On Sun, 24 Aug 2003 23:35:13 GMT, [EMAIL PROTECTED] wrote:

> Hello
> 
> For my simulations, I have to simulate paths of 
> Normal(mu*, sigma*).random variables.  It so happens, that each path
> has few [500] observations, AND sigma is large [compared to mu]. Thus
> the realized values of mu are often very different from mu*.   I
> discovered that in my particular application, it makes sense to take
> each path, 
> 1) estimate mu and sigma 
> 2) if mu is not equal to m*, then "massage the data" - for each path,
> convert the data into N(0,1) [by subtracting the estimated mu [for
> this path] from each observation, etc], and then convert the data into
> N(M*, Sigma*)I [by adding m* to each value, etc].
> 
> I was wondering whether there is a formal Statistics term describing
> this "massaging the data" procedure.  I really need to know this, to
> be able to put my research into context of existing work.


I think that for this one, you would just give the short conclusion -
you performed simulations  with random normal variables,
where the mu and sigma for every 500 trials were constrained
to the exact values ....

I wonder if you are doing it backwards, though.  One of the
important things about *randomized*  variables is that they 
are random.  It is not very 'random'  in description, if every
mean = 0, SD=1.  I'm asking this because I made that mistake
when I first set out to do Monte Carlo, 25 or so years ago --
I thought that the means and SDs should be fixed, as a 
general result.  Now I realize that there can be special reasons
to require fixed-means/ SDs, but the general case doesn't.


I have read that the LACK  of randomness of computer's
random-number-generators is a potential problem, for 
some very-large-scale  simulations.   In particular, I read about
simulations of breeding populations.  It should be important, 
for interactions over hundreds or thousands of generations, that
the external variations are modeled right.  
 - First, it might be too conservative to use Normal.
 - Second, if Normal is the proper, real-world assumption, THEN
the algorithms need to make variation in the means and variances.

I am willing to read more about the randomness of algorithms, 
too, but I don't have any references right now.

-- 
Rich Ulrich, [EMAIL PROTECTED]

http://www.pitt.edu/~wpilib/index.html
"Taxes are the price we pay for civilization."  Justice Holmes.
.
.
=================================================================
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