dear Dr Nicola,
I understand that the dispersion parameter 'k'
(=mean*mean/(variance-mean))
of the negative binomial = 1/alpha, when dispersion is
constant with no
depedent variable (note: alpha is the coefficient of
the estimated mean
relating dispersion (variance to mean ratio) with mean
in stata). But I want
to model 'k' as a function of the estimated mean 'm'
(i.e. k= k0+k1*m^beta),
where m is the mean corresponding to the age-class
given by a0+a1*age-class.
I would like to test whether k1=0 against the
alternatives k1<0 or k1>0. My
aim is to test whether the dispersion 'k' declines
with increasing
age-class.
I would like to know if I can model the relationship
of 'k' with 'm' using
the 'ztnb' option in stata. If possible, I request the
stata group to help
me out.
with regards
Subramanian Swaminathan
Vector Control Research Centre
(Indian Council of Medical Research)
Indira Nagar
Pondicherry - 605 006
INDIA
----- Original Message -----
From: <[email protected]>
To: <[email protected]>
Cc: <[email protected]>
Sent: Friday, February 22, 2008 11:27 PM
Subject: Re: st: how to use ztnb?
> Which is the question, here? Your knowledge of
-ztnb- seems to exceed
> mine. Your code
> xi:ztnb mf-counts i.age [fweight = freq] if
mf-counts>0, dispersion(mean)
> let you estimate a zero truncated NB mobel where the
dispersion (the
> parameter called dispersion in Stata) is a nonlinear
(exponential)
> function of the mean. As a default, it includes a
test about the
> hypothesis that the dispersion is a nonlinear
(exponential) function of
> the mean vs. the dispersion is a constant (if the
test is insignificant,
> then the dispersion is a constant).
> Nicola
>
> At 02.33 19/02/2008 -0500, Subramanian Swaminathan
wrote:
>>Dear stata listers,
>>
>>I am trying to fit a zero-truncated negative
binomial
>>model to my count
>>data. The data variables are mf-counts, age,
>>frequency. The model that I
>>would like to fit is of the form:
>>xi:ztnb mf-counts i.age [fweight = freq] if
>>mf-counts>0, dispersion(mean)
>>
>>Kindly note that my dispersion parameter is not a
>>constant but is a
>>non-linear function as given below:
>>
>>k = k0+beta*mean**gamma (0<gamma < infinity)
>>
>>where,
>>'k' is the dispersion parameter 'k' ( in stata,
k=log
>>(1/alpha) of the
>>negative bonomial model
>>
>>beta & gamma = coefficients to be estimated
>>
>>In this model I want to test the hypothesis that
>>beta=0 against beta not
>>equal to 0.
>>
>>could anyone help me out, how the stata code should
be
>>written to estimate
>>the parameters of the model?
>>
>>kindly note that i am new stata.
>
>
____________________________________________________________________________________
Be a better friend, newshound, and
know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
