2 Nov
2004
2 Nov
'04
4:10 a.m.
Good mornig everyone,
I have 2 questiions and 1 news.
How to control the number of digits in the presentation of matrix?
As for ps6 #1(b),
when I reparametarize sigma^2 by exp(r)
and get an estimated r and its se,
how can I un-reparametarize r's se into sigma^2's se?
FYI, if you plan to publish your publishable paper,
why don't you make it public at the next year's APSA?
http://www.apsanet.org/mtgs/
The propsal deadline comes on Nov. 15.
Kentaro
2 Nov
2 Nov
4:40 a.m.
if you
sigma <- exp(par[length(par)])
you
sigma <- log(par[length(par)])
after you're done with optimization.
Or do the inverse of whatever function you did to paramterize in
the log-likelihood function.
----- Original Message -----
From: "Kentaro Fukumoto" <fukumoto(a)dg8.so-net.ne.jp>
To: <gov2001-l(a)lists.fas.harvard.edu>
Sent: Tuesday, November 02, 2004 9:09 AM
Subject: [gov2001-l] ps6 and APSA
Good mornig everyone,
I have 2 questiions and 1 news.
How to control the number of digits in the presentation of
matrix?
As for ps6 #1(b),
when I reparametarize sigma^2 by exp(r)
and get an estimated r and its se, how can I un-reparametarize
r's se into sigma^2's se?
FYI, if you plan to publish your publishable paper,
why don't you make it public at the next year's APSA?
http://www.apsanet.org/mtgs/
The propsal deadline comes on Nov. 15.
Kentaro
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
4:44 a.m.
actually, I think you only need the first one. after estimation the thing
in the last element of par will be gamma, so at that point you exp it to
get sigma.
the key tho is that this gizmo works only if there exists an inverse
function (in this case log). if there doesn't, then formally we say the
function is not one-to-one (which means that you can't invert it). so it
has to be possible or bad things happen, but you don't usually need to use
it (with the exception of creating starting values perhaps if you like to
think in terms of sigma instead of the gamma it expects).
Gary
On Tue, 2 Nov 2004, Olivia Lau wrote:
if you
sigma <- exp(par[length(par)])
you
sigma <- log(par[length(par)])
after you're done with optimization.
Or do the inverse of whatever function you did to paramterize in
the log-likelihood function.
----- Original Message -----
From: "Kentaro Fukumoto" <fukumoto(a)dg8.so-net.ne.jp>
To: <gov2001-l(a)lists.fas.harvard.edu>
Sent: Tuesday, November 02, 2004 9:09 AM
Subject: [gov2001-l] ps6 and APSA
Good mornig everyone,
I have 2 questiions and 1 news.
How to control the number of digits in the presentation of
matrix?
As for ps6 #1(b),
when I reparametarize sigma^2 by exp(r)
and get an estimated r and its se, how can I un-reparametarize
r's se into sigma^2's se?
FYI, if you plan to publish your publishable paper,
why don't you make it public at the next year's APSA?
http://www.apsanet.org/mtgs/
The propsal deadline comes on Nov. 15.
Kentaro
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
5:24 a.m.
Hi all,
Thank you, Gary and Olivia,
But what I ask you is standard error, not estimated parameters themselves.
let
Y ~ N(y | mu, sigma^2)
sigma^2 <- exp(gamma)
and then I get the MLE of gamma.
Of course, I can get sigma^2.hat by way of exp(gamma.hat).
But I think exp(var(gamma)) is not var(sigma^2).
My idea is that I can only report SE of log(sigma^2), not that of sigma^2 itself,
by using MLE and reparameterization.
----- Original Message -----
From: "Olivia Lau" <olau(a)fas.harvard.edu>
To: <gov2001-l(a)lists.fas.harvard.edu>
Sent: Tuesday, November 02, 2004 9:40 AM
Subject: Re: [gov2001-l] ps6 and APSA
if you
sigma <- exp(par[length(par)])
you
sigma <- log(par[length(par)])
after you're done with optimization.
Or do the inverse of whatever function you did to paramterize in
the log-likelihood function.
----- Original Message -----
From: "Kentaro Fukumoto" <fukumoto(a)dg8.so-net.ne.jp>
To: <gov2001-l(a)lists.fas.harvard.edu>
Sent: Tuesday, November 02, 2004 9:09 AM
Subject: [gov2001-l] ps6 and APSA
Good mornig everyone,
I have 2 questiions and 1 news.
How to control the number of digits in the presentation of
matrix?
