17 Apr
2008
17 Apr
'08
2:44 p.m.
Hello classmates,
Despite my best efforts, I still have the same two questions I had
last night in this morning. I am listing them again in and
demonstrating what I tried to do to see if anyone has any other ideas.
First, for the ordered probit model, , zelig is giving me this error
message:
I thought it didn't know the variables were ordered, so I tried
imposing as.ordered() to the variables, but this didn't help very
much. I'm out of ideas on this one because I'm really not sure what
it means for the design to be rank-deficient.
Second, I am having trouble finding the bug in my negative binomial
log likelihood function:
ll.negbin <- function(par, X, Y){
theta <- X%*%par[1:ncol(X)]
gamma <- par[(ncol(X)+1)]
sigma2 <- exp(gamma) + 1
out <- sum(lgamma((theta/(sigma2 - 1)) + 1) - lgamma(Y+1) - lgamma
(theta/(sigma2 -1)) + Y*log((sigma2-1)/sigma2) - (theta/(sigma2 - 1))
*log(sigma2))
}
That is leading to this:
snctmodel <- zelig(RES ~ IMPORT + COST + TARGET +
COOP +
TARGET*COOP, model="oprobit", data=snct)
Warning message:
In function (formula, data, weights, start, ..., subset, na.action, :
design appears to be rank-deficient, so dropping some coefs
I don't know if it's giving me wrong coefficients because of this,
but it is a problem because it leads to these two errors as well:
#SIMULATIONS
import.low <- setx(snctmodel, fn = list(numeric =
mean, ordered =
median,others = mode), IMPORT=0)
import.high <- setx(snctmodel, fn = list(numeric =
mean, ordered =
median,others = mode), IMPORT=1)
coop.low <- setx(snctmodel, fn = list(numeric =
mean, ordered =
median,others = mode), COOP=1)
coop.high <- setx(snctmodel, fn = list(numeric =
mean, ordered =
median,others = mode), COOP=2)
coopsim <- sim(snctmodel, coop.low, coop.high, 1000)
Error in x[, -1] %*%
t(sim.coef) : non-conformable arguments
importsim <- sim(snctmodel, import.low,
import.high, 1000)
Error in x[, -1] %*% t(sim.coef) : non-conformable arguments
summary(whatif(data = snctmodel, cfact = import.high))
[1] "Preprocessing data
..."
Error in whatif(data = snctmodel, cfact = import.high) :
number of columns of 'cfact' and 'data' are not equal
opt.nb <- optim(c(0,0,0,0,0), ll.negbin, X = X, Y =
Y, method =
"BFGS", control = list(fnscale = -1), hessian = TRUE)
Error in optim(c(0, 0, 0, 0, 0), ll.negbin, X = X, Y = Y, method =
"BFGS", :
initial value in 'vmmin' is not finite
In addition: There were 28 warnings (use warnings() to see them)
#(the warnings are all "1: In fn(par, ...) : value out of range in
'lgamma'")
Apparently I've written my function such that it evaluates to -Inf or
NaN at certain points. I can get numbers sometimes depending on the
par values I get it, but changing the starting values in optim is not
much help.
I would appreciate any advice you could offer.
Thanks,
Keith
17 Apr
17 Apr
3:05 p.m.
Hi Keith,
I am the least expert, but I will venture an answer. For your ordered probit, you may want
to try the following:
snctmodel <- zelig(as.factor(RES) ~ IMPORT + COST + TARGET + COOP + TARGET*COOP,
model="oprobit", data=snct)
I wonder if we should use as.factor for the variables with more than one category.
Afamia
I thought it didn't know the variables were ordered, so I tried
imposing as.ordered() to the variables, but this didn't help very
much. I'm out of ideas on this one because I'm really not sure what
it means for the design to be rank-deficient.
Second, I am having trouble finding the bug in my negative binomial
log likelihood function:
ll.negbin <- function(par, X, Y){
theta <- X%*%par[1:ncol(X)]
gamma <- par[(ncol(X)+1)]
sigma2 <- exp(gamma) + 1
out <- sum(lgamma((theta/(sigma2 - 1)) + 1) - lgamma(Y+1) - lgamma
(theta/(sigma2 -1)) + Y*log((sigma2-1)/sigma2) - (theta/(sigma2 - 1))
*log(sigma2))
}
That is leading to this:
>> Keith Schnakenberg <keith.schnakenberg at
gmail.com> 4/17/2008 2:44 PM >>>
Hello classmates,
Despite my best efforts, I still have the same two questions I had
last night in this morning. I am listing them again in and
demonstrating what I tried to do to see if anyone has any other ideas.
