11 Aug
2005
11 Aug
'05
4:30 p.m.
Dear Olivia and Kosuke,
thank you very much for your kindness.
mlogit is taking the base-line category exactly as I thought
it was doing it.
But my problem remains, since the output is
z.out <- zelig(food ~ small + han + okl + tra,
model = "mlogit", data = alligator)
z.out
Call:
zelig(formula = food ~ small + han + okl + tra, model = "mlogit",
data = alligator)
Coefficients:
(Intercept):1 (Intercept):2 (Intercept):3 (Intercept):4 small:1 small:2
small:3 small:4 han:1 han:2 han:3
-0.6931 -13.4332 -13.4332 -13.4332 -0.2178 0.5949
-1.0014 -1.0256 1.7697 12.0783 12.6985
han:4 okl:1 okl:2 okl:3 okl:4 tra:1
tra:2 tra:3 tra:4
13.2221 2.5126 14.9878 14.5858 13.3380 1.0514
13.6991 13.5604 12.8731
Degrees of Freedom: 636 Total; 616 Residual
Residual Deviance: 418.9
Log-likelihood: -209.5
I assume that (for example) intercept1 refers to the intercept for the first level which
is not the base-line category (invertebrate). If that is so,
and I plug these estimates in the calculus of the log likelihood yields Log-likelihood:
-284.
Besides, the estimates are quite different from those reported by Agresti (2002):
AGRESTI (2002) mlogit
intercept Size Hancock Oklawaha Trafford intercept Size Hancock Oklawaha Trafford
Invertebrate -1.55 1.46 -1.66 0.94 1.12 Invertebrate -0.6931 -0.2178 1.7694 2.5126
1.0514
Reptile -3.31 -0.35 1.24 2.46 2.94 Reptile -13.433 0.5949 12.0783 14.9878 13.6991
Bird -2.09 -0.63 0.7 -0.65 1.09 Bird -13.433 -1.0014 12.6985 14.5858 13.5604
Other -1.9 0.33 0.83 0.01 1.52 Other -13.433 -1.0256 13.2221 13.338 12.8731
I believe I'm missing something really important here but I don't know what it
is, could you please give me a hint?
Thanks a lot in advance,
Alejandro
--------------------------------------------------------------------------------
Dear Alejandro,
All models that take a factor response (mlogit, oprobit, ologit, etc) as
the dependent variable use the first level as the base case.
Now you ask: What's the first level? For a numeric vector, the first
level is the smallest level. For a vector of character strings, the first
level is the first string value in alphabetical order.
Best,
Olivia
On Thu, 11 Aug 2005, Alejandro Buren wrote:
Dear all,
I'm new to Zelig, I found it looking for a way to fit multinomial logit models with
R.
I still haven't been able to figure out how "mlogit" chooses the base-line
category.
I tried running an example I'm familiarized with (Alligator Food Choice Example, Ex
7.1.2
from Agresti 2002 "Categorical Data Analysis"), but that didn't help me
to understand how it is working.
Could anyone help me out?
I'm pasting the code to run the example.
Any comment or help will be most appreciated,
alligator <- data.frame(
food =
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,3,3,4,4,5,5,5,5,5,5,5,5,1,1,1,1,1,1,1,2,3,4,4,4,5,5,5,5,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,4,5,5,5,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,3,3,3,3,3,3,4,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,3,4,5,5,5,5,5,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,3,3,3,3,3,3,4,4,4,5,5,5,5,5),
small =
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
han =
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
okl =
c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
tra =
c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1))
z.out <- zelig(food ~ small + han + okl + tra,
model = "mlogit", data = alligator)
--
Alejandro Daniel Buren
Graduate Student
Memorial University of Newfoundland
St. John's, Newfoundland A1B 3X9 Canada
e-mail: aburen@xxxxxx 
11 Aug
11 Aug
5:14 p.m.
New subject: mlogit help
Dear Alejandro,
The *last* level (rather than the first level) is the base category. In
our mexico example (type demo(mexico)), for example, "3" is the base
category.
As for the difference between the two results, if you find another
estimate that gives a greater value of the log-likelihood, your estimate
is definitely not MLE!! From your mesasge, it seems that VGAM results are
giving a better answer. Just to make sure they are calculating the same
log-likelihood (sometimes, different software drops different constants),
you might want to evaluate the log-likelihood for both estimates by hand.
Good luck,
Kosuke
-----------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
-----------------------------------------------------
On Thu, 11 Aug 2005, Alejandro Buren wrote:
Dear Olivia and Kosuke,
thank you very much for your kindness.
mlogit is taking the base-line category exactly as I thought
it was doing it.
But my problem remains, since the output is
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
z.out <- zelig(food ~ small + han + okl + tra,
model = "mlogit", data = alligator)
z.out
Call:
zelig(formula = food ~ small + han + okl + tra, model = "mlogit",
data = alligator)
Coefficients:
(Intercept):1 (Intercept):2 (Intercept):3 (Intercept):4 small:1 small:2
small:3 small:4 han:1 han:2 han:3
-0.6931 -13.4332 -13.4332 -13.4332 -0.2178 0.5949
-1.0014 -1.0256 1.7697 12.0783 12.6985
han:4 okl:1 okl:2 okl:3 okl:4 tra:1
tra:2 tra:3 tra:4
13.2221 2.5126 14.9878 14.5858 13.3380 1.0514
13.6991 13.5604 12.8731
Degrees of Freedom: 636 Total; 616 Residual
Residual Deviance: 418.9
Log-likelihood: -209.5
I assume that (for example) intercept1 refers to the intercept for the
first level which is not the base-line category (invertebrate). If that
is so,
and I plug these estimates in the calculus of the log likelihood yields
Log-likelihood: -284. Besides, the estimates are quite different from
those reported by Agresti (2002):
AGRESTI (2002) mlogit
intercept Size Hancock Oklawaha Trafford intercept Size Hancock Oklawaha
Trafford
Invertebrate -1.55 1.46 -1.66 0.94 1.12 Invertebrate -0.6931 -0.2178 1.7694 2.5126
1.0514
Reptile -3.31 -0.35 1.24 2.46 2.94 Reptile -13.433 0.5949 12.0783 14.9878 13.6991
Bird -2.09 -0.63 0.7 -0.65 1.09 Bird -13.433 -1.0014 12.6985 14.5858 13.5604
Other -1.9 0.33 0.83 0.01 1.52 Other -13.433 -1.0256 13.2221 13.338 12.8731
I believe I'm missing something really important here but I don't know what it
is, could you please give me a hint?
Thanks a lot in advance,
Alejandro
--------------------------------------------------------------------------------
Dear Alejandro,
All models that take a factor response (mlogit, oprobit, ologit, etc) as
the dependent variable use the first level as the base case.
Now you ask: What's the first level? For a numeric vector, the first
level is the smallest level. For a vector of character strings, the first
level is the first string value in alphabetical order.
Best,
Olivia
On Thu, 11 Aug 2005, Alejandro Buren wrote:
Dear all,
I'm new to Zelig, I found it looking for a way to fit multinomial logit models with
R.
I still haven't been able to figure out how "mlogit" chooses the base-line
category.
I tried running an example I'm familiarized with (Alligator Food Choice Example, Ex
7.1.2
from Agresti 2002 "Categorical Data Analysis"), but that didn't help me
to understand how it is working.
Could anyone help me out?
I'm pasting the code to run the example.
Any comment or help will be most appreciated,
alligator <- data.frame(
food =
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,3,3,4,4,5,5,5,5,5,5,5,5,1,1,1,1,1,1,1,2,3,4,4,4,5,5,5,5,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,4,5,5,5,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,3,3,3,3,3,3,4,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,3,4,5,5,5,5,5,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,3,3,3,3,3,3,4,4,4,5,5,5,5,5),
small =
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
han =
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
okl =
c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
tra =
c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1))
z.out <- zelig(food ~ small + han + okl + tra,
model = "mlogit", data = alligator)
--
Alejandro Daniel Buren
Graduate Student
Memorial University of Newfoundland
St. John's, Newfoundland A1B 3X9 Canada
e-mail: aburen@xxxxxx
12 Aug
12 Aug
9:02 a.m.
New subject: mlogit help
Dear Kosuke,
thanks for the tip, I've run the example, but there is something I'm still
missing.
Running the example, mlogit yields a set of parameters and a LogLikelihood;
when I use
that set of parameters to calculate the LogLikelihood myself (by hand) I get
a completely
different value.
mlogit is not working as I thougt it was, is there any document which
explains in more detail
than "Zelig: Everyone's statistical software" how it's working?
I would really appreciate it, since it's crucial for my research to be able
to fit these
multinomial logit models.
Thanks for your time and consideration,
-----------------------------------------------------
Alejandro Daniel Buren
Graduate Student
Memorial University of Newfoundland
St. John's, Newfoundland A1B 3X9 Canada
-----------------------------------------------------
----- Original Message -----
From: "Kosuke Imai" <kimai(a)Princeton.Edu>
To: "Alejandro Buren" <aled38(a)yahoo.com.ar>
Cc: <zelig(a)latte.harvard.edu>du>; "Olivia Lau" <>
Sent: Thursday, August 11, 2005 6:44 PM
Subject: Re: [zelig] Re: mlogit help
Dear Alejandro,
The *last* level (rather than the first level) is the base category. In
our mexico example (type demo(mexico)), for example, "3" is the base
category.
As for the difference between the two results, if you find another
estimate that gives a greater value of the log-likelihood, your estimate
is definitely not MLE!! From your mesasge, it seems that VGAM results are
giving a better answer. Just to make sure they are calculating the same
log-likelihood (sometimes, different software drops different constants),
you might want to evaluate the log-likelihood for both estimates by hand.
Good luck,
Kosuke
-----------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
-----------------------------------------------------
On Thu, 11 Aug 2005, Alejandro Buren wrote:
___________________________________________________________
1GB gratis, Antivirus y Antispam
Correo Yahoo!, el mejor correo web del mundo
http://correo.yahoo.com.ar
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Olivia and Kosuke,
thank you very much for your kindness.
mlogit is taking the base-line category exactly as I thought
it was doing it.
