significance of spectal peak with spectrum()
Sebastian Leuzinger wrote:
thanks a lot. I am interested in the more complex case where the interest is about a specific frequency being significant, not "at least one frequency being significantly different from the backgrond white noise". I have discussed this issue with very knowledgable people in the field who could not help me either. I would be interested in any references as well.
http://bayes.wustl.edu/ where you can download Bretthorst's book "Bayesian Spectrum Analysis and Parameter Estimation" Kjetil
On Friday 16 September 2005 10:36, you wrote:
Sebastian Leuzinger wrote:
the null hypothesis would be: one particular frequency peak is not
significantly different from the background noise.
So you want to know, e.g., whether there is something going on at 1000 Hz? This is difficult: If you are considering the periodogram to be a density, then you do not know the distribution of the value of a single frequency, because it depends on the stuff going on at other frequencies. Second point is (and already asked): "Kind of [background] noise"? The only really easy test is for the Null "signal is white noise", hence H1 is "at least one non-white-noisy frequency". [If somebody knows a really good book or papers that cover other cases than the trivial one mentioned above, I am very interested to hear about them, BTW.] If you have another kind of noise (such as blue or pink noise), things become even worse. Uwe Ligges
On Friday 16 September 2005 09:28, you wrote:
Sebastian Leuzinger wrote:
Hello, has anybody got a simple recepie to test the significance level
of the peaks after using spectrum() ?
What is you null hypothesis?
- Kind of noise?
- One particular frequency is noisy or all noisy?
- ...
Uwe Ligges
(R-version 2.0.1, linux SuSE9.3)
--
Kjetil Halvorsen.
Peace is the most effective weapon of mass construction.
-- Mahdi Elmandjra
No virus found in this outgoing message. Checked by AVG Anti-Virus.