First, for the ordered probit model, , zelig is giving me this error
message:
snctmodel <- zelig(RES ~ IMPORT + COST + TARGET +
COOP +
TARGET*COOP, model="oprobit", data=snct)
Warning message:
In function (formula, data, weights, start, ..., subset, na.action, :
design appears to be rank-deficient, so dropping some coefs
I don't know if it's giving me wrong coefficients because of this,
but it is a problem because it leads to these two errors as well:
#SIMULATIONS
import.low <- setx(snctmodel, fn = list(numeric =
mean, ordered =
median,others = mode), IMPORT=0)
import.high <- setx(snctmodel, fn = list(numeric =
mean, ordered =
median,others = mode), IMPORT=1)
coop.low <- setx(snctmodel, fn = list(numeric =
mean, ordered =
median,others = mode), COOP=1)
coop.high <- setx(snctmodel, fn = list(numeric =
mean, ordered =
median,others = mode), COOP=2)
coopsim <- sim(snctmodel, coop.low, coop.high, 1000)
Error in x[, -1] %*%
t(sim.coef) : non-conformable arguments
importsim <- sim(snctmodel, import.low,
import.high, 1000)
Error in x[, -1] %*% t(sim.coef) : non-conformable arguments
summary(whatif(data = snctmodel, cfact = import.high))
[1] "Preprocessing data
..."
Error in whatif(data = snctmodel, cfact = import.high) :
number of columns of 'cfact' and 'data' are not equal
opt.nb <- optim(c(0,0,0,0,0), ll.negbin, X = X, Y =
Y, method =
"BFGS", control = list(fnscale = -1), hessian = TRUE)
Error in optim(c(0, 0, 0, 0, 0), ll.negbin, X = X, Y = Y, method =
"BFGS", :
initial value in 'vmmin' is not finite
In addition: There were 28 warnings (use warnings() to see them)
#(the warnings are all "1: In fn(par, ...) : value out of range in
'lgamma'")
Apparently I've written my function such that it evaluates to -Inf or
NaN at certain points. I can get numbers sometimes depending on the
par values I get it, but changing the starting values in optim is not
much help.
I would appreciate any advice you could offer.
Thanks,
Keith
_______________________________________________
gov2001-l mailing list
gov2001-l at lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
3:50 p.m.
Thanks Afamia, I appreciate the reply. This does not seem to fix the
problem, but perhaps I can play around with it and figure something out.
On Apr 17, 2008, at 12:05 PM, Afamia Kaddour wrote:
Hi Keith,
I am the least expert, but I will venture an answer. For your
ordered probit, you may want to try the following:
snctmodel <- zelig(as.factor(RES) ~ IMPORT + COST + TARGET + COOP
+ TARGET*COOP, model="oprobit", data=snct)
I wonder if we should use as.factor for the variables with more
than one category.
Afamia
I thought it didn't know the variables were ordered, so I tried
imposing as.ordered() to the variables, but this didn't help very
much. I'm out of ideas on this one because I'm really not sure what
it means for the design to be rank-deficient.
Second, I am having trouble finding the bug in my negative binomial
log likelihood function:
ll.negbin <- function(par, X, Y){
theta <- X%*%par[1:ncol(X)]
gamma <- par[(ncol(X)+1)]
sigma2 <- exp(gamma) + 1
out <- sum(lgamma((theta/(sigma2 - 1)) + 1) - lgamma(Y+1) - lgamma
(theta/(sigma2 -1)) + Y*log((sigma2-1)/sigma2) - (theta/(sigma2 - 1))
*log(sigma2))
}
That is leading to this:
>> Keith Schnakenberg <keith.schnakenberg
at gmail.com> 4/17/2008 2:44
>> PM >>>
Hello classmates,
Despite my best efforts, I still have the same two questions I had
last night in this morning. I am listing them again in and
demonstrating what I tried to do to see if anyone has any other ideas.
