27 Oct
2004
27 Oct
'04
5:45 p.m.
Hi,
Does anybody know...
1) How to clean up a R console?
2) In LaTeX, how to show the date other than \today by \maketitle?
3) When I do coauthor work,
I highlight what parts I revise on Word.
How can we do this by XEmacs?
4) what is the natural conjugate prior distribution
of negative binomial (poisson * gamma) or beta?
(I do not ask beta is conjugate for what distribution
(actually, for bernoulli related distributions.))
Thank you in advance!
Kentaro
27 Oct
27 Oct
5:54 p.m.
On Wed, 27 Oct 2004, Kentaro Fukumoto wrote:
Hi,
Does anybody know...
1) How to clean up a R console?
2) In LaTeX, how to show the date other than \today by \maketitle?
you can put \today wherever you like.
3) When I do coauthor work,
I highlight what parts I revise on Word.
How can we do this by XEmacs?
there's something called CVS (concurrent versions system) that does this
even better than Word. believe it or not, its easy with CVS and latex for
2 people to be editing the same document at the same time without any
confusion. for now while you're getting used to everything else I'd skip
CVS, but later its something to look into; there's a link at my homepage.
i use it all the time.
4) what is the natural conjugate prior distribution
of negative binomial (poisson * gamma) or beta?
(I do not ask beta is conjugate for what distribution
(actually, for bernoulli related distributions.))
by conjugate, Kentaro means the choice of a prior distribution such that
the integral works out analytically without approximation. the conjugate
prior for the poisson is the gamma, and the result of the integral which
we've done in class is the negative binomial.
Gary
Thank you in advance!
Kentaro
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
28 Oct
28 Oct
7:20 p.m.
Hi all,
Thank you Gary and Olivia,
I think C^{a certain key} and C^{a certain
key} clean up without quitting R.
But I don't remember which keys.
I'll try it!
My question was misleading.
This is not related to the lecture, but to our final paper.
I know the conjugate prior for the poisson is the gamma.
Then, what the conjugate prior for the negative binomial is?
Kentaro
1) How to
clean up a R console?
Quit R, close the buffer, start R again. 3) When I do
coauthor work,
I highlight what parts I revise on Word.
How can we do this by XEmacs?
there's something called CVS (concurrent versions system) 4) what is the
natural conjugate prior distribution
of negative binomial (poisson * gamma) or beta?
(I do not ask beta is conjugate for what distribution
(actually, for bernoulli related distributions.))
by conjugate, Kentaro means the choice of a prior distribution such that
the integral works out analytically without approximation. the conjugate
prior for the poisson is the gamma, and the result of the integral which
we've done in class is the negative binomial. 
29 Oct
29 Oct
6:23 a.m.
C-c C-o cleans up the last output to the R buffer, but won't
clean up everything else. (Useful if you've just printed a big
matrix or data frame unintentionally.) To clean up the rest of
your buffer, quit R, save your workspace, close the buffer, and
start R again. If you have a large buffer > 3000 lines, you
should do this immediately.
There is no well-defined conjugate prior for the negative
binomial distribution. If you wish to use this for your paper
(I don't think you should), you will need to write down
* the prior p(phi, sigma2) (which may be p(phi) x p(sigma),
if you assume that the priors are independent),
* the negative binomial likelihood likelihood p(y|phi,
sigma2) and
* make sure that the joint density (the posterior) is a
proper probability distribution (that integrates to 1).
As I seem to recall, a mixture distribution like the negative
binomial can't be modeled via a simple one step procedure -- you
have to implement a hierarchical model to capture the dependency
of phi on sigma and vice versa.
Not every distribution has a conjugate prior -- these are
mathematical anomolies that give rise to a class of priors which
may or may not be able to model the actual prior *beliefs* for
which you would like ot account for in p(theta). In other
words, the assumption of the p(theta) functional form actually
restricts the type of prior information that you can include.
Olivia
----- Original Message -----
From: "Kentaro Fukumoto" <fukumoto(a)dg8.so-net.ne.jp>
To: <gov2001-l(a)lists.fas.harvard.edu>
Sent: Thursday, October 28, 2004 11:16 PM
Subject: Re: [gov2001-l] 4 questiuons
Hi all,
Thank you Gary and Olivia,
I think C^{a certain key} and C^{a
certain key} clean up
without quitting R.
But I don't remember which keys.
I'll try it!
My question was misleading.
This is not related to the lecture, but to our final paper.
I know the conjugate prior for the poisson is the gamma.
Then, what the conjugate prior for the negative binomial is?
Kentaro
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
1) How to
clean up a R console?
