R help - Oct 2009 - When modeling with negbin from the aod package...

Hi,
When modeling with negbin from the aod package, parameters for a given count

y | lambda~Poisson(lambda)

with lambda following a Gamma distribution Gamma(r, theta)

are estimated.
The intercept is called phi.
Some other parameters may be also be estimated from factors in the
data: the estimates returned for all these would be in accordance with
the Value listing in the negbin entry in the aod reference pdf, that
is, they are objects "of formal class "glimML"", that is
they are
maximum likelihood estimates. Would someone tell me if I'm correct?
regards,
shfets

alexander russell <ssv736 <at> gmail.com> writes:
> 
> Hi,
> When modeling with negbin from the aod package, parameters for a given
count
> 
> y | lambda~Poisson(lambda)
> 
> with lambda following a Gamma distribution Gamma(r, theta)
> 
> are estimated.
> The intercept is called phi.
> Some other parameters may be also be estimated from factors in the
> data: the estimates returned for all these would be in accordance with
> the Value listing in the negbin entry in the aod reference pdf, that
> is, they are objects "of formal class "glimML"", that
is they are
> maximum likelihood estimates. Would someone tell me if I'm correct?
> regards,
> shfets

  Sounds about right, except that I'm not sure what you mean
by "the intercept is called phi" -- phi is an overdispersion
parameter (var(y) = mu+phi*mu^2, so the distribution approaches
Poisson as phi -> 0 )

  You can look at help("glim-ML"), or at the innards of the
negbin function (just type negbin at the R prompt) for more
information ...

---------- Forwarded message ----------
From: alexander russell <ssv736@gmail.com>
Date: Tue, Oct 20, 2009 at 4:34 PM
Subject: Re: [R] When modeling with negbin from the aod package...
To: Matthieu Lesnoff <matthieu.lesnoff@gmail.com>


Hello again,

It seems that, though we have a simple estimate of the variance, phi, with
negbin, some models seek to create a formula for the variance.

For example, I think Bolker has modeled  variance in the Lily_sum data set
as

*nlikfun = function(a, c, d) {*

*+ k = c * exp(d * flowers)*

*+ -sum(dnbinom(seedlings, mu = a, size = k, log = TRUE))*

*+ }*

(book, p. 420)

if I'm correct?
Is it fair to say that the 'random' argument for negbin will model
variance
this way only if the independent variable on which the variance depends in
the "right-hand formula" is categorical or is a factor with a few
levels
only?
regards, s
  On Thu, Oct 15, 2009 at 5:53 PM, Matthieu Lesnoff <
matthieu.lesnoff@gmail.com> wrote:
> >that is they are
> > maximum likelihood estimates. Would someone tell me if I'm
correct?
>
> yes, all parameters estimated with negbin (aod) are ML estimates
>
> Regards
>
> Matthieu
>
>
	[[alternative HTML version deleted]]