10 May
2004
10 May
'04
5:17 p.m.
the theorem only says that if you let it run it will eventually finish,
but there are problems sometimes. almost all the time the problem is due
to difficult or impossible datasets, such as if your panel design is
entirely missing for all the variables for respondents who didn't answer
one of the waves of the survey. I suggest you start with just a couple of
variables that do not have much much missingness, get that to converge,
and then add other variables one or a few at a time and see what happens.
The problem may be entirely in one variable or set of variables, which you
could then remove.
Best of luck,
Gary King
: Gary King, King(a)Harvard.Edu http://GKing.Harvard.Edu :
: Center for Basic Research Direct (617) 495-2027 :
: in the Social Sciences Assistant (617) 495-9271 :
: 34 Kirkland Street, Rm. 2 HU-MIT DC (617) 495-4734 :
: Harvard U, Cambridge, MA 02138 eFax (617) 812-8581 :
On Mon, 10 May 2004, Jose A. Aleman wrote:
To whom it may concern:
I have been working with Amelia for the past few days. My database consists
of short panels (the number of countries is large but the number of years
small) and I can't get Amelia to converge. I am using the minimum number of
variables I think the algorithm needs to make good guesses but it just runs
on forever. I have dropped the number of observations but then the program
does not have enough data to make imputations. Perhaps panel designs are a
bit more demanding and I should change the settings, but I'm not completely
sure of what to change, since some of the variables exhibit high
stationarity and some don't. What is the most time I can expect Amelia to
run for in these situations?
Thanks,
Jose Aleman
PhD Candidate
Politics Department
130 Corwin Hall
Princeton, NJ 08544
609.937.0190
609.258.2147
<http://www.princeton.edu/~jaaleman>
11 May
11 May
3:47 p.m.
New subject: Amelia does not converge
Prof. King,
Thanks for your prompt reply. I forgot to mention that I'm working with
event count data rather than surveys. The two variables that most interest
me, however, are the ones with high missingness...
Jose Aleman
PhD Candidate
Politics Department
130 Corwin Hall
Princeton, NJ 08544
609.937.0190
609.258.2147
-----Original Message-----
From: Gary King [mailto:king@harvard.edu]
Sent: Monday, May 10, 2004 5:17 PM
To: Jose A. Aleman
Cc: Amelia Listserv
Subject: Re: Amelia does not converge
the theorem only says that if you let it run it will eventually finish, but
there are problems sometimes. almost all the time the problem is due to
difficult or impossible datasets, such as if your panel design is entirely
missing for all the variables for respondents who didn't answer one of the
waves of the survey. I suggest you start with just a couple of variables
that do not have much much missingness, get that to converge, and then add
other variables one or a few at a time and see what happens. The problem
may be entirely in one variable or set of variables, which you could then
remove.
Best of luck,
Gary King
: Gary King, King(a)Harvard.Edu http://GKing.Harvard.Edu :
: Center for Basic Research Direct (617) 495-2027 :
: in the Social Sciences Assistant (617) 495-9271 :
: 34 Kirkland Street, Rm. 2 HU-MIT DC (617) 495-4734 :
: Harvard U, Cambridge, MA 02138 eFax (617) 812-8581 :
On Mon, 10 May 2004, Jose A. Aleman wrote:
To whom it may concern:
I have been working with Amelia for the past few days. My database
consists of short panels (the number of countries is large but the
number of years
small) and I can't get Amelia to converge. I am using the minimum number
of
variables I think the algorithm needs to make good
guesses but it just
runs
on forever. I have dropped the number of observations
but then the program
does not have enough data to make imputations. Perhaps panel designs are a
bit more demanding and I should change the settings, but I'm not
completely
sure of what to change, since some of the variables
exhibit high
stationarity and some don't. What is the most time I can expect Amelia to
run for in these situations?
Thanks,
Jose Aleman
PhD Candidate
Politics Department
130 Corwin Hall
Princeton, NJ 08544
609.937.0190
609.258.2147
<http://www.princeton.edu/~jaaleman>
4:24 p.m.
New subject: Amelia does not converge
i'd still suggest following the strategy, below. you may need to increase
the prior value in _AMempri to get it to work. but be aware that the more
missingness, the more model-dependent the imputations will be. i.e., if
there is no information even Amelia won't find it for you!
Best of luck,
Gary
On Tue, 11 May 2004, Jose A. Aleman wrote:
Prof. King,
Thanks for your prompt reply. I forgot to mention that I'm working with
event count data rather than surveys. The two variables that most interest
me, however, are the ones with high missingness...
Jose Aleman
PhD Candidate
Politics Department
130 Corwin Hall
Princeton, NJ 08544
609.937.0190
609.258.2147
-----Original Message-----
From: Gary King [mailto:king@harvard.edu]
Sent: Monday, May 10, 2004 5:17 PM
To: Jose A. Aleman
Cc: Amelia Listserv
Subject: Re: Amelia does not converge
the theorem only says that if you let it run it will eventually finish, but
there are problems sometimes. almost all the time the problem is due to
difficult or impossible datasets, such as if your panel design is entirely
missing for all the variables for respondents who didn't answer one of the
waves of the survey. I suggest you start with just a couple of variables
that do not have much much missingness, get that to converge, and then add
other variables one or a few at a time and see what happens. The problem
may be entirely in one variable or set of variables, which you could then
remove.
Best of luck,
Gary King
: Gary King, King(a)Harvard.Edu http://GKing.Harvard.Edu :
: Center for Basic Research Direct (617) 495-2027 :
: in the Social Sciences Assistant (617) 495-9271 :
: 34 Kirkland Street, Rm. 2 HU-MIT DC (617) 495-4734 :
: Harvard U, Cambridge, MA 02138 eFax (617) 812-8581 :
On Mon, 10 May 2004, Jose A. Aleman wrote:
To whom it may concern:
I have been working with Amelia for the past few days. My database
consists of short panels (the number of countries is large but the
number of years
small) and I can't get Amelia to converge. I am using the minimum number
of
variables I think the algorithm needs to make
good guesses but it just
runs
on forever. I have dropped the number of
observations but then the program
does not have enough data to make imputations. Perhaps panel designs are a
bit more demanding and I should change the settings, but I'm not
completely
sure of what to change, since some of the
variables exhibit high
stationarity and some don't. What is the most time I can expect Amelia to
run for in these situations?
Thanks,
Jose Aleman
PhD Candidate
Politics Department
130 Corwin Hall
Princeton, NJ 08544
609.937.0190
609.258.2147
<http://www.princeton.edu/~jaaleman>
7294
days inactive
7295
days old
amelia@lists.gking.harvard.edu
2 comments
2 participants
participants (2)
-
Gary King -
Jose A. Aleman