Quit R, close the buffer, start R again. 3) When I
do coauthor work,
I highlight what parts I revise on Word.
How can we do this by XEmacs?
there's something called CVS (concurrent versions system) 4) what
is the natural conjugate prior distribution
of negative binomial (poisson * gamma) or beta?
(I do not ask beta is conjugate for what distribution
(actually, for bernoulli related distributions.))
by conjugate, Kentaro means the choice of a prior
distribution such that
the integral works out analytically without approximation.
the conjugate
prior for the poisson is the gamma, and the result of the
integral which
we've done in class is the negative binomial.
6:31 a.m.
actually since the negbin model has an analytical form, you can use it
with the same likelihood etc procedures as you do for the normal.
the math and applications are different of course, but its logically the
same process, and it also has 2 parameters.
the beauty of simulation you should remember, tho, is that you dno't need
to compute everything analytically. even if we couldn't come up with the
form of the negbin density, we could draw from the gamma, stick that in
the Possion and draw from the binomial.
if you really care about conjugate priors and such, you might have a look
at the book Bayesian Data Analysis by Andrew Gelman et al. There's a good
appendix with all these.
And for a more general book on a gazillion distributions, there is 4
volume set called Distributions in Statistics by Johnson and Kotz.
Gary
On Fri, 29 Oct 2004, Olivia Lau wrote:
C-c C-o cleans up the last output to the R buffer, but
won't
clean up everything else. (Useful if you've just printed a big
matrix or data frame unintentionally.) To clean up the rest of
your buffer, quit R, save your workspace, close the buffer, and
start R again. If you have a large buffer > 3000 lines, you
should do this immediately.
There is no well-defined conjugate prior for the negative
binomial distribution. If you wish to use this for your paper
(I don't think you should), you will need to write down
* the prior p(phi, sigma2) (which may be p(phi) x p(sigma),
if you assume that the priors are independent),
* the negative binomial likelihood likelihood p(y|phi,
sigma2) and
* make sure that the joint density (the posterior) is a
proper probability distribution (that integrates to 1).
As I seem to recall, a mixture distribution like the negative
binomial can't be modeled via a simple one step procedure -- you
have to implement a hierarchical model to capture the dependency
of phi on sigma and vice versa.
Not every distribution has a conjugate prior -- these are
mathematical anomolies that give rise to a class of priors which
may or may not be able to model the actual prior *beliefs* for
which you would like ot account for in p(theta). In other
words, the assumption of the p(theta) functional form actually
restricts the type of prior information that you can include.
Olivia
----- Original Message -----
From: "Kentaro Fukumoto" <fukumoto(a)dg8.so-net.ne.jp>
To: <gov2001-l(a)lists.fas.harvard.edu>
Sent: Thursday, October 28, 2004 11:16 PM
Subject: Re: [gov2001-l] 4 questiuons
Hi all,
Thank you Gary and Olivia,
I think C^{a certain key} and C^{a
certain key} clean up
without quitting R.
But I don't remember which keys.
I'll try it!
My question was misleading.
This is not related to the lecture, but to our final paper.
I know the conjugate prior for the poisson is the gamma.
Then, what the conjugate prior for the negative binomial is?
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
1) How to
clean up a R console?
Quit R, close the buffer, start R again. 3) When I
do coauthor work,
I highlight what parts I revise on Word.
How can we do this by XEmacs?
there's something called CVS (concurrent versions system) 4) what
is the natural conjugate prior distribution
of negative binomial (poisson * gamma) or beta?
(I do not ask beta is conjugate for what distribution
(actually, for bernoulli related distributions.))
by conjugate, Kentaro means the choice of a prior
distribution such that
the integral works out analytically without approximation.
the conjugate
prior for the poisson is the gamma, and the result of the
integral which
we've done in class is the negative binomial. 
28 Oct
28 Oct
6:39 a.m.
New subject: [gov2001-l] identifying coordinates
Good morning everyone,
In question 3, I'm trying to identify the coordinates of phi and sigma2
(that are separate vectors) that correspond to my maximum value of my grid
output. I'm using the command:
phi[output == max(output)], and
sigma[output == max(output)]
but for some reason is giving me in both cases an "NA" answer. It works fine
by identifying the maximum when I do:
output[output == max(output)]
Could someone explain me what I'm doing wrong? Thanks a lot!!
Cecilia
7 a.m.
New subject: [gov2001-l] identifying coordinates
I had the same problem a few minutes ago.
When you do phi[output == max(output)], it is giving you something
like phi[571], where 571 is the number of the entry in <output> where
<output? is at a maximum -- not the x coordinate or the y-coordinate,
but the entry number if you listed them all in order like (1,1),
(1,2), (1,3) . . (1,length(sigmasq)), (2,1) . . (length(phi),
length(sigmasq)). It doesn't know that you want the y coordinate of
the max(output). So it gives you na because your phi vector is not as
long as 571 or whatever that location is.
