Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

A bit of confusion, xcorr output vs. corr output

35 views
Skip to first unread message

JS

unread,
Oct 15, 2009, 5:03:07 AM10/15/09
to
Hi group,

It seemed my post from yesterday fell into a black hole.. Couldn't
find it when I wanted to see if anybody replied, anyway. So here goes
again. Should be simple for someone with a bit of knowledge in
statistics and MatLab (not me for the former..):

I have two time-series that I cross correlate to find the optimal
shift, i.e. the delay of one relative to the other where the signals
are the most similar. I do that by doing this:

xcorr(signal1,signal2,5,'coeff')

I get a number of correlation-coefficients out between -1 and 1. My
question is, can I use the maximum correlation coefficient from this
vector to say anything about significance? That is, are they Pearson
r
values that I can look up in a table, e.g. 20 degrees of freedom,
p=0.05 --> r large enough to be significant or not?

I know that the number that comes out of the "corr" function in the
statistics toolbox is this type of
value, but am I then supposed to write my own routine for shifting
the
signals, padding with zeros, etc.?

Any help is greatly appreciated. Thanks!

/J

JS

unread,
Oct 15, 2009, 9:04:59 AM10/15/09
to

Ok - I think I've gotten a little wiser.. I think that I can get
"correct" correlation coefficients out of the cross-correlation (i.e.
Pearson r-values that can be tested for significance), if I do this:

c = xcov(signal1,signal2,5,'coef')

Anybody care to confirm?

Thanks!

/JS

JS

unread,
Oct 15, 2009, 9:12:22 AM10/15/09
to
On Oct 15, 11:03 am, JS <joe778...@gmail.com> wrote:

Ok, I think I've gotten a little wiser. I think I can get "correct"
correlation coefficients (i.e. Pearson r-values that can be tested for
significance) in the cross correlation vector if I do this:

c = xcov(signal1,signal2,5,'coeff')

0 new messages