But my problem remains, since the output is
z.out <- zelig(food ~ small + han + okl + tra,
model = "mlogit", data =
alligator)
z.out
Call:
zelig(formula = food ~ small + han + okl + tra, model = "mlogit",
data = alligator)
Coefficients:
(Intercept):1 (Intercept):2 (Intercept):3 (Intercept):4 small:1
small:2 small:3 small:4 han:1 han:2
han:3
-0.6931 -13.4332 -13.4332 -13.4332 -0.2178
0.5949 -1.0014 -1.0256 1.7697 12.0783
12.6985
han:4 okl:1 okl:2 okl:3 okl:4
tra:1 tra:2 tra:3 tra:4
13.2221 2.5126 14.9878 14.5858 13.3380
1.0514 13.6991 13.5604 12.8731
Degrees of Freedom: 636 Total; 616 Residual
Residual Deviance: 418.9
Log-likelihood: -209.5
I assume that (for example) intercept1 refers to the intercept for the
first level which is not the base-line category (invertebrate). If that
is so,
and I plug these estimates in the calculus of the log likelihood yields
Log-likelihood: -284. Besides, the estimates are quite different from
those reported by Agresti (2002):
AGRESTI (2002) mlogit
intercept Size Hancock Oklawaha Trafford intercept Size Hancock
Oklawaha Trafford
Invertebrate -1.55 1.46 -1.66 0.94 1.12
Invertebrate -0.6931 -0.2178 1.7694 2.5126 1.0514
Reptile -3.31 -0.35 1.24 2.46 2.94 Reptile -13.433 0.5949 12.0783
14.9878 13.6991
Bird -2.09 -0.63 0.7 -0.65 1.09 Bird -13.433 -1.0014 12.6985
14.5858 13.5604
Other -1.9 0.33 0.83 0.01 1.52 Other -13.433 -1.0256 13.2221
13.338 12.8731
I believe I'm missing something really important here but I don't know
what it is, could you please give me a hint?
Thanks a lot in advance,
Alejandro
--------------------------------------------------------------------------------
Dear Alejandro,
All models that take a factor response (mlogit, oprobit, ologit, etc) as
the dependent variable use the first level as the base case.
Now you ask: What's the first level? For a numeric vector, the first
level is the smallest level. For a vector of character strings, the
first
level is the first string value in alphabetical order.
Best,
Olivia
On Thu, 11 Aug 2005, Alejandro Buren wrote:
Dear all,
I'm new to Zelig, I found it looking for a way to fit multinomial logit
models with R.
I still haven't been able to figure out how "mlogit" chooses the
base-line category.
I tried running an example I'm familiarized with (Alligator Food Choice
Example, Ex 7.1.2
from Agresti 2002 "Categorical Data Analysis"), but that didn't help
me to understand how it is working.
Could anyone help me out?
I'm pasting the code to run the example.
Any comment or help will be most appreciated,
alligator <- data.frame(
food =
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,3,3,4,4,5,5,5,5,5,5,5,5,1,1,1,1,1,1,1,2,3,4,4,4,5,5,5,5,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,4,5,5,5,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,3,3,3,3,3,3,4,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,3,4,5,5,5,5,5,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,3,3,3,3,3,3,4,4,4,5,5,5,5,5),
small =
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
han =
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
okl =
c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
tra =
c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1))
z.out <- zelig(food ~ small + han + okl + tra,
model = "mlogit", data
= alligator)
--
Alejandro Daniel Buren
Graduate Student
Memorial University of Newfoundland
St. John's, Newfoundland A1B 3X9 Canada
e-mail: aburen@xxxxxx 
11:52 a.m.
New subject: mlogit help
Our working theory anyway, is that the zelig documentation details exactly
how the program works. We've tested it but you never know of course with
code... if you're doing it 'by hand,' I'd suggest you pair your code
down
to its bare essentials -- a dichotomous outcome, one equation, just a
constant term, etc. see if that matches. if it does, then add one
feature at a time (a covariate, etc.). If you can isolate an issue please
do let us know.
Thanks,
Gary
---
Gary King
Institute for Quantitative Social Science
Harvard University, 34 Kirkland St, Cambridge, MA 02138
http://GKing.Harvard.Edu, King(a)Harvard.Edu
Direct 617-495-2027, Assistant 495-9271, eFax 812-8581
On Fri, 12 Aug 2005, Alejandro Buren wrote:
Dear Kosuke,
thanks for the tip, I've run the example, but there is something I'm still
missing.
Running the example, mlogit yields a set of parameters and a LogLikelihood;
when I use
that set of parameters to calculate the LogLikelihood myself (by hand) I get
a completely
different value.
mlogit is not working as I thougt it was, is there any document which
explains in more detail
than "Zelig: Everyone's statistical software" how it's working?
I would really appreciate it, since it's crucial for my research to be able
to fit these
multinomial logit models.
Thanks for your time and consideration,
-----------------------------------------------------
Alejandro Daniel Buren
Graduate Student
Memorial University of Newfoundland
St. John's, Newfoundland A1B 3X9 Canada
-----------------------------------------------------
----- Original Message ----- From: "Kosuke Imai" <kimai(a)Princeton.Edu>
To: "Alejandro Buren" <aled38(a)yahoo.com.ar>
Cc: <zelig(a)latte.harvard.edu>du>; "Olivia Lau" <>
Sent: Thursday, August 11, 2005 6:44 PM
Subject: Re: [zelig] Re: mlogit help
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Alejandro,
The *last* level (rather than the first level) is the base category. In
our mexico example (type demo(mexico)), for example, "3" is the base
category.
As for the difference between the two results, if you find another
estimate that gives a greater value of the log-likelihood, your estimate
is definitely not MLE!! From your mesasge, it seems that VGAM results are
giving a better answer. Just to make sure they are calculating the same
log-likelihood (sometimes, different software drops different constants),
you might want to evaluate the log-likelihood for both estimates by hand.
Good luck,
Kosuke
-----------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
-----------------------------------------------------
On Thu, 11 Aug 2005, Alejandro Buren wrote:
___________________________________________________________
1GB gratis, Antivirus y Antispam Correo Yahoo!, el mejor correo web del mundo
http://correo.yahoo.com.ar
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Olivia and Kosuke,
thank you very much for your kindness.
mlogit is taking the base-line category exactly as I thought
it was doing it.
But my problem remains, since the output is
z.out <- zelig(food ~ small + han + okl +
tra, model = "mlogit", data
= alligator)
z.out
Call:
zelig(formula = food ~ small + han + okl + tra, model = "mlogit",
data = alligator)
Coefficients:
(Intercept):1 (Intercept):2 (Intercept):3 (Intercept):4 small:1
small:2 small:3 small:4 han:1 han:2 han:3
-0.6931 -13.4332 -13.4332 -13.4332 -0.2178
0.5949 -1.0014 -1.0256 1.7697 12.0783 12.6985
han:4 okl:1 okl:2 okl:3 okl:4
tra:1 tra:2 tra:3 tra:4
13.2221 2.5126 14.9878 14.5858 13.3380
1.0514 13.6991 13.5604 12.8731
Degrees of Freedom: 636 Total; 616 Residual
Residual Deviance: 418.9
Log-likelihood: -209.5
I assume that (for example) intercept1 refers to the intercept for the
first level which is not the base-line category (invertebrate). If that
is so,
and I plug these estimates in the calculus of the log likelihood yields
Log-likelihood: -284. Besides, the estimates are quite different from
those reported by Agresti (2002):
AGRESTI (2002) mlogit
intercept Size Hancock Oklawaha Trafford intercept Size Hancock
Oklawaha Trafford
Invertebrate -1.55 1.46 -1.66 0.94 1.12 Invertebrate -0.6931
-0.2178 1.7694 2.5126 1.0514
Reptile -3.31 -0.35 1.24 2.46 2.94 Reptile -13.433 0.5949 12.0783
14.9878 13.6991
Bird -2.09 -0.63 0.7 -0.65 1.09 Bird -13.433 -1.0014 12.6985
14.5858 13.5604
Other -1.9 0.33 0.83 0.01 1.52 Other -13.433 -1.0256 13.2221
13.338 12.8731
I believe I'm missing something really important here but I don't know
what it is, could you please give me a hint?
Thanks a lot in advance,
Alejandro
--------------------------------------------------------------------------------
Dear Alejandro,
All models that take a factor response (mlogit, oprobit, ologit, etc) as
the dependent variable use the first level as the base case.
Now you ask: What's the first level? For a numeric vector, the first
level is the smallest level. For a vector of character strings, the
first
level is the first string value in alphabetical order.
Best,
Olivia
On Thu, 11 Aug 2005, Alejandro Buren wrote:
Dear all,
I'm new to Zelig, I found it looking for a way to fit multinomial
logit models with R.
I still haven't been able to figure out how "mlogit" chooses the
base-line category.
I tried running an example I'm familiarized with (Alligator Food
Choice Example, Ex 7.1.2
from Agresti 2002 "Categorical Data Analysis"), but that didn't help
me to understand how it is working.
Could anyone help me out?
I'm pasting the code to run the example.
Any comment or help will be most appreciated,
> alligator <- data.frame(
food =
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,3,3,4,4,5,5,5,5,5,5,5,5,1,1,1,1,1,1,1,2,3,4,4,4,5,5,5,5,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,4,5,5,5,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,3,3,3,3,3,3,4,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,3,4,5,5,5,5,5,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,3,3,3,3,3,3,4,4,4,5,5,5,5,5),
small =
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
han =
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
okl =
c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
tra =
c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1))
> z.out <- zelig(food ~ small + han + okl + tra, model = "mlogit",
> data = alligator)
--
Alejandro Daniel Buren
Graduate Student
Memorial University of Newfoundland
St. John's, Newfoundland A1B 3X9 Canada
e-mail: aburen@xxxxxx
12:20 p.m.
New subject: mlogit help
Alejandro --
It looks like zelig works (correctly) as described in the earlier
exchange. I have attached a simple comparison using R's optimizer vs.
zelig (and the default normalization).
Cheers,
Dan
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Daniel A. Powers, Ph.D.
