Re: MLE of Negative Binomial Model
- From: cksinick@xxxxxxxxx
- Date: Tue, 8 Jan 2008 19:37:59 -0800 (PST)
On Dec 19 2007, 9:01 am, burgera...@xxxxxxxxx wrote:
I'm attempting to write an algorithm (in Matlab) that estimates the
parameters in a negative binomial regression via Newton-Raphson
method.
My parameters are:
- Beta (coefficient vector for independent variables); and
- Theta ( = 1/alpha, where alpha is the dispersion parameter).
As initial values, I use Beta estimates from a Poisson model and a ML
estimate of Theta based on the response data only.
I am running into 2 problems:
1) While the algorithm works for simple datasets it seems quite
unstable for others, drifting away from the correct solution.
2) An output I require is the covariance matrix. I am calculating this
as the negative inverse of the Hessian, but in cases where my
algorithm finds the correct Beta & Theta, my covariance matrix does
not match what I expect (ie, the covariance matrix produced by
LIMDEP).
Does anyone know a more stable algorithm for the NB model or a way to
add stability to the Newton-Raphson approach?
Is my computation of the covariance matrix correct? How does LIMDEP
calculate the covariance matrix?
Thankyou in advance,
Sean
i am attempting to write a program in matlab to estimates the nubmer
of accidents on two lane rural roads.the dependent variable is
accident and the independent variables are carriageway width,road
condition,shoulder width,shoulder condition,land use,gradient number
of curves number of intersections,number of busstops,number of
curves.i don't know much about Mtalab.Could u please help me?
Sini C K
.
- Follow-Ups:
- Re: MLE of Negative Binomial Model
- From: burgerater
- Re: MLE of Negative Binomial Model
- Prev by Date: Re: multiple comparisons
- Next by Date: on Hypergeometric distribution, can total number of balls in urn be zero?
- Previous by thread: Generalized Inverses
- Next by thread: Re: MLE of Negative Binomial Model
- Index(es):
Relevant Pages
|
