On Tue, 28 Jan 2003 19:14:05 GMT, "Mountain Bikn' Guy" <[EMAIL PROTECTED]> wrote:
[ snip, much... ] > I now feel comfortable that I do not need to use autoregression techniques > even though I have an autocorrelated time series. > [ ... ] X is autocorrelated, and Y is autocorrelated? One characterization that I like is to say, this will act like a test-situation for t-test or correlation that has a much tinier number of degrees of freedom. Consider: You do not really have more *information* if you measure air temperature (say) every minute.... Is somebody's climate question (if that were the topic) really worth 24 time points per day? 24x60? 24x60x60 ? In the important sense of 'information', you can't manufacture 'information' by measuring the same thing over and over again, when there is nothing independent. One way to 'fix' some autocorrelation is to skip all the points that are close enough to be correlated. That's true even if you have to drop 9 out of 10 points, or 999 out of 1000. Check for cross-correlation, and if there is not any, then that changes the next questions. I would not worry about applying diagnostics until I spent some time in figuring whether the series *ought* to be correlated, for some logical reason. That should give me a hint of which critical components need to be disentangled, which *changes* ought to provide a sensitive test of co-variation. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