Department of Sociology
University of Texas at Austin
1 University Station A1700
Austin, TX 78712-0118
phone: 512-232-6335
fax: 512-471-1748
dpowers(a)mail.la.utexas.edu
On Fri, 12 Aug 2005, Gary King wrote:
Our working theory anyway, is that the zelig documentation details exactly
how the program works. We've tested it but you never know of course with
code... if you're doing it 'by hand,' I'd suggest you pair your code
down
to its bare essentials -- a dichotomous outcome, one equation, just a
constant term, etc. see if that matches. if it does, then add one
feature at a time (a covariate, etc.). If you can isolate an issue please
do let us know.
Thanks,
Gary
---
Gary King
Institute for Quantitative Social Science
Harvard University, 34 Kirkland St, Cambridge, MA 02138
http://GKing.Harvard.Edu, King(a)Harvard.Edu
Direct 617-495-2027, Assistant 495-9271, eFax 812-8581
On Fri, 12 Aug 2005, Alejandro Buren wrote:
Dear Kosuke,
thanks for the tip, I've run the example, but there is something I'm still
missing.
Running the example, mlogit yields a set of parameters and a LogLikelihood;
when I use
that set of parameters to calculate the LogLikelihood myself (by hand) I get
a completely
different value.
mlogit is not working as I thougt it was, is there any document which
explains in more detail
than "Zelig: Everyone's statistical software" how it's working?
I would really appreciate it, since it's crucial for my research to be able
to fit these
multinomial logit models.
Thanks for your time and consideration,
-----------------------------------------------------
Alejandro Daniel Buren
Graduate Student
Memorial University of Newfoundland
St. John's, Newfoundland A1B 3X9 Canada
-----------------------------------------------------
----- Original Message ----- From: "Kosuke Imai" <kimai(a)Princeton.Edu>
To: "Alejandro Buren" <aled38(a)yahoo.com.ar>
Cc: <zelig(a)latte.harvard.edu>du>; "Olivia Lau" <>
Sent: Thursday, August 11, 2005 6:44 PM
Subject: Re: [zelig] Re: mlogit help
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Alejandro,
The *last* level (rather than the first level) is the base category. In
our mexico example (type demo(mexico)), for example, "3" is the base
category.
As for the difference between the two results, if you find another
estimate that gives a greater value of the log-likelihood, your estimate
is definitely not MLE!! From your mesasge, it seems that VGAM results are
giving a better answer. Just to make sure they are calculating the same
log-likelihood (sometimes, different software drops different constants),
you might want to evaluate the log-likelihood for both estimates by hand.
Good luck,
Kosuke
-----------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
-----------------------------------------------------
On Thu, 11 Aug 2005, Alejandro Buren wrote:
___________________________________________________________
1GB gratis, Antivirus y Antispam Correo Yahoo!, el mejor correo web del mundo
http://correo.yahoo.com.ar
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Olivia and Kosuke,
thank you very much for your kindness.
mlogit is taking the base-line category exactly as I thought
it was doing it.
But my problem remains, since the output is
> z.out <- zelig(food ~ small + han + okl + tra, model = "mlogit", data
> = alligator)
> z.out
Call:
zelig(formula = food ~ small + han + okl + tra, model = "mlogit",
data = alligator)
Coefficients:
(Intercept):1 (Intercept):2 (Intercept):3 (Intercept):4 small:1
small:2 small:3 small:4 han:1 han:2 han:3
-0.6931 -13.4332 -13.4332 -13.4332 -0.2178
0.5949 -1.0014 -1.0256 1.7697 12.0783 12.6985
han:4 okl:1 okl:2 okl:3 okl:4
tra:1 tra:2 tra:3 tra:4
13.2221 2.5126 14.9878 14.5858 13.3380
1.0514 13.6991 13.5604 12.8731
Degrees of Freedom: 636 Total; 616 Residual
Residual Deviance: 418.9
Log-likelihood: -209.5
I assume that (for example) intercept1 refers to the intercept for the
first level which is not the base-line category (invertebrate). If that
is so,
and I plug these estimates in the calculus of the log likelihood yields
Log-likelihood: -284. Besides, the estimates are quite different from
those reported by Agresti (2002):
AGRESTI (2002) mlogit
intercept Size Hancock Oklawaha Trafford intercept Size Hancock
Oklawaha Trafford
Invertebrate -1.55 1.46 -1.66 0.94 1.12 Invertebrate -0.6931
-0.2178 1.7694 2.5126 1.0514
Reptile -3.31 -0.35 1.24 2.46 2.94 Reptile -13.433 0.5949 12.0783
14.9878 13.6991
Bird -2.09 -0.63 0.7 -0.65 1.09 Bird -13.433 -1.0014 12.6985
14.5858 13.5604
Other -1.9 0.33 0.83 0.01 1.52 Other -13.433 -1.0256 13.2221
13.338 12.8731
I believe I'm missing something really important here but I don't know
what it is, could you please give me a hint?
Thanks a lot in advance,
Alejandro
--------------------------------------------------------------------------------
Dear Alejandro,
All models that take a factor response (mlogit, oprobit, ologit, etc) as
the dependent variable use the first level as the base case.
Now you ask: What's the first level? For a numeric vector, the first
level is the smallest level. For a vector of character strings, the
first
level is the first string value in alphabetical order.
Best,
Olivia
On Thu, 11 Aug 2005, Alejandro Buren wrote:
> Dear all,
> I'm new to Zelig, I found it looking for a way to fit multinomial
> logit models with R.
> I still haven't been able to figure out how "mlogit" chooses the
> base-line category.
> I tried running an example I'm familiarized with (Alligator Food
> Choice Example, Ex 7.1.2
> from Agresti 2002 "Categorical Data Analysis"), but that didn't
help
> me to understand how it is working.
> Could anyone help me out?
> I'm pasting the code to run the example.
> Any comment or help will be most appreciated,
>
>
> > alligator <- data.frame(
> food =
>
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,3,3,4,4,5,5,5,5,5,5,5,5,1,1,1,1,1,1,1,2,3,4,4,4,5,5,5,5,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,4,5,5,5,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,3,3,3,3,3,3,4,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,3,4,5,5,5,5,5,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,3,3,3,3,3,3,4,4,4,5,5,5,5,5),
> small =
>
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
> han =
>
c(1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
> okl =
>
c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
> tra =
>
c(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1))
>
>
>
> > z.out <- zelig(food ~ small + han + okl + tra, model = "mlogit",
> > data = alligator)
>
> --
> Alejandro Daniel Buren
> Graduate Student
> Memorial University of Newfoundland
> St. John's, Newfoundland A1B 3X9 Canada
>
> e-mail: aburen@xxxxxx
2:49 p.m.
New subject: mlogit help
Hi Alejandro,
We have the documentation at http://gking.harvard.edu/zelig/docs/
_TT_mlogit_TT__Multino.html In the next version, we will add the fact
that the last level of the outcome variable is used as the base
category in the multinomial logit model. If there is anything else
unclear about the documentation, please let us know since our goal is
to write a self-contained
As for the likelihood, I would follow Gary's suggestion of doing
something very simple - using a simulated data set where you know the
truth for example. I would also double-check the log-likelihood
formula you are using: make sure you do all the calculation on the
log-scale - sometimes multiplying and dividing by very small or large
numbers make the computer calculation incorrect. Doing all the
calculation on the log-scale should help.
Good luck,
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics Fax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
---------------------------------------------------------
On Aug 12, 2005, at 9:02 AM, Alejandro Buren wrote:
Dear Kosuke,
thanks for the tip, I've run the example, but there is something
I'm still missing.
Running the example, mlogit yields a set of parameters and a
LogLikelihood; when I use
that set of parameters to calculate the LogLikelihood myself (by
hand) I get a completely
different value.
mlogit is not working as I thougt it was, is there any document
which explains in more detail
than "Zelig: Everyone's statistical software" how it's working?
I would really appreciate it, since it's crucial for my research to
be able to fit these
multinomial logit models.
Thanks for your time and consideration,
-----------------------------------------------------
Alejandro Daniel Buren
Graduate Student
Memorial University of Newfoundland
St. John's, Newfoundland A1B 3X9 Canada
-----------------------------------------------------
----- Original Message ----- From: "Kosuke Imai" <kimai(a)Princeton.Edu>
To: "Alejandro Buren" <aled38(a)yahoo.com.ar>
Cc: <zelig(a)latte.harvard.edu>du>; "Olivia Lau" <>
Sent: Thursday, August 11, 2005 6:44 PM
Subject: Re: [zelig] Re: mlogit help
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Alejandro,
The *last* level (rather than the first level) is the base
category. In
our mexico example (type demo(mexico)), for example, "3" is the base
category.
As for the difference between the two results, if you find another
estimate that gives a greater value of the log-likelihood, your
estimate
is definitely not MLE!! From your mesasge, it seems that VGAM
results are
giving a better answer. Just to make sure they are calculating the
same
log-likelihood (sometimes, different software drops different
constants),
you might want to evaluate the log-likelihood for both estimates
by hand.
Good luck,
Kosuke
-----------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
-----------------------------------------------------
On Thu, 11 Aug 2005, Alejandro Buren wrote:
> Dear Olivia and Kosuke,
> thank you very much for your kindness.
> mlogit is taking the base-line category exactly as I thought
> it was doing it.
> But my problem remains, since the output is
>
> > z.out <- zelig(food ~ small + han + okl + tra, model =
> "mlogit", data = > alligator)
> > z.out
> Call:
> zelig(formula = food ~ small + han + okl + tra, model = "mlogit",
> data = alligator)
>
> Coefficients:
> (Intercept):1 (Intercept):2 (Intercept):3 (Intercept):4
> small:1 small:2 small:3 small:4 han:1
> han:2 han:3
> -0.6931 -13.4332 -13.4332 -13.4332
> -0.2178 0.5949 -1.0014 -1.0256 1.7697
> 12.0783 12.6985
> han:4 okl:1 okl:2 okl:3
> okl:4 tra:1 tra:2 tra:3 tra:4
> 13.2221 2.5126 14.9878 14.5858
> 13.3380 1.0514 13.6991 13.5604 12.8731
>
> Degrees of Freedom: 636 Total; 616 Residual
> Residual Deviance: 418.9
> Log-likelihood: -209.5
>
>
> I assume that (for example) intercept1 refers to the intercept
> for the
> first level which is not the base-line category (invertebrate).
