On Sat, 26 Apr 2003, Stanislav Markus wrote:
Dear all,
I want to use Amelia on a time-series cross-sectional dataset, and -
having successfully loaded the input file, and started the iterations
(Windows version) - have some questions... Any help would be greatly
appreciated!
1) Overall, is it ok to try Amelia on a large dataset consisting of
4000+ observations and 50 variables, for 150 states over time - should I
keep any caveats in mind, apart from those mentioned in the manual?
(Currently, Amelia is taking a few minutes for each iteration; I guess
I'll leave it running for the night...)
50 is a lot. more than 40 is unplesant, altho possible. count the number
of params, p*(p+3)/2. its a lot. if you have trouble, cut it down to 10
and then gradually increase.
2) For global_AMlagvs, what's the rule of thumb concerning the variables
that should be lagged? If, for example, one model uses 5 independent
variables, all of which are lagged by 5 years (with respect to the
dependent variable) - should I lag just those in Amelia, if I want to
reestimate the model with imputed data?
start with the variables you'll want in the analysis model. then
gradually add variables that you think will help predict.
3) For global _AMusets (time trend option), should I choose to include
it if the original model had dummies for all years?
dummies for all years makes it the functional form completely flexible.
so ideally you'd include the dummies. tho that may be asking a lot.
4) I wonder whether I should change prior from its
default setting..
only if you need to.
Gary
Thanks!
Stan
****************************
Stanislav Markus
Ph.D. Candidate
Harvard University
Department of Government
e: smarkus(a)fas.harvard.edu
t: 617.513.5407
-----Original Message-----
From: gov2001-l-admin(a)fas.harvard.edu
[mailto:gov2001-l-admin@fas.harvard.edu] On Behalf Of
dhopkins(a)fas.harvard.edu
Sent: Thursday, April 24, 2003 7:59 PM
To: David Konisky
Cc: gov2001-l(a)fas.harvard.edu
Subject: Re: [gov2001-l] amelia
Dear David,
I'd recommend the Gauss version. Here's how to use it:
1. use DBMS copy to transfer the data into Gauss format (.dat)
2. put the data in the GAUSS36 directory
3. create a GAUSS program; this is more simple than it sounds. It
needs to
look like the below, where the ordinal, nominal, and fully specified
variables
are listed (see the Amelia instructions for more). The "tab10" is your
dataset.
4. Name this program, and place it, too, in the GAUSS36 folder.
5. Open GAUSS36 and type "run programname"--the rest is automatic.
new;
library amelia;
_AMords= {21};
_AMnoms= {1,2,3,10,15,16,17,18,19};
_AMfully = {7,8,11,22};
dbuf = amelia("tab10");
Best,
Dan
Quoting David Konisky <konisky(a)mit.edu>du>:
Has anyone had success using the Windows version
of amelia? I have
been
unable to input my dataset; the program either
fails to recognize the
file
or crashes altogether. I have tried numerous
formats, including
Ascii,
txt, excel, comma delimited, and Stata without
any luck.
Any suggestions would be greatly appreciated.
David
_______________________________________________
gov2001-l mailing list
gov2001-l(a)fas.harvard.edu
http://www.fas.harvard.edu/mailman/listinfo/gov2001-l