As for ps6 #1(b),
when I reparametarize sigma^2 by exp(r)
and get an estimated r and its se, how can I un-reparametarize
r's se into sigma^2's se?
FYI, if you plan to publish your publishable paper,
why don't you make it public at the next year's APSA?
http://www.apsanet.org/mtgs/
The propsal deadline comes on Nov. 15.
Kentaro
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
5:26 a.m.
simulation: the answer to all questions about quantities of interest.
Gary
On Tue, 2 Nov 2004, Kentaro Fukumoto wrote:
Hi all,
Thank you, Gary and Olivia,
But what I ask you is standard error, not estimated parameters themselves.
let
Y ~ N(y | mu, sigma^2)
sigma^2 <- exp(gamma)
and then I get the MLE of gamma.
Of course, I can get sigma^2.hat by way of exp(gamma.hat).
But I think exp(var(gamma)) is not var(sigma^2).
My idea is that I can only report SE of log(sigma^2), not that of sigma^2 itself,
by using MLE and reparameterization.
----- Original Message -----
From: "Olivia Lau" <olau(a)fas.harvard.edu>
To: <gov2001-l(a)lists.fas.harvard.edu>
Sent: Tuesday, November 02, 2004 9:40 AM
Subject: Re: [gov2001-l] ps6 and APSA
if you
sigma <- exp(par[length(par)])
you
sigma <- log(par[length(par)])
after you're done with optimization.
Or do the inverse of whatever function you did to paramterize in
the log-likelihood function.
----- Original Message -----
From: "Kentaro Fukumoto" <fukumoto(a)dg8.so-net.ne.jp>
To: <gov2001-l(a)lists.fas.harvard.edu>
Sent: Tuesday, November 02, 2004 9:09 AM
Subject: [gov2001-l] ps6 and APSA
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
Good mornig everyone,
I have 2 questiions and 1 news.
How to control the number of digits in the presentation of
matrix?
As for ps6 #1(b),
when I reparametarize sigma^2 by exp(r)
and get an estimated r and its se, how can I un-reparametarize
r's se into sigma^2's se?
FYI, if you plan to publish your publishable paper,
why don't you make it public at the next year's APSA?
http://www.apsanet.org/mtgs/
The propsal deadline comes on Nov. 15.
Kentaro
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l 
3:18 p.m.
Oh yes, seeing is believing.
Regards,
Paul.
Quoting Gary King <king(a)harvard.edu>du>:
simulation: the answer to all questions about quantities of interest.
Gary
On Tue, 2 Nov 2004, Kentaro Fukumoto wrote:
Hi all,
Thank you, Gary and Olivia,
But what I ask you is standard error, not estimated parameters themselves.
let
Y ~ N(y | mu, sigma^2)
sigma^2 <- exp(gamma)
and then I get the MLE of gamma.
Of course, I can get sigma^2.hat by way of exp(gamma.hat).
But I think exp(var(gamma)) is not var(sigma^2).
My idea is that I can only report SE of log(sigma^2), not that of sigma^2
itself,
by using MLE and reparameterization.
----- Original Message -----
From: "Olivia Lau" <olau(a)fas.harvard.edu>
To: <gov2001-l(a)lists.fas.harvard.edu>
Sent: Tuesday, November 02, 2004 9:40 AM
Subject: Re: [gov2001-l] ps6 and APSA
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
if you
sigma <- exp(par[length(par)])
you
sigma <- log(par[length(par)])
after you're done with optimization.
Or do the inverse of whatever function you did to paramterize in
the log-likelihood function.
----- Original Message -----
From: "Kentaro Fukumoto" <fukumoto(a)dg8.so-net.ne.jp>
To: <gov2001-l(a)lists.fas.harvard.edu>
Sent: Tuesday, November 02, 2004 9:09 AM
Subject: [gov2001-l] ps6 and APSA
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
Good mornig everyone,
I have 2 questiions and 1 news.
How to control the number of digits in the presentation of
matrix?
As for ps6 #1(b),
when I reparametarize sigma^2 by exp(r)
and get an estimated r and its se, how can I un-reparametarize
r's se into sigma^2's se?
FYI, if you plan to publish your publishable paper,
why don't you make it public at the next year's APSA?
http://www.apsanet.org/mtgs/
The propsal deadline comes on Nov. 15.
Kentaro
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
7138
days inactive
7138
days old
gov2001@lists.gking.harvard.edu
5 comments
4 participants
participants (4)
-
fukumoto＠dg8.so-net.ne.jp -
king＠harvard.edu -
mschoi＠fas.harvard.edu -
olau＠fas.harvard.edu