> If that
> is so,
> and I plug these estimates in the calculus of the log likelihood
> yields
> Log-likelihood: -284. Besides, the estimates are quite different
> from
> those reported by Agresti (2002):
>
> AGRESTI (2002) mlogit
> intercept Size Hancock Oklawaha Trafford intercept Size
> Hancock Oklawaha Trafford
> Invertebrate -1.55 1.46 -1.66 0.94 1.12 Invertebrate
> -0.6931 -0.2178 1.7694 2.5126 1.0514
> Reptile -3.31 -0.35 1.24 2.46 2.94 Reptile -13.433 0.5949
> 12.0783 14.9878 13.6991
> Bird -2.09 -0.63 0.7 -0.65 1.09 Bird -13.433 -1.0014
> 12.6985 14.5858 13.5604
> Other -1.9 0.33 0.83 0.01 1.52 Other -13.433 -1.0256
> 13.2221 13.338 12.8731
>
>
>
> I believe I'm missing something really important here but I don't
> know what it is, could you please give me a hint?
>
> Thanks a lot in advance,
>
> Alejandro
>
> --------------------------------------------------------------------
> ------------
>
>
> Dear Alejandro,
>
> All models that take a factor response (mlogit, oprobit, ologit,
> etc) as
> the dependent variable use the first level as the base case.
>
> Now you ask: What's the first level? For a numeric vector, the
> first
> level is the smallest level. For a vector of character strings,
> the first
> level is the first string value in alphabetical order.
>
> Best,
>
> Olivia
>
> On Thu, 11 Aug 2005, Alejandro Buren wrote:
>
> > Dear all,
> > I'm new to Zelig, I found it looking for a way to fit
> multinomial logit > models with R.
> > I still haven't been able to figure out how "mlogit" chooses
> the > base-line category.
> > I tried running an example I'm familiarized with (Alligator
> Food Choice > Example, Ex 7.1.2
> > from Agresti 2002 "Categorical Data Analysis"), but that
> didn't help > me to understand how it is working.
> > Could anyone help me out?
> > I'm pasting the code to run the example.
> > Any comment or help will be most appreciated,
> >
> >
> > > alligator <- data.frame(
> > food = > c
> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,3,3,4,4,5,5,5
> ,
> 5,5,5,5,5,1,1,1,1,1,1,1,2,3,4,4,4,5,5,5,5,5,1,1,1,1,1,2,2,2,2,2,2,2,
> 2,2,2,2,3,4,5,5,5,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,3,3,3,3,
> 3,3,4,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,3,4,5,5,5,5,5,1,1,1,1,1,1,
> 1,1,2,2,2,2,2,2,2,3,3,3,3,3,3,4,4,4,5,5,5,5,5),
> > small = > c
> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
> ,
> 1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
> > han = > c
> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
> ,
> 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
> > okl = > c
> (0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
> ,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
> > tra = > c
> (0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
> ,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1))
> >
> >
> >
> > > z.out <- zelig(food ~ small + han + okl + tra, model =
> "mlogit", data > > = alligator)
> >
> > --
> > Alejandro Daniel Buren
> > Graduate Student
> > Memorial University of Newfoundland
> > St. John's, Newfoundland A1B 3X9 Canada
> >
> > e-mail: aburen@xxxxxx
>
>
>
___________________________________________________________ 1GB
gratis, Antivirus y Antispam Correo Yahoo!, el mejor correo web del
mundo http://correo.yahoo.com.ar
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
14 Aug
14 Aug
2:50 p.m.
New subject: mlogit help
Kosuke --
Any hints on obtaining the coefficient vector from the zelig mlogit model?
Apparently it is not of the form
m1$coef, or variants thereof, as documented.
Thanks,
Dan
%%%%%%%%%%%%%%%%%%%%%%%%
Daniel A. Powers, Ph.D.
Department of Sociology
University of Texas at Austin
1 University Station A1700
Austin, TX 78712-0118
phone: 512-232-6335
fax: 512-471-1748
dpowers(a)mail.la.utexas.edu
-----Original Message-----
From: owner-zelig(a)latte.harvard.edu
[mailto:owner-zelig@latte.harvard.edu]On Behalf Of Kosuke Imai
Sent: Friday, August 12, 2005 1:49 PM
To: Alejandro Buren
Cc: zelig(a)latte.harvard.edu; olau(a)fas.harvard.edu
Subject: Re: [zelig] Re: mlogit help
Hi Alejandro,
We have the documentation at http://gking.harvard.edu/zelig/docs/
_TT_mlogit_TT__Multino.html In the next version, we will add the fact
that the last level of the outcome variable is used as the base
category in the multinomial logit model. If there is anything else
unclear about the documentation, please let us know since our goal is
to write a self-contained
As for the likelihood, I would follow Gary's suggestion of doing
something very simple - using a simulated data set where you know the
truth for example. I would also double-check the log-likelihood
formula you are using: make sure you do all the calculation on the
log-scale - sometimes multiplying and dividing by very small or large
numbers make the computer calculation incorrect. Doing all the
calculation on the log-scale should help.
Good luck,
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics Fax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
---------------------------------------------------------
On Aug 12, 2005, at 9:02 AM, Alejandro Buren wrote:
Dear Kosuke,
thanks for the tip, I've run the example, but there is something
I'm still missing.
Running the example, mlogit yields a set of parameters and a
LogLikelihood; when I use
that set of parameters to calculate the LogLikelihood myself (by
hand) I get a completely
different value.
mlogit is not working as I thougt it was, is there any document
which explains in more detail
than "Zelig: Everyone's statistical software" how it's working?
I would really appreciate it, since it's crucial for my research to
be able to fit these
multinomial logit models.
Thanks for your time and consideration,
-----------------------------------------------------
Alejandro Daniel Buren
Graduate Student
Memorial University of Newfoundland
St. John's, Newfoundland A1B 3X9 Canada
-----------------------------------------------------
----- Original Message ----- From: "Kosuke Imai" <kimai(a)Princeton.Edu>
To: "Alejandro Buren" <aled38(a)yahoo.com.ar>
Cc: <zelig(a)latte.harvard.edu>du>; "Olivia Lau" <>
Sent: Thursday, August 11, 2005 6:44 PM
Subject: Re: [zelig] Re: mlogit help
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Alejandro,
The *last* level (rather than the first level) is the base
category. In
our mexico example (type demo(mexico)), for example, "3" is the base
category.
As for the difference between the two results, if you find another
estimate that gives a greater value of the log-likelihood, your
estimate
is definitely not MLE!! From your mesasge, it seems that VGAM
results are
giving a better answer. Just to make sure they are calculating the
same
log-likelihood (sometimes, different software drops different
constants),
you might want to evaluate the log-likelihood for both estimates
by hand.
Good luck,
Kosuke
-----------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
-----------------------------------------------------
On Thu, 11 Aug 2005, Alejandro Buren wrote:
> Dear Olivia and Kosuke,
> thank you very much for your kindness.
> mlogit is taking the base-line category exactly as I thought
> it was doing it.
> But my problem remains, since the output is
>
> > z.out <- zelig(food ~ small + han + okl + tra, model =
> "mlogit", data = > alligator)
> > z.out
> Call:
> zelig(formula = food ~ small + han + okl + tra, model = "mlogit",
> data = alligator)
>
> Coefficients:
> (Intercept):1 (Intercept):2 (Intercept):3 (Intercept):4
> small:1 small:2 small:3 small:4 han:1
> han:2 han:3
> -0.6931 -13.4332 -13.4332 -13.4332
> -0.2178 0.5949 -1.0014 -1.0256 1.7697
> 12.0783 12.6985
> han:4 okl:1 okl:2 okl:3
> okl:4 tra:1 tra:2 tra:3 tra:4
> 13.2221 2.5126 14.9878 14.5858
> 13.3380 1.0514 13.6991 13.5604 12.8731
>
> Degrees of Freedom: 636 Total; 616 Residual
> Residual Deviance: 418.9
> Log-likelihood: -209.5
>
>
> I assume that (for example) intercept1 refers to the intercept
> for the
> first level which is not the base-line category (invertebrate).
> If that
> is so,
> and I plug these estimates in the calculus of the log likelihood
> yields
> Log-likelihood: -284. Besides, the estimates are quite different
> from
> those reported by Agresti (2002):
>
> AGRESTI (2002) mlogit
> intercept Size Hancock Oklawaha Trafford intercept Size
> Hancock Oklawaha Trafford
> Invertebrate -1.55 1.46 -1.66 0.94 1.12 Invertebrate
> -0.6931 -0.2178 1.7694 2.5126 1.0514
> Reptile -3.31 -0.35 1.24 2.46 2.94 Reptile -13.433 0.5949
> 12.0783 14.9878 13.6991
> Bird -2.09 -0.63 0.7 -0.65 1.09 Bird -13.433 -1.0014
> 12.6985 14.5858 13.5604
> Other -1.9 0.33 0.83 0.01 1.52 Other -13.433 -1.0256
> 13.2221 13.338 12.8731
>
>
>
> I believe I'm missing something really important here but I don't
> know what it is, could you please give me a hint?
>
> Thanks a lot in advance,
>
> Alejandro
>
> --------------------------------------------------------------------
> ------------
>
>
> Dear Alejandro,
>
> All models that take a factor response (mlogit, oprobit, ologit,
> etc) as
> the dependent variable use the first level as the base case.
>
> Now you ask: What's the first level? For a numeric vector, the
> first
> level is the smallest level. For a vector of character strings,
> the first
> level is the first string value in alphabetical order.
>
> Best,
>
> Olivia
>
> On Thu, 11 Aug 2005, Alejandro Buren wrote:
>
> > Dear all,
> > I'm new to Zelig, I found it looking for a way to fit
> multinomial logit > models with R.
> > I still haven't been able to figure out how "mlogit" chooses
> the > base-line category.
> > I tried running an example I'm familiarized with (Alligator
> Food Choice > Example, Ex 7.1.2
> > from Agresti 2002 "Categorical Data Analysis"), but that
> didn't help > me to understand how it is working.
> > Could anyone help me out?
> > I'm pasting the code to run the example.
