What he said. Linear least squares (y = at + b where t is time) tends to 
find trends where none exist, when you're fitting it to nonstationary 
time series data. That is, standard tests of significance can produce 
results that have been shown to be spurious, for this model.

Beyond that, the model y=at + b is pretty much useless for time series 
data that has been stationarized (which often means converted into a 
series of differences from previous sample)

On 7/8/2010 6:15 PM, Patrick Haskell wrote:
> Mann-Kendall is a non-parametric statistical test to determine the 
> existence or absence of a trend at a given statistical significance.  
> The slope of the line describing the trend could still be determined 
> using a least-squares fit (or other suitable method) after the 
> determination of whether a trend exists.
>
>
> ------------------------------------------------------------------------
> *From:* Denis Bogan <[log in to unmask]>
> *To:* [log in to unmask]
> *Sent:* Thu, July 8, 2010 5:09:37 PM
> *Subject:* Re: [SKIVT-L] Nice Work Wes
>
> Yes.  Nice work.  I do have a curiosity question.  I am unfamiliar 
> with the Mann-Kendall test but since it yields a straight line fit I 
> am wondering what is the advantage over a straight line lease squares 
> fit.  This would also give correlation coefficients.  You must have 
> tried it.
> - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 
> SkiVt-L is brought to you by the University of Vermont.
>
> To unsubscribe, visit http://list.uvm.edu/archives/skivt-l.html
>
>
> - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 
> SkiVt-L is brought to you by the University of Vermont.
>
> To unsubscribe, visit http://list.uvm.edu/archives/skivt-l.html
>


- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
SkiVt-L is brought to you by the University of Vermont.

To unsubscribe, visit http://list.uvm.edu/archives/skivt-l.html