Rich Ulrich <[EMAIL PROTECTED]> writes:
> Typo?  But "included the /N in the square root"  is exactly what
> Don expects.  What you say is written, and what Don says, is 
>     SD= sqrt(SS/N)
> not, 
>     SD= sqrt(SS)/N
> which would produce the statement in (e).
> 
> The paper had no apparent typo.  Am I missing something?

Just that I shouldn't try to do maths at 5am when in a state of panic.

Please ignore everything I said about problems with the formula. It 
transpires that if you calculate the standard deviation and plot log 
s.d. vs log bin size you get a gradient which is H-1, and if you 
calculate the variance and plot log var vs log bin size, you get a
gradient which is 2H-2. (As expected, as log x^2 = 2 log x). Likewise, 
if you use the standard deviation, the "cutoff" gradients for 
different conclusions are -0.33 and -0.5, and if you use the variance 
they are -0.66 and -1.0. [1]

So I'm pretty certain that my formula is correct (it was my text 
description that was wrong). My question about dealing with holes in
time series remains. I've looked at about 10 different textbooks on
statistics as applied to the space and atmospheric sciences, and
I think only one of them had anything to say about data holes. 
Unfortunately the data analysis method it was describing is very 
unlike mine and so there was no way for me to apply their technique. 
Applying a regression analysis to fill in gaps just isn't practical
with 80000 points x 7 sites x 2 pollutants and a high level of
intermittency in the data. 

helen-louise
[EMAIL PROTECTED]


[1] The reason for my extreme confusion was there is another method 
very similar to the aggregated variance which gives a result of 2H-1,
and I therefore ended up thinking that H-1, 2H-2 and 2H-1 were all 
results of _the same method_!

.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to