> > Any comment or help will be most appreciated,
> >
> >
> > > alligator <- data.frame(
> > food = > c
> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,3,3,4,4,5,5,5
> ,
> 5,5,5,5,5,1,1,1,1,1,1,1,2,3,4,4,4,5,5,5,5,5,1,1,1,1,1,2,2,2,2,2,2,2,
> 2,2,2,2,3,4,5,5,5,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,3,3,3,3,
> 3,3,4,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,3,4,5,5,5,5,5,1,1,1,1,1,1,
> 1,1,2,2,2,2,2,2,2,3,3,3,3,3,3,4,4,4,5,5,5,5,5),
> > small = > c
> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
> ,
> 1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
> > han = > c
> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
> ,
> 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
> > okl = > c
> (0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
> ,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
> > tra = > c
> (0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
> ,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1))
> >
> >
> >
> > > z.out <- zelig(food ~ small + han + okl + tra, model =
> "mlogit", data > > = alligator)
> >
> > --
> > Alejandro Daniel Buren
> > Graduate Student
> > Memorial University of Newfoundland
> > St. John's, Newfoundland A1B 3X9 Canada
> >
> > e-mail: aburen@xxxxxx
>
>
>
___________________________________________________________ 1GB
gratis, Antivirus y Antispam Correo Yahoo!, el mejor correo web del
mundo http://correo.yahoo.com.ar
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
3:29 p.m.
New subject: mlogit help
The manual says z.out$coefficients
see http://gking.harvard.edu/zelig/docs/Output_Values18.html
And in general names(z.out) will give you a list of the options.
Gary
---
Gary King,
Institute for Quantitative Social Science
Harvard University, 34 Kirkland St, Cambridge, MA 02138
http://GKing.Harvard.Edu, King(a)Harvard.Edu
Direct 617-495-2027, Assistant 495-9271, eFax 812-8581
On Sun, 14 Aug 2005, Dan Powers wrote:
Kosuke --
Any hints on obtaining the coefficient vector from the zelig mlogit model?
Apparently it is not of the form
m1$coef, or variants thereof, as documented.
Thanks,
Dan
%%%%%%%%%%%%%%%%%%%%%%%%
Daniel A. Powers, Ph.D.
Department of Sociology
University of Texas at Austin
1 University Station A1700
Austin, TX 78712-0118
phone: 512-232-6335
fax: 512-471-1748
dpowers(a)mail.la.utexas.edu
-----Original Message-----
From: owner-zelig(a)latte.harvard.edu
[mailto:owner-zelig@latte.harvard.edu]On Behalf Of Kosuke Imai
Sent: Friday, August 12, 2005 1:49 PM
To: Alejandro Buren
Cc: zelig(a)latte.harvard.edu; olau(a)fas.harvard.edu
Subject: Re: [zelig] Re: mlogit help
Hi Alejandro,
We have the documentation at http://gking.harvard.edu/zelig/docs/
_TT_mlogit_TT__Multino.html In the next version, we will add the fact
that the last level of the outcome variable is used as the base
category in the multinomial logit model. If there is anything else
unclear about the documentation, please let us know since our goal is
to write a self-contained
As for the likelihood, I would follow Gary's suggestion of doing
something very simple - using a simulated data set where you know the
truth for example. I would also double-check the log-likelihood
formula you are using: make sure you do all the calculation on the
log-scale - sometimes multiplying and dividing by very small or large
numbers make the computer calculation incorrect. Doing all the
calculation on the log-scale should help.
Good luck,
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics Fax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
---------------------------------------------------------
On Aug 12, 2005, at 9:02 AM, Alejandro Buren wrote:
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Kosuke,
thanks for the tip, I've run the example, but there is something
I'm still missing.
Running the example, mlogit yields a set of parameters and a
LogLikelihood; when I use
that set of parameters to calculate the LogLikelihood myself (by
hand) I get a completely
different value.
mlogit is not working as I thougt it was, is there any document
which explains in more detail
than "Zelig: Everyone's statistical software" how it's working?
I would really appreciate it, since it's crucial for my research to
be able to fit these
multinomial logit models.
Thanks for your time and consideration,
-----------------------------------------------------
Alejandro Daniel Buren
Graduate Student
Memorial University of Newfoundland
St. John's, Newfoundland A1B 3X9 Canada
-----------------------------------------------------
----- Original Message ----- From: "Kosuke Imai" <kimai(a)Princeton.Edu>
To: "Alejandro Buren" <aled38(a)yahoo.com.ar>
Cc: <zelig(a)latte.harvard.edu>du>; "Olivia Lau" <>
Sent: Thursday, August 11, 2005 6:44 PM
Subject: Re: [zelig] Re: mlogit help
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Alejandro,
The *last* level (rather than the first level) is the base
category. In
our mexico example (type demo(mexico)), for example, "3" is the base
category.
As for the difference between the two results, if you find another
estimate that gives a greater value of the log-likelihood, your
estimate
is definitely not MLE!! From your mesasge, it seems that VGAM
results are
giving a better answer. Just to make sure they are calculating the
same
log-likelihood (sometimes, different software drops different
constants),
you might want to evaluate the log-likelihood for both estimates
by hand.
Good luck,
Kosuke
-----------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
-----------------------------------------------------
On Thu, 11 Aug 2005, Alejandro Buren wrote:
> Dear Olivia and Kosuke,
> thank you very much for your kindness.
> mlogit is taking the base-line category exactly as I thought
> it was doing it.
> But my problem remains, since the output is
>
>> z.out <- zelig(food ~ small + han + okl + tra, model =
> "mlogit", data = > alligator)
>> z.out
> Call:
> zelig(formula = food ~ small + han + okl + tra, model = "mlogit",
> data = alligator)
>
> Coefficients:
> (Intercept):1 (Intercept):2 (Intercept):3 (Intercept):4
> small:1 small:2 small:3 small:4 han:1
> han:2 han:3
> -0.6931 -13.4332 -13.4332 -13.4332
> -0.2178 0.5949 -1.0014 -1.0256 1.7697
> 12.0783 12.6985
> han:4 okl:1 okl:2 okl:3
> okl:4 tra:1 tra:2 tra:3 tra:4
> 13.2221 2.5126 14.9878 14.5858
> 13.3380 1.0514 13.6991 13.5604 12.8731
>
> Degrees of Freedom: 636 Total; 616 Residual
> Residual Deviance: 418.9
> Log-likelihood: -209.5
>
>
> I assume that (for example) intercept1 refers to the intercept
> for the
> first level which is not the base-line category (invertebrate).
> If that
> is so,
> and I plug these estimates in the calculus of the log likelihood
> yields
> Log-likelihood: -284. Besides, the estimates are quite different
> from
> those reported by Agresti (2002):
>
> AGRESTI (2002) mlogit
> intercept Size Hancock Oklawaha Trafford intercept Size
> Hancock Oklawaha Trafford
> Invertebrate -1.55 1.46 -1.66 0.94 1.12 Invertebrate
> -0.6931 -0.2178 1.7694 2.5126 1.0514
> Reptile -3.31 -0.35 1.24 2.46 2.94 Reptile -13.433 0.5949
> 12.0783 14.9878 13.6991
> Bird -2.09 -0.63 0.7 -0.65 1.09 Bird -13.433 -1.0014
> 12.6985 14.5858 13.5604
> Other -1.9 0.33 0.83 0.01 1.52 Other -13.433 -1.0256
> 13.2221 13.338 12.8731
>
>
>
> I believe I'm missing something really important here but I don't
> know what it is, could you please give me a hint?
>
> Thanks a lot in advance,
>
> Alejandro
>
> --------------------------------------------------------------------
> ------------
>
>
> Dear Alejandro,
>
> All models that take a factor response (mlogit, oprobit, ologit,
> etc) as
> the dependent variable use the first level as the base case.
>
> Now you ask: What's the first level? For a numeric vector, the
> first
> level is the smallest level. For a vector of character strings,
> the first
> level is the first string value in alphabetical order.
>
> Best,
>
> Olivia
>
> On Thu, 11 Aug 2005, Alejandro Buren wrote:
>
>> Dear all,
>> I'm new to Zelig, I found it looking for a way to fit
> multinomial logit > models with R.
>> I still haven't been able to figure out how "mlogit" chooses
> the > base-line category.
>> I tried running an example I'm familiarized with (Alligator
> Food Choice > Example, Ex 7.1.2
>> from Agresti 2002 "Categorical Data Analysis"), but that
> didn't help > me to understand how it is working.
>> Could anyone help me out?
>> I'm pasting the code to run the example.
>> Any comment or help will be most appreciated,
>>
>>
>>> alligator <- data.frame(
>> food = > c
> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,3,3,4,4,5,5,5
> ,
> 5,5,5,5,5,1,1,1,1,1,1,1,2,3,4,4,4,5,5,5,5,5,1,1,1,1,1,2,2,2,2,2,2,2,
> 2,2,2,2,3,4,5,5,5,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,3,3,3,3,
> 3,3,4,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,3,4,5,5,5,5,5,1,1,1,1,1,1,
> 1,1,2,2,2,2,2,2,2,3,3,3,3,3,3,4,4,4,5,5,5,5,5),
>> small = > c
> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
> ,
> 1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
>> han = > c
> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
> ,
> 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
>> okl = > c
> (0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
> ,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
>> tra = > c
> (0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
> ,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1))
>>
>>
>>
>>> z.out <- zelig(food ~ small + han + okl + tra, model =
> "mlogit", data > > = alligator)
>>
>> --
>> Alejandro Daniel Buren
>> Graduate Student
>> Memorial University of Newfoundland
>> St. John's, Newfoundland A1B 3X9 Canada
>>
>> e-mail: aburen@xxxxxx
>
>
>
___________________________________________________________ 1GB
gratis, Antivirus y Antispam Correo Yahoo!, el mejor correo web del
mundo http://correo.yahoo.com.ar
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
4:18 p.m.
New subject: mlogit help
mlogit, like bprobit and blogit, use slots. Depending on the version of
Zelig that you have installed, you can use $ to extract slot elements. If $
doesn't work, either update your version of Zelig or use @ (the slot
extractor).
Best, Olivia
----- Original Message -----
From: "Gary King" <king(a)harvard.edu>
To: "Dan Powers" <dpowers(a)mail.la.utexas.edu>
Cc: "Kosuke Imai" <kimai(a)princeton.edu>du>; <zelig(a)latte.harvard.edu>du>;
<olau(a)fas.harvard.edu>
Sent: Sunday, August 14, 2005 3:29 PM
Subject: RE: [zelig] Re: mlogit help
The manual says z.out$coefficients
see http://gking.harvard.edu/zelig/docs/Output_Values18.html
And in general names(z.out) will give you a list of the options.
