 |
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
RE: st: ice and random-number seed
Removing the weights does not seem to fix the problem.
However, I am able to work around the problem and obtain reproducible results by manually setting the seed both before -ice- and before -mim-. The -seed- option in -ice- does not need to be used. For example:
set seed 1234
ice var1 var2 var3 , m(2)
set seed 1234
mim: reg var1 var2 var3
Thank you for all of your suggestion.
Tait
-----Original Message-----
From: Rodrigo Alfaro A. [mailto:[email protected]]
Sent: Friday, August 22, 2008 10:16 AM
To: [email protected]
Cc: Medina, Tait Runnfeldt
Subject: RE: st: ice and random-number seed
///
Tait,
I think that your case is more complicated than Maarten's example.
(1) You are using weights. If I recall correctly ICE produces results
using iterations => order matters. My suggestion is to try your problem
without weights.
(2) Sorting could make some differences here. You could try imputing
one dataset each time using a specific seed. It is expected the
correlations between imputed variables (original y replication) should
be one.
Under (1) you have no solution. Under (2) you could control the problem
imputing 1 dataset at the time with specific seeds. We had problem (2)
with user-written routine hotdeck.
Rodrigo.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/