Skip to content

squared coherency and cross-spectrum

4 messages · Ling Jin, Spencer Graves, Kjetil Halvorsen

#
Hi All,

I have two time series, each has length 354. I tried to calculate the 
coherency^2 between them, but the value I got is always 1. On a website, 
it says: " Note that if the ensemble averaging were to be omitted, the 
coherency (squared) would be 1, independent of the data". Does any of 
you know how to specify properly in R in order to get more useful 
coherency? The examples in the help do give coherencies that are not 1s, 
but I did not notice any special specification.

Next question is on co-spectrum. When I supply "spectrum" function with 
multiple time series, it only gives me spectrum (smoothed periodogram) 
of individual time series. Is there any way I can get the 
cross-spectrum? I believe R has calculated it, but I could not find in 
the returned values.

Attached is the smoothed periodogram of the two time series.

Thanks a lot!

Ling

-------------- next part --------------
A non-text attachment was scrubbed...
Name: test.spectrum.png
Type: image/png
Size: 4596 bytes
Desc: not available
Url : https://stat.ethz.ch/pipermail/r-help/attachments/20051201/1beca4ef/test.spectrum.png
4 days later
#
I haven't seen a reply, so I will comment even though I've never used 
"coherency" / "coherence" nor "spectrum".  RSiteSearch("coherence") 
produced 13 hits, the third of which looked like it might be relevant to 
your question 
(http://finzi.psych.upenn.edu/R/Rhelp02a/archive/37640.html). 
RSiteSearch("coherency") produced 12 hits, at least some of which look 
like they might help you.  In my cursory review, it looked like at least 
one of the "coherence" / "coherency" hits also mentioned the 
co-spectrum.  Whether that's true or not, the examples with "?spectrum" 
includes the statement, "for multivariate examples see the help for 
spec.pgram".  If you still have a question for this listserve after 
reviewing these references, PLEASE do read the posting guide! 
'www.R-project.org/posting-guide.html'.  I believe that people who 
follow more closely that posting guide tend to receive quicker, more 
useful answers than those who don't.

	  I hope you won't mind if I now ask you a question:  What can you get 
from "coherency" and "co-spectrum" that you can't get as easily from 
autocorrelation and partial autocorrelation functions, including the 
cross-correlations?

	  hope this helps.
	  spencer graves
Ling Jin wrote:

            

  
    
#
Spencer Graves wrote:
Although the question is not to me,I try to answer,as I am planning to 
try to use this techniques!  What I hope to get from them is on what 
time scale to time series is correlated, or more succinctly, the high 
frequency variation we can se in both series (ground and sattelite 
measurement of same phenomenon), are they correlated?

Kjetil
#
Thanks for the information. I tried to use coherency to look at 
correlations of two time series at different time scales. More 
specifically, on what time scale, my model predictions are more coherent 
with the observed values.

Ling
Spencer Graves wrote: