But order of observations has everything to do with a time series, doesn't it? And autoregression models use prior values of the time series to generate current estimates, so randomly sorting the data would invalidate everything for a real time series.. By these standards, I'm not dealing with a time series and any correlation among dependent variables could just be an artifact due to lack of random sampling -- I'm sampling at regular time intervals. Of course, lack of random sampling has its own issues, but my question is related to whether I should treat my data as a time series.
I assume your "Nope!" refers to my last question, which was : > >Knowing that I _can_ remove the autocorrelation, can I proceed to perform > >parametric regression analysis without actually randomly sorting the data > >and treat this as a non-time-series analysis? Regards, Steve "Dick Startz" <[EMAIL PROTECTED]> wrote in message news:[EMAIL PROTECTED]... > Nope! > > The order of observations has no effect on coefficient estimates or > prediction from a regression. But if the errors in the regression (not > just the values of the dependent variable) are correlated with one > another, then a regression may not be the best thing to do for > variety of reasons. > > Sorting the data doesn't make the correlation go away, it just hides > it. It used to be that observation 1 was correlated with observation 2 > and observation 2 with observation 3, etc. Now 1 is correlated with > 393 (or wherever 2 got randomly sorted to), etc. > > -Dick Startz > > On Mon, 27 Jan 2003 23:35:17 GMT, "Mountain Bikn' Guy" <[EMAIL PROTECTED]> > wrote: > > >My dependent variable fits at least one definition of a time series: "If you > >take a sequence of equally spaced readings, this is called a time series." > >Furthermore, there is very strong autocorrelation (near 1) in the dependent > >variable -- when tested in the order the data is collected. However, I can > >randomly resort all the data (dependent plus independent variables) so that > >there is no longer any autocorrelation and this does not affect the > >predictive ability of the independent variables. So I'm thinking that I am > >not dealing with a time series. Any thoughts? > > > >Any arguments in favor of using time series analyses? > > > >Knowing that I _can_ remove the autocorrelation, can I proceed to perform > >parametric regression analysis without actually randomly sorting the data > >and treat this as a non-time-series analysis? > > > >TIA > > > >Steve > > > > > > > > > > ---------------------- > Richard Startz [EMAIL PROTECTED] > Lundberg Startz Associates . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
