft(x,m2,1);
    Y = fft(y,m2,1);
    if isreal(x) && isreal(y)
        c1 = ifft(X.*conj(Y),[],1,'symmetric');
    else
        c1 = ifft(X.*conj(Y),[],1);
    end
    % Keep only the lags we want and move negative lags before positive
    % lags.
    c = [c1(m2 - mxl + (1:mxl)); c1(1:mxl+1)];

How that differs from the R-implementation I cannot tell. (But without any thinking) This looks like an fft-based convolution between x and y. Exactly which way around to go here requires attention to detail, but you can compare this with the R-version, and check which is what with the wikipedia-description: Cross-correlation.

HTH

[Lizzy Cedeno]

Thank you! Yes, from what I've gathered in the past couple of days is that both implementations differ in the order of the arguments: R Studio's ccf(t+k, t) vs MATLAB crosscorr(t, t+k).