st: Re: [question on -gammafit- and -pgamma-]

[Nick Cox <njcoxstata@gmail.com>]

It is always helpful to specify a subject for your email.

This question refers to -gammafit- and -pgamma- from SSC, which are
user-written programs.
Statalist etiquette is to explain where user-written programs referred
to come from.

Peter quotes the syntax

by n:gammafit y, level(95)

which -gammafit- supports. However, as is standard with -by:- support,
this syntax is of limited use for _saving_ results, as after the
first, each invocation of -gammafit- stomps over the previous set of
saved results, so although parameter estimates are displayed in turn,
only the last such set is present in memory once all are done.

Peter correctly reports that -pgamma- does not support the -by()-
option (nor does it support the -by:- prefix).

I am not aware that Stata supports doubly subscripted variables.
Perhaps Peter means a Stata or Mata matrix.

To get results for different groups and put them together, Peter has
to arrange that himself. But this code shows that the steps are
simple.

sysuse auto, clear
matrix results = J(2,2,.)
gammafit mpg if foreign
matrix results[1,1] = e(alpha)
matrix results[1,2] = e(beta)
pgamma mpg if foreign, gen(tp1 ep1)
gammafit mpg if !foreign
matrix results[2,1] = e(alpha)
matrix results[2,2] = e(beta)
pgamma mpg if !foreign, gen(tp2 ep2)
matrix rownames results = Foreign Domestic
matrix colnames results = alpha beta
matrix li results
scatter tp1 ep1 || scatter tp2 ep2 , legend(order(1 "Foreign" 2 "Domestic"))

For a group variable with many more than 2 classes, it would be better
to set up a loop over the groups.

The help for -gammafit- includes a link to the help for -maximize-,
which explains how to set initial values.

Nick

On Sat, Mar 5, 2011 at 4:47 PM, Peter Maclean <pmaclean2011@yahoo.com> wrote:

> I am new is Stata modelling.
> I am running the following model:
>
> by n:gammafit y, level(95)
> pgamma y, gen(g_p e_p)
>
> where n is  a group variable.
> I would like to save the estimated parameters by gammafit (alpha,beta,alpha+beta
> and lower and upper limit of both alpha and beta) in a variable 'result(n,j).'
>
>
> Also, I would like pgamma to estimate the probabilities (g_p and e_p) by
> variable n.
> The pgamma module does not accept the 'by' option. In addition, is there a way
> to give starting values for the gammafit module?

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/