Hello,
I've spent a lot of time during the past month trying to get negative
multinomial regression models for clustered event counts as described in
(Guang Guo. 1996. "Negative Multinomial Regression Models For Clustered
Event Counts." Sociological Methodology 26: 113-132., abstract at
http://depts.washington.edu/socmeth2/4abst96.htm) implemented in R. A
FORTRAN version of the method described in the Guo paper is available at
http://www.stat.unipg.it/stat/statlib/general/negmul.
I've been approaching this problem by modifying the loglik glm.nb function
as such:
## original loglik function in glm.nb
## loglik <- function(n, th, mu, y) {
## sum(lgamma(th + y) - lgamma(th) - lgamma(y + 1) + th *
## log(th) + y * log(mu + (y == 0)) - (th + y) * log(th +
## mu))
## }
loglik <- function(n, th, mu, y, mu.grouped, y.grouped) {
sum( th*log(th))*length(y.grouped) +
sum(y*log(mu+ (y == 0)))+
sum(lgamma(y.grouped+ th) -lgamma(th) -
(y.grouped+th)*log(mu.grouped+ th))
}
where y.grouped and mu.grouped are the sums of the clusters, as in
y.grouped <- tapply(Y,group,sum)
mu.grouped <- tapply(mu,group,sum)
where group represents the level used to identify the clusters.
I've made some other small changes to glm.nb on the calls to theta.ml such
that
##original call to theta.ml
## th <- as.vector(theta.ml(Y, mu, n, limit = control$maxit,
## trace = control$trace > 2))
th <- as.vector(theta.ml(y.grouped, mu.grouped, length(y.grouped), limit
= control$maxit,
trace = control$trace > 2))
At first I thought all I needed was to make some small changes and modify
the loglik function in glm.nb but implementation now appears to be more
complex than just that.
I'm looking for any help/advice in what I should do in order to implement
negative multinomial regression models in R. What more needs to be done
after modifying the loglik function in glm.nb?
Of course, if it has already been done that would be nice too (though I
don't think so).
I have not been able to use the web interface to log on to the mailing lists
(I keep getting an error message) and can only review the archives, so it
would be much appreciated if you would cc any response to my email address
below.
Sincerely,
Maurice Masliah
Senior Researcher
iTRANS Consulting Inc.
100 York Boulevard
Suite 300, Richmond Hill
L4B 1J8, Ontario
CANADA
Tel:905-882-4100 ext.5295
Fax:905-882-1557
e-mail: [EMAIL PROTECTED]
[[alternate HTML version deleted]]
______________________________________________
[EMAIL PROTECTED] mailing list
http://www.stat.math.ethz.ch/mailman/listinfo/r-help