Gary
---
Gary King,
Institute for Quantitative Social Science
Harvard University, 34 Kirkland St, Cambridge, MA 02138
http://GKing.Harvard.Edu, King(a)Harvard.Edu
Direct 617-495-2027, Assistant 495-9271, eFax 812-8581
On Sun, 14 Aug 2005, Dan Powers wrote:
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Kosuke --
Any hints on obtaining the coefficient vector from the zelig mlogit
model?
Apparently it is not of the form
m1$coef, or variants thereof, as documented.
Thanks,
Dan
%%%%%%%%%%%%%%%%%%%%%%%%
Daniel A. Powers, Ph.D.
Department of Sociology
University of Texas at Austin
1 University Station A1700
Austin, TX 78712-0118
phone: 512-232-6335
fax: 512-471-1748
dpowers(a)mail.la.utexas.edu
-----Original Message-----
From: owner-zelig(a)latte.harvard.edu
[mailto:owner-zelig@latte.harvard.edu]On Behalf Of Kosuke Imai
Sent: Friday, August 12, 2005 1:49 PM
To: Alejandro Buren
Cc: zelig(a)latte.harvard.edu; olau(a)fas.harvard.edu
Subject: Re: [zelig] Re: mlogit help
Hi Alejandro,
We have the documentation at http://gking.harvard.edu/zelig/docs/
_TT_mlogit_TT__Multino.html In the next version, we will add the fact
that the last level of the outcome variable is used as the base
category in the multinomial logit model. If there is anything else
unclear about the documentation, please let us know since our goal is
to write a self-contained
As for the likelihood, I would follow Gary's suggestion of doing
something very simple - using a simulated data set where you know the
truth for example. I would also double-check the log-likelihood
formula you are using: make sure you do all the calculation on the
log-scale - sometimes multiplying and dividing by very small or large
numbers make the computer calculation incorrect. Doing all the
calculation on the log-scale should help.
Good luck,
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics Fax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
---------------------------------------------------------
On Aug 12, 2005, at 9:02 AM, Alejandro Buren wrote:
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Kosuke,
thanks for the tip, I've run the example, but there is something
I'm still missing.
Running the example, mlogit yields a set of parameters and a
LogLikelihood; when I use
that set of parameters to calculate the LogLikelihood myself (by
hand) I get a completely
different value.
mlogit is not working as I thougt it was, is there any document
which explains in more detail
than "Zelig: Everyone's statistical software" how it's working?
I would really appreciate it, since it's crucial for my research to
be able to fit these
multinomial logit models.
Thanks for your time and consideration,
-----------------------------------------------------
Alejandro Daniel Buren
Graduate Student
Memorial University of Newfoundland
St. John's, Newfoundland A1B 3X9 Canada
-----------------------------------------------------
----- Original Message ----- From: "Kosuke Imai" <kimai(a)Princeton.Edu>
To: "Alejandro Buren" <aled38(a)yahoo.com.ar>
Cc: <zelig(a)latte.harvard.edu>du>; "Olivia Lau" <>
Sent: Thursday, August 11, 2005 6:44 PM
Subject: Re: [zelig] Re: mlogit help
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Alejandro,
The *last* level (rather than the first level) is the base
category. In
our mexico example (type demo(mexico)), for example, "3" is the base
category.
As for the difference between the two results, if you find another
estimate that gives a greater value of the log-likelihood, your
estimate
is definitely not MLE!! From your mesasge, it seems that VGAM
results are
giving a better answer. Just to make sure they are calculating the
same
log-likelihood (sometimes, different software drops different
constants),
you might want to evaluate the log-likelihood for both estimates
by hand.
Good luck,
Kosuke
-----------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics eFax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
-----------------------------------------------------
On Thu, 11 Aug 2005, Alejandro Buren wrote:
> Dear Olivia and Kosuke,
> thank you very much for your kindness.
> mlogit is taking the base-line category exactly as I thought
> it was doing it.
> But my problem remains, since the output is
>
>> z.out <- zelig(food ~ small + han + okl + tra, model =
> "mlogit", data = > alligator)
>> z.out
> Call:
> zelig(formula = food ~ small + han + okl + tra, model = "mlogit",
> data = alligator)
>
> Coefficients:
> (Intercept):1 (Intercept):2 (Intercept):3 (Intercept):4
> small:1 small:2 small:3 small:4 han:1
> han:2 han:3
> -0.6931 -13.4332 -13.4332 -13.4332
> -0.2178 0.5949 -1.0014 -1.0256 1.7697
> 12.0783 12.6985
> han:4 okl:1 okl:2 okl:3
> okl:4 tra:1 tra:2 tra:3 tra:4
> 13.2221 2.5126 14.9878 14.5858
> 13.3380 1.0514 13.6991 13.5604 12.8731
>
> Degrees of Freedom: 636 Total; 616 Residual
> Residual Deviance: 418.9
> Log-likelihood: -209.5
>
>
> I assume that (for example) intercept1 refers to the intercept
> for the
> first level which is not the base-line category (invertebrate).
> If that
> is so,
> and I plug these estimates in the calculus of the log likelihood
> yields
> Log-likelihood: -284. Besides, the estimates are quite different
> from
> those reported by Agresti (2002):
>
> AGRESTI (2002) mlogit
> intercept Size Hancock Oklawaha Trafford intercept Size
> Hancock Oklawaha Trafford
> Invertebrate -1.55 1.46 -1.66 0.94 1.12 Invertebrate
> -0.6931 -0.2178 1.7694 2.5126 1.0514
> Reptile -3.31 -0.35 1.24 2.46 2.94 Reptile -13.433 0.5949
> 12.0783 14.9878 13.6991
> Bird -2.09 -0.63 0.7 -0.65 1.09 Bird -13.433 -1.0014
> 12.6985 14.5858 13.5604
> Other -1.9 0.33 0.83 0.01 1.52 Other -13.433 -1.0256
> 13.2221 13.338 12.8731
>
>
>
> I believe I'm missing something really important here but I don't
> know what it is, could you please give me a hint?
>
> Thanks a lot in advance,
>
> Alejandro
>
> --------------------------------------------------------------------
> ------------
>
>
> Dear Alejandro,
>
> All models that take a factor response (mlogit, oprobit, ologit,
> etc) as
> the dependent variable use the first level as the base case.
>
> Now you ask: What's the first level? For a numeric vector, the
> first
> level is the smallest level. For a vector of character strings,
> the first
> level is the first string value in alphabetical order.
>
> Best,
>
> Olivia
>
> On Thu, 11 Aug 2005, Alejandro Buren wrote:
>
>> Dear all,
>> I'm new to Zelig, I found it looking for a way to fit
> multinomial logit > models with R.
>> I still haven't been able to figure out how "mlogit" chooses
> the > base-line category.
>> I tried running an example I'm familiarized with (Alligator
> Food Choice > Example, Ex 7.1.2
>> from Agresti 2002 "Categorical Data Analysis"), but that
> didn't help > me to understand how it is working.
>> Could anyone help me out?
>> I'm pasting the code to run the example.
>> Any comment or help will be most appreciated,
>>
>>
>>> alligator <- data.frame(
>> food = > c
> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,3,3,4,4,5,5,5
> ,
> 5,5,5,5,5,1,1,1,1,1,1,1,2,3,4,4,4,5,5,5,5,5,1,1,1,1,1,2,2,2,2,2,2,2,
> 2,2,2,2,3,4,5,5,5,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,3,3,3,3,
> 3,3,4,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,3,4,5,5,5,5,5,1,1,1,1,1,1,
> 1,1,2,2,2,2,2,2,2,3,3,3,3,3,3,4,4,4,5,5,5,5,5),
>> small = > c
> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
> ,
> 1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
>> han = > c
> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
> ,
> 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
>> okl = > c
> (0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
> ,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
>> tra = > c
> (0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
> ,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
> 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
> 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1))
>>
>>
>>
>>> z.out <- zelig(food ~ small + han + okl + tra, model =
> "mlogit", data > > = alligator)
>>
>> --
>> Alejandro Daniel Buren
>> Graduate Student
>> Memorial University of Newfoundland
>> St. John's, Newfoundland A1B 3X9 Canada
>>
>> e-mail: aburen@xxxxxx
>
>
>
___________________________________________________________ 1GB
gratis, Antivirus y Antispam Correo Yahoo!, el mejor correo web del
mundo http://correo.yahoo.com.ar
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
8:17 p.m.
New subject: mlogit help
Dan,
Try
coef(z.out)
or
z.out$coefficients
And, make sure, as Olivia suggests, to update Zelig to the most
current version.
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics Fax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
---------------------------------------------------------
On Aug 14, 2005, at 4:18 PM, Olivia Lau wrote:
mlogit, like bprobit and blogit, use slots. Depending
on the
version of Zelig that you have installed, you can use $ to extract
slot elements. If $ doesn't work, either update your version of
Zelig or use @ (the slot extractor).
Best, Olivia
----- Original Message ----- From: "Gary King" <king(a)harvard.edu>
To: "Dan Powers" <dpowers(a)mail.la.utexas.edu>
Cc: "Kosuke Imai" <kimai(a)princeton.edu>du>; <zelig(a)latte.harvard.edu>du>;
<olau(a)fas.harvard.edu>
Sent: Sunday, August 14, 2005 3:29 PM
Subject: RE: [zelig] Re: mlogit help
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
The manual says z.out$coefficients
see http://gking.harvard.edu/zelig/docs/Output_Values18.html
And in general names(z.out) will give you a list of the options.
Gary
---
Gary King,
Institute for Quantitative Social Science
Harvard University, 34 Kirkland St, Cambridge, MA 02138
http://GKing.Harvard.Edu, King(a)Harvard.Edu
Direct 617-495-2027, Assistant 495-9271, eFax 812-8581
On Sun, 14 Aug 2005, Dan Powers wrote:
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Kosuke --
Any hints on obtaining the coefficient vector from the zelig
mlogit model?
Apparently it is not of the form
m1$coef, or variants thereof, as documented.
Thanks,
Dan
%%%%%%%%%%%%%%%%%%%%%%%%
Daniel A. Powers, Ph.D.