I solved this by getting the number of the entry (in the above case,
571), and dividing it by the length of phi for the phi coordinate and
the length of sigmasq for the sigmasq coordinate. so the phi[output ==
max(output)] is phi[571/length(phi)].
I hope that makes sense.
Andy
On Thu, 28 Oct 2004 10:39:36 -0400, Cecilia Vidal
<cecilia_vidal(a)harvard.edu> wrote:
Good morning everyone,
In question 3, I'm trying to identify the coordinates of phi and sigma2
(that are separate vectors) that correspond to my maximum value of my grid
output. I'm using the command:
phi[output == max(output)], and
sigma[output == max(output)]
but for some reason is giving me in both cases an "NA" answer. It works fine
by identifying the maximum when I do:
output[output == max(output)]
Could someone explain me what I'm doing wrong? Thanks a lot!!
Cecilia
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
7:44 a.m.
New subject: [gov2001-l] identifying coordinates
Just a note one what I wrote below -- I wouldn't try exactly what I
suggest, since it gave me some wrong answers. I'm not yet sure why.
But I tried a similar way and it worked properly.
ae.
On Thu, 28 Oct 2004 11:00:42 -0400, Andrew Eggers <aeggers(a)gmail.com> wrote:
I had the same problem a few minutes ago.
When you do phi[output == max(output)], it is giving you something
like phi[571], where 571 is the number of the entry in <output> where
<output? is at a maximum -- not the x coordinate or the y-coordinate,
but the entry number if you listed them all in order like (1,1),
(1,2), (1,3) . . (1,length(sigmasq)), (2,1) . . (length(phi),
length(sigmasq)). It doesn't know that you want the y coordinate of
the max(output). So it gives you na because your phi vector is not as
long as 571 or whatever that location is.
I solved this by getting the number of the entry (in the above case,
571), and dividing it by the length of phi for the phi coordinate and
the length of sigmasq for the sigmasq coordinate. so the phi[output ==
max(output)] is phi[571/length(phi)].
I hope that makes sense.
Andy
On Thu, 28 Oct 2004 10:39:36 -0400, Cecilia Vidal
<cecilia_vidal(a)harvard.edu> wrote:
Good morning everyone,
In question 3, I'm trying to identify the coordinates of phi and sigma2
(that are separate vectors) that correspond to my maximum value of my grid
output. I'm using the command:
phi[output == max(output)], and
sigma[output == max(output)]
but for some reason is giving me in both cases an "NA" answer. It works fine
by identifying the maximum when I do:
output[output == max(output)]
Could someone explain me what I'm doing wrong? Thanks a lot!!
Cecilia
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
7:59 a.m.
New subject: [gov2001-l] identifying coordinates
First, use which to create an index with the row and column
location of the maximum:
idx <- which(joint, keep.ind = TRUE)
idx is a vector of length 2 where the first element corresponds
to the row location of the maximum and the second element
corresponds to the column location of the maximum.
Now use these two numbers to extract the corresponding phi and
sigma2 values from the appropriate row and column indices.
----- Original Message -----
From: "Andrew Eggers" <aeggers(a)gmail.com>
To: <cecilia_vidal(a)harvard.edu>du>;
<gov2001-l(a)lists.fas.harvard.edu>
Sent: Thursday, October 28, 2004 11:00 AM
Subject: Re: [gov2001-l] identifying coordinates
I had the same problem a few minutes ago.
When you do phi[output == max(output)], it is giving you
something
like phi[571], where 571 is the number of the entry in
<output> where
<output? is at a maximum -- not the x coordinate or the
y-coordinate,
but the entry number if you listed them all in order like
(1,1),
(1,2), (1,3) . . (1,length(sigmasq)), (2,1) . . (length(phi),
length(sigmasq)). It doesn't know that you want the y
coordinate of
the max(output). So it gives you na because your phi vector is
not as
long as 571 or whatever that location is.
I solved this by getting the number of the entry (in the above
case,
571), and dividing it by the length of phi for the phi
coordinate and
the length of sigmasq for the sigmasq coordinate. so the
phi[output ==
max(output)] is phi[571/length(phi)].
I hope that makes sense.