First, for the ordered probit model, , zelig is giving me this error
message:
snctmodel <- zelig(RES ~ IMPORT + COST +
TARGET + COOP +
TARGET*COOP, model="oprobit", data=snct)
Warning message:
In function (formula, data, weights, start, ..., subset, na.action, :
design appears to be rank-deficient, so dropping some coefs
I don't know if it's giving me wrong coefficients because of this,
but it is a problem because it leads to these two errors as well:
#SIMULATIONS
import.low <- setx(snctmodel, fn =
list(numeric = mean, ordered =
median,others = mode), IMPORT=0)
import.high <- setx(snctmodel, fn =
list(numeric = mean, ordered =
median,others = mode), IMPORT=1)
coop.low <- setx(snctmodel, fn = list(numeric
= mean, ordered =
median,others = mode), COOP=1)
coop.high <- setx(snctmodel, fn = list(numeric
= mean, ordered =
median,others = mode), COOP=2)
coopsim <- sim(snctmodel, coop.low, coop.high, 1000)
Error in x[, -1] %*%
t(sim.coef) : non-conformable arguments
importsim <- sim(snctmodel, import.low,
import.high, 1000)
Error in x[, -1] %*% t(sim.coef) : non-conformable arguments
summary(whatif(data = snctmodel, cfact = import.high))
[1] "Preprocessing
data ..."
Error in whatif(data = snctmodel, cfact = import.high) :
number of columns of 'cfact' and 'data' are not equal
opt.nb <- optim(c(0,0,0,0,0), ll.negbin, X =
X, Y = Y, method =
"BFGS", control = list(fnscale = -1), hessian = TRUE)
Error in optim(c(0, 0, 0, 0, 0), ll.negbin, X = X, Y = Y, method =
"BFGS", :
initial value in 'vmmin' is not finite
In addition: There were 28 warnings (use warnings() to see them)
#(the warnings are all "1: In fn(par, ...) : value out of range in
'lgamma'")
Apparently I've written my function such that it evaluates to -Inf or
NaN at certain points. I can get numbers sometimes depending on the
par values I get it, but changing the starting values in optim is not
much help.
I would appreciate any advice you could offer.
Thanks,
Keith
_______________________________________________
gov2001-l mailing list
gov2001-l at lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l at lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l 
19 Apr
19 Apr
3:06 p.m.
Keith,
Apologies for the posting if you've already solved your problems.
Problem 1
The section code suggests using these formulations:
z.out <- zelig(as.factor(RES) ~ IMPORT+COVERT+TARGET+COOP
+TARGET:COOP, model="oprobit", data=snct)
z.out1 <- zelig(as.factor(RES) ~ IMPORT+COVERT+TARGET*COOP,
model="oprobit", data=snct)
but I tried yours too:
z.out2 <- zelig(as.factor(RES) ~ IMPORT+COVERT+TARGET+COOP
+TARGET*COOP, model="oprobit", data=snct)
and got the exact same results as the above two - all without your
errors. So maybe something has happened to the dataset as you've been
working on it. Have you tried clearing everything and starting again?
Problem 2
The homework sheet suggests using the Zelig parameterization and
Jenn / Jens are right - it's a bit easier (or at least more intuitive
to me.)
But there seem to be several bugs in your code:
ll.negbin <- function(par, X, Y){
theta <- X%*%par[1:ncol(X)]
I think theta should be exp(X %*% Beta) - this is probably where some
of your values out of range are coming from as R may be trying to
evaluate the log of a negative in the lgamma function. See the gamma
function graph in mathworld: http://mathworld.wolfram.com/
GammaFunction.html
gamma <- par[(ncol(X)+1)]
sigma2 <- exp(gamma) + 1
out <- sum(lgamma((theta/(sigma2 - 1)) + 1)
I think this should be lgamma((theta/(sigma2 - 1)) + Y)
- lgamma(Y+1)
you can drop this term (and you can write simply as factorial(Y))
- lgamma
(theta/(sigma2 -1)) + Y*log((sigma2-1)/sigma2) - (theta/(sigma2 - 1))
*log(sigma2))
}
Jeremy
On 17 Apr 2008, at 20:50, Keith Schnakenberg wrote:
Thanks Afamia, I appreciate the reply. This does not
seem to fix the
problem, but perhaps I can play around with it and figure something
out.
