First off this isn't for school. I have a few questions relating to cross-correlations, normal distributions, and confidence intervals and would appreciate any guidance.

In no particular order:

1) What is the best way to combine cross-correlations?
I've noticed that where h = g+f, xcorr(h) != xcorr(g) + xcorr(f)
also, xcorrSquared(h) != xcorrSquared(g) + xcorrSquared(f)
Should I used the above line as an aprroximation, or is there a better algorithim for getting the exact cross correlation between two functions made up of several different function pairs with known correlations

2) once the above correlations are known for the two "super" functions, would it be legitimate to treat the observed correlation one time unit in the future as a stochastic process (uniformly distributed), and would the population size be the number of function pairs that went into making the two major functions? something about this doesn't seem right.

I'm trying to find a confidence interval for the correlation between the two major functions, given historical correlation of several minor function pairs. What would be the best way to go about this?

10/22/2006 8:41:42 PM

i think the lack of replies means that you already know 100x more than we do, im even a stat. major, but only a sophomore, cant help ya out. sorry

10/23/2006 3:44:54 AM

ok well let me say this then:


suppose: f(x) = g(x) + h(x)

does xcorr(f(x)) = xcorr(g(x)) + xcorr(h(x))?

where xcorr is cross correlation, because it seems that my numbers are implying that this equality doesn't exactly hold, but may be an unbiased estimator, anyone?

10/23/2006 1:25:56 PM

http://mathworld.wolfram.com/Cross-CorrelationTheorem.html
http://en.wikipedia.org/wiki/Cross-correlation

Glancing at those two pages tells me the problem you are trying to solve is probably over most of our heads. Would recommend picking up some older textbooks from the library for this one.

10/23/2006 1:36:17 PM

thanks for the links

10/24/2006 12:58:48 PM