I can't say if anything has changed since last summer as my executable is still on the 23 Jul 2004 version (trying to find a convenient time for everyone on the network license to update...), but I can tell how Stata (in this executable at least) deals with "illegally high" seeds...
The first argument of the KISS generator is updated mod 2^32 - so if you give -set seed- an argument greater than or equal to 2^32, this is just treated mod 2^32. In your case:
. set seed 4753533809
. di c(seed)
Xe0dcdff5c43f462544a474abacbdd93d062c
. set seed `=4753533809-2^32'
. di c(seed)
Xe0dcdff5c43f462544a474abacbdd93d062c
This still leaves a bit of a hole from 2^31 to 2^32-1. Any argument # in this range is treated as 2^32-#, for example:
. set seed 3000000000
. di c(seed)
Xfb7d4374c43f462544a474abacbdd93d0a9a
. set seed `=2^32-3000000000'
. di c(seed)
Xfb7d4374c43f462544a474abacbdd93d0a9a
None of this explains why the output of your do-file has changed. Have you tried rerunning the do-file twice? Does it give the same results both times? Have you tried running it twice with a lower ("legal") seed? Some procedures just do not seem to be reproducible...
Hope this helps
David
[email protected]
-----Original Message-----
From: Roger Newson [mailto:[email protected]]
Sent: 07 February 2005 21:49
To: [email protected]
Subject: st: Query on -set seed-
Fellow Statalisters (especially StataCorp):
According to -[R] functions-, the -set seed- command can take any seed from
0 to 2^31-1 inclusive, and, if I specify a seed with one of these values in
a do-file under -version- control and then use the -uniform()- function,
then the same result will be produced every time I execute that do-file
with the same input data. My query is, will this still be the case if I set
the seed to an even higher integer? I ask because, last July, I
inadvertently wrote a do-file containing the statements
set seed 4753533809;
gene double ranord2=uniform();
and executed it on 15 July 2004 under Stata 8.2 under Windows 2000,
producing some random-looking results and no complaints about the illegally
high seed. However, when I re-executed the same do-file today (7 September
2005) with the same input dataset under the Stata 8.2 update dated 10
January 2005, it produced a completely different set of random-looking
results, and still no complaints about the same illegally high seed. Have
there been any unpublicised revisions to -set seed- or to -uniform()-
affecting the response to an illegally high seed?
Best wishes (and thanks in advance)
Roger
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