On Apr 17, 2008, at 12:05 PM, Afamia Kaddour wrote:
Dr Jeremy Hodgen
Senior Lecturer in Mathematics Education
King's College London
Department of Education and Professional Studies
Franklin-Wilkins Building
Waterloo Bridge Wing
150 Stamford Street
London SE1 9NH
Tel: 020 7848 3102
Fax: 020 7848 3182
E-mail: jeremy.hodgen at kcl.ac.uk
Hi Keith,
I am the least expert, but I will venture an answer. For your
ordered probit, you may want to try the following:
snctmodel <- zelig(as.factor(RES) ~ IMPORT + COST + TARGET + COOP
+ TARGET*COOP, model="oprobit", data=snct)
I wonder if we should use as.factor for the variables with more
than one category.
Afamia
I thought it didn't know the variables were ordered, so I tried
imposing as.ordered() to the variables, but this didn't help very
much. I'm out of ideas on this one because I'm really not sure what
it means for the design to be rank-deficient.
Second, I am having trouble finding the bug in my negative binomial
log likelihood function:
ll.negbin <- function(par, X, Y){
theta <- X%*%par[1:ncol(X)]
gamma <- par[(ncol(X)+1)]
sigma2 <- exp(gamma) + 1
out <- sum(lgamma((theta/(sigma2 - 1)) + 1) - lgamma(Y+1) - lgamma
(theta/(sigma2 -1)) + Y*log((sigma2-1)/sigma2) - (theta/(sigma2 - 1))
*log(sigma2))
}
That is leading to this:
_______________________________________________
gov2001-l mailing list
gov2001-l at lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l >> Keith Schnakenberg <keith.schnakenberg
at gmail.com> 4/17/2008 2:44
>> PM >>>
Hello classmates,
Despite my best efforts, I still have the same two questions I had
last night in this morning. I am listing them again in and
demonstrating what I tried to do to see if anyone has any other
ideas.
First, for the ordered probit model, , zelig is giving me this error
message:
snctmodel <- zelig(RES ~ IMPORT + COST +
TARGET + COOP +
TARGET*COOP, model="oprobit", data=snct)
Warning message:
In function (formula, data, weights, start, ..., subset,
na.action, :
design appears to be rank-deficient, so dropping some coefs
I don't know if it's giving me wrong coefficients because of this,
but it is a problem because it leads to these two errors as well:
#SIMULATIONS
import.low <- setx(snctmodel, fn =
list(numeric = mean, ordered =
median,others = mode), IMPORT=0)
import.high <- setx(snctmodel, fn =
list(numeric = mean, ordered =
median,others = mode), IMPORT=1)
coop.low <- setx(snctmodel, fn = list(numeric
= mean, ordered =
median,others = mode), COOP=1)
coop.high <- setx(snctmodel, fn = list(numeric
= mean, ordered =
median,others = mode), COOP=2)
coopsim <- sim(snctmodel, coop.low, coop.high, 1000)
Error in x[, -1] %*%
t(sim.coef) : non-conformable arguments
importsim <- sim(snctmodel, import.low,
import.high, 1000)
Error in x[, -1] %*% t(sim.coef) : non-conformable arguments
summary(whatif(data = snctmodel, cfact = import.high))
[1] "Preprocessing
data ..."
Error in whatif(data = snctmodel, cfact = import.high) :
number of columns of 'cfact' and 'data' are not equal
opt.nb <- optim(c(0,0,0,0,0), ll.negbin, X =
X, Y = Y, method =
"BFGS", control = list(fnscale = -1), hessian = TRUE)
Error in optim(c(0, 0, 0, 0, 0), ll.negbin, X = X, Y = Y, method =
"BFGS", :
initial value in 'vmmin' is not finite
In addition: There were 28 warnings (use warnings() to see them)
#(the warnings are all "1: In fn(par, ...) : value out of range in
'lgamma'")
Apparently I've written my function such that it evaluates to -Inf or
NaN at certain points. I can get numbers sometimes depending on the
par values I get it, but changing the starting values in optim is not
much help.
I would appreciate any advice you could offer.
Thanks,
Keith
_______________________________________________
gov2001-l mailing list
gov2001-l at lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
_______________________________________________
gov2001-l mailing list
gov2001-l at lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
5873
days inactive
5875
days old
gov2001@lists.gking.harvard.edu
3 comments
3 participants
participants (3)
-
AKADDOUR＠hsph.harvard.edu -
jeremy.hodgen＠kcl.ac.uk -
keith.schnakenberg＠gmail.com