Department of Sociology
University of Texas at Austin
1 University Station A1700
Austin, TX 78712-0118
phone: 512-232-6335
fax: 512-471-1748
dpowers(a)mail.la.utexas.edu
-----Original Message-----
From: owner-zelig(a)latte.harvard.edu
[mailto:owner-zelig@latte.harvard.edu]On Behalf Of Kosuke Imai
Sent: Friday, August 12, 2005 1:49 PM
To: Alejandro Buren
Cc: zelig(a)latte.harvard.edu; olau(a)fas.harvard.edu
Subject: Re: [zelig] Re: mlogit help
Hi Alejandro,
We have the documentation at http://gking.harvard.edu/zelig/docs/
_TT_mlogit_TT__Multino.html In the next version, we will add the
fact
that the last level of the outcome variable is used as the base
category in the multinomial logit model. If there is anything else
unclear about the documentation, please let us know since our
goal is
to write a self-contained
As for the likelihood, I would follow Gary's suggestion of doing
something very simple - using a simulated data set where you know
the
truth for example. I would also double-check the log-likelihood
formula you are using: make sure you do all the calculation on the
log-scale - sometimes multiplying and dividing by very small or
large
numbers make the computer calculation incorrect. Doing all the
calculation on the log-scale should help.
Good luck,
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics Fax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
---------------------------------------------------------
On Aug 12, 2005, at 9:02 AM, Alejandro Buren wrote:
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Kosuke,
thanks for the tip, I've run the example, but there is something
I'm still missing.
Running the example, mlogit yields a set of parameters and a
LogLikelihood; when I use
that set of parameters to calculate the LogLikelihood myself (by
hand) I get a completely
different value.
mlogit is not working as I thougt it was, is there any document
which explains in more detail
than "Zelig: Everyone's statistical software" how it's working?
I would really appreciate it, since it's crucial for my research to
be able to fit these
multinomial logit models.
Thanks for your time and consideration,
-----------------------------------------------------
Alejandro Daniel Buren
Graduate Student
Memorial University of Newfoundland
St. John's, Newfoundland A1B 3X9 Canada
-----------------------------------------------------
----- Original Message ----- From: "Kosuke Imai"
<kimai(a)Princeton.Edu>
To: "Alejandro Buren" <aled38(a)yahoo.com.ar>
Cc: <zelig(a)latte.harvard.edu>du>; "Olivia Lau" <>
Sent: Thursday, August 11, 2005 6:44 PM
Subject: Re: [zelig] Re: mlogit help
> Dear Alejandro,
> The *last* level (rather than the first level) is the base
> category. In
> our mexico example (type demo(mexico)), for example, "3" is the
> base
> category.
> As for the difference between the two results, if you find
> another
> estimate that gives a greater value of the log-likelihood, your
> estimate
> is definitely not MLE!! From your mesasge, it seems that VGAM
> results are
> giving a better answer. Just to make sure they are calculating the
> same
> log-likelihood (sometimes, different software drops different
> constants),
> you might want to evaluate the log-likelihood for both estimates
> by hand.
> Good luck,
> Kosuke
>
> -----------------------------------------------------
> Kosuke Imai Office: Corwin Hall 041
> Assistant Professor Phone: 609-258-6601
> Department of Politics eFax: 973-556-1929
> Princeton University Email: kimai(a)Princeton.Edu
> Princeton, NJ 08544-1012 http://imai.princeton.edu
> -----------------------------------------------------
>
> On Thu, 11 Aug 2005, Alejandro Buren wrote:
>
>
>
>> Dear Olivia and Kosuke,
>> thank you very much for your kindness.
>> mlogit is taking the base-line category exactly as I thought
>> it was doing it.
>> But my problem remains, since the output is
>>
>>
>>> z.out <- zelig(food ~ small + han + okl + tra, model =
>>>
>> "mlogit", data = > alligator)
>>
>>> z.out
>>>
>> Call:
>> zelig(formula = food ~ small + han + okl + tra, model = "mlogit",
>> data = alligator)
>>
>> Coefficients:
>> (Intercept):1 (Intercept):2 (Intercept):3 (Intercept):4
>> small:1 small:2 small:3 small:4 han:1
>> han:2 han:3
>> -0.6931 -13.4332 -13.4332 -13.4332
>> -0.2178 0.5949 -1.0014 -1.0256 1.7697
>> 12.0783 12.6985
>> han:4 okl:1 okl:2 okl:3
>> okl:4 tra:1 tra:2 tra:3 tra:4
>> 13.2221 2.5126 14.9878 14.5858
>> 13.3380 1.0514 13.6991 13.5604 12.8731
>>
>> Degrees of Freedom: 636 Total; 616 Residual
>> Residual Deviance: 418.9
>> Log-likelihood: -209.5
>>
>>
>> I assume that (for example) intercept1 refers to the intercept
>> for the
>> first level which is not the base-line category (invertebrate).
>> If that
>> is so,
>> and I plug these estimates in the calculus of the log likelihood
>> yields
>> Log-likelihood: -284. Besides, the estimates are quite different
>> from
>> those reported by Agresti (2002):
>>
>> AGRESTI (2002) mlogit
>> intercept Size Hancock Oklawaha Trafford intercept Size
>> Hancock Oklawaha Trafford
>> Invertebrate -1.55 1.46 -1.66 0.94 1.12 Invertebrate
>> -0.6931 -0.2178 1.7694 2.5126 1.0514
>> Reptile -3.31 -0.35 1.24 2.46 2.94 Reptile -13.433 0.5949
>> 12.0783 14.9878 13.6991
>> Bird -2.09 -0.63 0.7 -0.65 1.09 Bird -13.433 -1.0014
>> 12.6985 14.5858 13.5604
>> Other -1.9 0.33 0.83 0.01 1.52 Other -13.433 -1.0256
>> 13.2221 13.338 12.8731
>>
>>
>>
>> I believe I'm missing something really important here but I don't
>> know what it is, could you please give me a hint?
>>
>> Thanks a lot in advance,
>>
>> Alejandro
>>
>> -----------------------------------------------------------------
>> ---
>> ------------
>>
>>
>> Dear Alejandro,
>>
>> All models that take a factor response (mlogit, oprobit, ologit,
>> etc) as
>> the dependent variable use the first level as the base case.
>>
>> Now you ask: What's the first level? For a numeric vector, the
>> first
>> level is the smallest level. For a vector of character strings,
>> the first
>> level is the first string value in alphabetical order.
>>
>> Best,
>>
>> Olivia
>>
>> On Thu, 11 Aug 2005, Alejandro Buren wrote:
>>
>>
>>> Dear all,
>>> I'm new to Zelig, I found it looking for a way to fit
>>>
>> multinomial logit > models with R.
>>
>>> I still haven't been able to figure out how "mlogit" chooses
>>>
>> the > base-line category.
>>
>>> I tried running an example I'm familiarized with (Alligator
>>>
>> Food Choice > Example, Ex 7.1.2
>>
>>> from Agresti 2002 "Categorical Data Analysis"), but that
>>>
>> didn't help > me to understand how it is working.
>>
>>> Could anyone help me out?
>>> I'm pasting the code to run the example.
>>> Any comment or help will be most appreciated,
>>>
>>>
>>>
>>>> alligator <- data.frame(
>>>>
>>> food = > c
>>>
>> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,3,3,4,4,5,
>> 5,5
>> ,
>> 5,5,5,5,5,1,1,1,1,1,1,1,2,3,4,4,4,5,5,5,5,5,1,1,1,1,1,2,2,2,2,2,2
>> ,2,
>> 2,2,2,2,3,4,5,5,5,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,3,3,3
>> ,3,
>> 3,3,4,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,3,4,5,5,5,5,5,1,1,1,1,1
>> ,1,
>> 1,1,2,2,2,2,2,2,2,3,3,3,3,3,3,4,4,4,5,5,5,5,5),
>>
>>> small = > c
>>>
>> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
>> 1,1
>> ,
>> 1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1
>> ,1,
>> 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>> ,0,
>> 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0
>> ,0,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
>>
>>> han = > c
>>>
>> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
>> 1,1
>> ,
>> 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0
>> ,0,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>> ,0,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>> ,0,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
>>
>>> okl = > c
>>>
>> (0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>> 0,0
>> ,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1
>> ,1,
>> 1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
>> ,1,
>> 1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>> ,0,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
>>
>>> tra = > c
>>>
>> (0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>> 0,0
>> ,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>> ,0,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>> ,0,
>> 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
>> ,1,
>> 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1))
>>
>>>
>>>
>>>
>>>
>>>> z.out <- zelig(food ~ small + han + okl + tra, model =
>>>>
>> "mlogit", data > > = alligator)
>>
>>>
>>> --
>>> Alejandro Daniel Buren
>>> Graduate Student
>>> Memorial University of Newfoundland
>>> St. John's, Newfoundland A1B 3X9 Canada
>>>
>>> e-mail: aburen@xxxxxx
>>>
>>
>>
>>
>>
___________________________________________________________ 1GB
gratis, Antivirus y Antispam Correo Yahoo!, el mejor correo web del
mundo http://correo.yahoo.com.ar
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
15 Aug
15 Aug
11:55 a.m.
New subject: mlogit help
Olivia --
Thanks. I think I had run into the slot/list issue before. Upgrading to
the recent version works.
Cheers,
Dan
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Daniel A. Powers, Ph.D.
Department of Sociology
University of Texas at Austin
1 University Station A1700
Austin, TX 78712-0118
phone: 512-232-6335
fax: 512-471-1748
dpowers(a)mail.la.utexas.edu
On Sun, 14 Aug 2005, Olivia Lau wrote:
mlogit, like bprobit and blogit, use slots. Depending
on the version of
Zelig that you have installed, you can use $ to extract slot elements. If $
doesn't work, either update your version of Zelig or use @ (the slot
extractor).
Best, Olivia
----- Original Message -----
From: "Gary King" <king(a)harvard.edu>
To: "Dan Powers" <dpowers(a)mail.la.utexas.edu>
Cc: "Kosuke Imai" <kimai(a)princeton.edu>du>;
<zelig(a)latte.harvard.edu>du>;
<olau(a)fas.harvard.edu>
Sent: Sunday, August 14, 2005 3:29 PM
Subject: RE: [zelig] Re: mlogit help
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
The manual says z.out$coefficients
see http://gking.harvard.edu/zelig/docs/Output_Values18.html
And in general names(z.out) will give you a list of the options.