Andy
On Thu, 28 Oct 2004 10:39:36 -0400, Cecilia Vidal
<cecilia_vidal(a)harvard.edu> wrote:
Good morning everyone,
In question 3, I'm trying to identify the coordinates of phi
and sigma2
(that are separate vectors) that correspond to my maximum
value of my grid
output. I'm using the command:
phi[output == max(output)], and
sigma[output == max(output)]
but for some reason is giving me in both cases an "NA"
answer. It works fine
by identifying the maximum when I do:
output[output == max(output)]
Could someone explain me what I'm doing wrong? Thanks a lot!!
Cecilia
_______________________________________________
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
29 Oct
29 Oct
7:45 a.m.
New subject: [gov2001-l] identifying coordinates
Things need from Mexican friends:
(1) Each year's national tax schedule and local tax schedule: such as
consumption tax rate, income tax rate, etc.
(2) Each year's national tax information: such as consumption tax revenue,
income tax revenue, payroll tax revenue, etc.
(3) Census survey about income and expenditure, tax information for each
year.
(4) Any information about health care policy change in years which we have
datasets.
Notice that if possible, we would like to have income, expenditure, health
expenditure, tax information from two difference national sources.
Chunling Lu, PH.D.
Senior Research Associate
Harvard University
Initiative for Global Health
104 Mt. Auburn Street
Cambridge, MA 02138
Telephone: 617-495-4866
Fax: 617-495-8231
Email: clu(a)camail.harvard.edu
-----Original Message-----
From: gov2001-l-bounces(a)lists.fas.harvard.edu
[mailto:gov2001-l-bounces@lists.fas.harvard.edu]On Behalf Of Cecilia
Vidal
Sent: Thursday, October 28, 2004 10:40 AM
To: gov2001-l(a)lists.fas.harvard.edu
Subject: [gov2001-l] identifying coordinates
Good morning everyone,
In question 3, I'm trying to identify the coordinates of phi and sigma2
(that are separate vectors) that correspond to my maximum value of my grid
output. I'm using the command:
phi[output == max(output)], and
sigma[output == max(output)]
but for some reason is giving me in both cases an "NA" answer. It works fine
by identifying the maximum when I do:
output[output == max(output)]
Could someone explain me what I'm doing wrong? Thanks a lot!!
Cecilia
_______________________________________________
gov2001-l mailing list
gov2001-l(a)lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
9:53 a.m.
New subject: [gov2001-l] sorry guys, pushed the wrong button...
Sorry guys, accidently pushed the wrong button, please ignore this message.
Chunling
-----Original Message-----
From: gov2001-l-bounces(a)lists.fas.harvard.edu
[mailto:gov2001-l-bounces@lists.fas.harvard.edu]On Behalf Of Chunling Lu
Sent: Friday, October 29, 2004 11:46 AM
To: cecilia_vidal(a)harvard.edu; gov2001-l(a)lists.fas.harvard.edu
Subject: RE: [gov2001-l] identifying coordinates
Things need from Mexican friends:
(1) Each year's national tax schedule and local tax schedule: such as
consumption tax rate, income tax rate, etc.
(2) Each year's national tax information: such as consumption tax revenue,
income tax revenue, payroll tax revenue, etc.
(3) Census survey about income and expenditure, tax information for each
year.
(4) Any information about health care policy change in years which we have
datasets.
Notice that if possible, we would like to have income, expenditure, health
expenditure, tax information from two difference national sources.
Chunling Lu, PH.D.
Senior Research Associate
Harvard University
Initiative for Global Health
104 Mt. Auburn Street
Cambridge, MA 02138
Telephone: 617-495-4866
Fax: 617-495-8231
Email: clu(a)camail.harvard.edu
-----Original Message-----
From: gov2001-l-bounces(a)lists.fas.harvard.edu
[mailto:gov2001-l-bounces@lists.fas.harvard.edu]On Behalf Of Cecilia
Vidal
Sent: Thursday, October 28, 2004 10:40 AM
To: gov2001-l(a)lists.fas.harvard.edu
Subject: [gov2001-l] identifying coordinates
Good morning everyone,
In question 3, I'm trying to identify the coordinates of phi and sigma2
(that are separate vectors) that correspond to my maximum value of my grid
output. I'm using the command:
phi[output == max(output)], and
sigma[output == max(output)]
but for some reason is giving me in both cases an "NA" answer. It works fine
by identifying the maximum when I do:
output[output == max(output)]
Could someone explain me what I'm doing wrong? Thanks a lot!!
Cecilia
_______________________________________________
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
7124
days inactive
7126
days old
gov2001@lists.gking.harvard.edu
11 comments
6 participants
participants (6)
-
aeggers＠gmail.com -
cecilia_vidal＠harvard.edu -
chunling_lu＠harvard.edu -
fukumoto＠dg8.so-net.ne.jp -
king＠harvard.edu -
olau＠fas.harvard.edu