> -----Original Message-----
> From: Olsen Chris [mailto:[EMAIL PROTECTED]]
> Sent: Sunday, September 29, 2002 6:23 AM
> To: David Heiser
> Subject: RE: Regression CIs (was: Normally Distributed ANOVA FACTORS?)
>
>
> David --
>
> > At least right now, Marsaglia's KISS RNG is a pretty good
> > RNG.
>
> I am not familiar with this -- do you have a reference?
>
> Thanks in advance!
>
> -- Chris
>
> Chris Olsen
> George Washington High School
> 2205 Forest Drive SE
> Cedar Rapids, IA
>
> (319)-398-2161
-------------------------------------------------------------------------
Information on it is small.
Marsaglia talked about it in STAT-L some years ago. He calls it "KISS" (Keep
It Simple Stupid). He suggested it as a reference source in his Diehard Test
Suite (CDROM or Internet download). The actual program in Fortran is at
http://www.fortran.com/fortran/kiss.f90. You can adapt it to C++ or VBA. It
is best done as an assembly level subroutine.
It combines the advantages of a linear congruential generator (See Knuth,
"The Art of Computer Programming", vol 2, for a very detailed mathematical
and algorithmic discourse on LCGs) with shift register generators and a
multiply-with-carry generator. In short he combines all the good RNG methods
into a neat single routine.
Knuth says that a RNG that passes all tests is likely to be a random
sequence (Knuth, page 149).
However there appears to be no references that prove that KISS generates an
infinity-distributed sequence. However even this approach leads one to
conclude that the binary sequence 01010101010101010101010101010101 is not a
random sequence. The probability that KISS could generate it is small, but
under the equidistributed sequence criteria, p is not zero (See Knuth, page
177).
David Heiser
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