Gary
---
Gary King,
Institute for Quantitative Social Science
Harvard University, 34 Kirkland St, Cambridge, MA 02138
http://GKing.Harvard.Edu, King(a)Harvard.Edu
Direct 617-495-2027, Assistant 495-9271, eFax 812-8581
On Sun, 14 Aug 2005, Dan Powers wrote:
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Kosuke --
Any hints on obtaining the coefficient vector from the zelig mlogit
model?
Apparently it is not of the form
m1$coef, or variants thereof, as documented.
Thanks,
Dan
%%%%%%%%%%%%%%%%%%%%%%%%
Daniel A. Powers, Ph.D.
Department of Sociology
University of Texas at Austin
1 University Station A1700
Austin, TX 78712-0118
phone: 512-232-6335
fax: 512-471-1748
dpowers(a)mail.la.utexas.edu
-----Original Message-----
From: owner-zelig(a)latte.harvard.edu
[mailto:owner-zelig@latte.harvard.edu]On Behalf Of Kosuke Imai
Sent: Friday, August 12, 2005 1:49 PM
To: Alejandro Buren
Cc: zelig(a)latte.harvard.edu; olau(a)fas.harvard.edu
Subject: Re: [zelig] Re: mlogit help
Hi Alejandro,
We have the documentation at http://gking.harvard.edu/zelig/docs/
_TT_mlogit_TT__Multino.html In the next version, we will add the fact
that the last level of the outcome variable is used as the base
category in the multinomial logit model. If there is anything else
unclear about the documentation, please let us know since our goal is
to write a self-contained
As for the likelihood, I would follow Gary's suggestion of doing
something very simple - using a simulated data set where you know the
truth for example. I would also double-check the log-likelihood
formula you are using: make sure you do all the calculation on the
log-scale - sometimes multiplying and dividing by very small or large
numbers make the computer calculation incorrect. Doing all the
calculation on the log-scale should help.
Good luck,
Kosuke
---------------------------------------------------------
Kosuke Imai Office: Corwin Hall 041
Assistant Professor Phone: 609-258-6601
Department of Politics Fax: 973-556-1929
Princeton University Email: kimai(a)Princeton.Edu
Princeton, NJ 08544-1012 http://imai.princeton.edu
---------------------------------------------------------
On Aug 12, 2005, at 9:02 AM, Alejandro Buren wrote:
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
Dear Kosuke,
thanks for the tip, I've run the example, but there is something
I'm still missing.
Running the example, mlogit yields a set of parameters and a
LogLikelihood; when I use
that set of parameters to calculate the LogLikelihood myself (by
hand) I get a completely
different value.
mlogit is not working as I thougt it was, is there any document
which explains in more detail
than "Zelig: Everyone's statistical software" how it's working?
I would really appreciate it, since it's crucial for my research to
be able to fit these
multinomial logit models.
Thanks for your time and consideration,
-----------------------------------------------------
Alejandro Daniel Buren
Graduate Student
Memorial University of Newfoundland
St. John's, Newfoundland A1B 3X9 Canada
-----------------------------------------------------
----- Original Message ----- From: "Kosuke Imai" <kimai(a)Princeton.Edu>
To: "Alejandro Buren" <aled38(a)yahoo.com.ar>
Cc: <zelig(a)latte.harvard.edu>du>; "Olivia Lau" <>
Sent: Thursday, August 11, 2005 6:44 PM
Subject: Re: [zelig] Re: mlogit help
> Dear Alejandro,
> The *last* level (rather than the first level) is the base
> category. In
> our mexico example (type demo(mexico)), for example, "3" is the base
> category.
> As for the difference between the two results, if you find another
> estimate that gives a greater value of the log-likelihood, your
> estimate
> is definitely not MLE!! From your mesasge, it seems that VGAM
> results are
> giving a better answer. Just to make sure they are calculating the
> same
> log-likelihood (sometimes, different software drops different
> constants),
> you might want to evaluate the log-likelihood for both estimates
> by hand.
> Good luck,
> Kosuke
>
> -----------------------------------------------------
> Kosuke Imai Office: Corwin Hall 041
> Assistant Professor Phone: 609-258-6601
> Department of Politics eFax: 973-556-1929
> Princeton University Email: kimai(a)Princeton.Edu
> Princeton, NJ 08544-1012 http://imai.princeton.edu
> -----------------------------------------------------
>
> On Thu, 11 Aug 2005, Alejandro Buren wrote:
>
>
>> Dear Olivia and Kosuke,
>> thank you very much for your kindness.
>> mlogit is taking the base-line category exactly as I thought
>> it was doing it.
>> But my problem remains, since the output is
>>
>>> z.out <- zelig(food ~ small + han + okl + tra, model =
>> "mlogit", data = > alligator)
>>> z.out
>> Call:
>> zelig(formula = food ~ small + han + okl + tra, model = "mlogit",
>> data = alligator)
>>
>> Coefficients:
>> (Intercept):1 (Intercept):2 (Intercept):3 (Intercept):4
>> small:1 small:2 small:3 small:4 han:1
>> han:2 han:3
>> -0.6931 -13.4332 -13.4332 -13.4332
>> -0.2178 0.5949 -1.0014 -1.0256 1.7697
>> 12.0783 12.6985
>> han:4 okl:1 okl:2 okl:3
>> okl:4 tra:1 tra:2 tra:3 tra:4
>> 13.2221 2.5126 14.9878 14.5858
>> 13.3380 1.0514 13.6991 13.5604 12.8731
>>
>> Degrees of Freedom: 636 Total; 616 Residual
>> Residual Deviance: 418.9
>> Log-likelihood: -209.5
>>
>>
>> I assume that (for example) intercept1 refers to the intercept
>> for the
>> first level which is not the base-line category (invertebrate).
>> If that
>> is so,
>> and I plug these estimates in the calculus of the log likelihood
>> yields
>> Log-likelihood: -284. Besides, the estimates are quite different
>> from
>> those reported by Agresti (2002):
>>
>> AGRESTI (2002) mlogit
>> intercept Size Hancock Oklawaha Trafford intercept Size
>> Hancock Oklawaha Trafford
>> Invertebrate -1.55 1.46 -1.66 0.94 1.12 Invertebrate
>> -0.6931 -0.2178 1.7694 2.5126 1.0514
>> Reptile -3.31 -0.35 1.24 2.46 2.94 Reptile -13.433 0.5949
>> 12.0783 14.9878 13.6991
>> Bird -2.09 -0.63 0.7 -0.65 1.09 Bird -13.433 -1.0014
>> 12.6985 14.5858 13.5604
>> Other -1.9 0.33 0.83 0.01 1.52 Other -13.433 -1.0256
>> 13.2221 13.338 12.8731
>>
>>
>>
>> I believe I'm missing something really important here but I don't
>> know what it is, could you please give me a hint?
>>
>> Thanks a lot in advance,
>>
>> Alejandro
>>
>> --------------------------------------------------------------------
>> ------------
>>
>>
>> Dear Alejandro,
>>
>> All models that take a factor response (mlogit, oprobit, ologit,
>> etc) as
>> the dependent variable use the first level as the base case.
>>
>> Now you ask: What's the first level? For a numeric vector, the
>> first
>> level is the smallest level. For a vector of character strings,
>> the first
>> level is the first string value in alphabetical order.
>>
>> Best,
>>
>> Olivia
>>
>> On Thu, 11 Aug 2005, Alejandro Buren wrote:
>>
>>> Dear all,
>>> I'm new to Zelig, I found it looking for a way to fit
>> multinomial logit > models with R.
>>> I still haven't been able to figure out how "mlogit" chooses
>> the > base-line category.
>>> I tried running an example I'm familiarized with (Alligator
>> Food Choice > Example, Ex 7.1.2
>>> from Agresti 2002 "Categorical Data Analysis"), but that
>> didn't help > me to understand how it is working.
>>> Could anyone help me out?
>>> I'm pasting the code to run the example.
>>> Any comment or help will be most appreciated,
>>>
>>>
>>>> alligator <- data.frame(
>>> food = > c
>> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,3,3,4,4,5,5,5
>> ,
>> 5,5,5,5,5,1,1,1,1,1,1,1,2,3,4,4,4,5,5,5,5,5,1,1,1,1,1,2,2,2,2,2,2,2,
>> 2,2,2,2,3,4,5,5,5,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,3,3,3,3,
>> 3,3,4,5,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,3,3,4,5,5,5,5,5,1,1,1,1,1,1,
>> 1,1,2,2,2,2,2,2,2,3,3,3,3,3,3,4,4,4,5,5,5,5,5),
>>> small = > c
>> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
>> ,
>> 1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,
>> 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>> 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,0,0,0,0,0,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
>>> han = > c
>> (1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1
>> ,
>> 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
>>> okl = > c
>> (0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>> ,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,
>> 1,1,1,1,1,1,1,1,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
>> 1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0),
>>> tra = > c
>> (0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0
>> ,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>> 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
>> 0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,
>> 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1))
>>>
>>>
>>>
>>>> z.out <- zelig(food ~ small + han + okl + tra, model =
>> "mlogit", data > > = alligator)
>>>
>>> --
>>> Alejandro Daniel Buren
>>> Graduate Student
>>> Memorial University of Newfoundland
>>> St. John's, Newfoundland A1B 3X9 Canada
>>>
>>> e-mail: aburen@xxxxxx
>>
>>
>>
___________________________________________________________ 1GB
gratis, Antivirus y Antispam Correo Yahoo!, el mejor correo web del
mundo http://correo.yahoo.com.ar
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
-
Zelig Mailing List, served by Harvard-MIT Data Center
Send messages: zelig(a)latte.harvard.edu
[un]subscribe Options: http://lists.hmdc.harvard.edu/?info=zelig
Zelig program information: http://gking.harvard.edu/zelig/
6852
days inactive
6856
days old
10 comments
7 participants
participants (7)
-
Alejandro Buren -
Dan Powers -
Daniel A. Powers
-
Gary King -
Kosuke Imai -
Kosuke Imai
-
Olivia Lau