I just want to thank both Nick and Stephen for their comments.
Yes, indeed my data is groubed into finite-limit-intervals with known limit
values. I will try both with -intreg- which seems indeed appealing, and
through adjsuting alternative distribution codes and see how it goes.
Following Nick's suggestion, I took a look in Lindsey's texts and I did
find some helpful directions, thanks.
Dimitris
Nick Cox wrote:
>
> Stephen's forgotten, or is too modest
> to allude to the fact, that he is co-author
> of the programs mentioned. They all allow
> frequency weights and would thus work
> with frequency distributions to the extent
> that it was acceptable to represent each
> bin, class or interval by its midpoint (or some other
> summary point, such as the geometric mean
> of the endpoints). The -intreg- approach
> is, however, clearly appealing in various
> ways.
>
> I think J.K. Lindsey has several texts
> based on the argument that the central problem
> (meaning, really, the central problem)
> in statistics is limited-dependent
> variable regression, including distribution
> fitting as one of several cases. This claim
> will seem either obvious or ridiculous, depending
> on what you do.
>
> Nick
> [email protected]
>
> Stephen P Jenkins
>
> > Nick Cox helpfully suggested that you look at his code for
> > ML estimation
> > of the parameters describing several distributions.
> > I note, however, that Dimitri referred to a "frequency
> > distribution",
> > which suggests that the data are in grouped (a.k.a. banded) form
> > (numbers of obs within intervals the boundaries of which
> > have specified
> > values). Nick's programs assume unit record data ('micro'
> > data; one row
> > for each obs). As far as I know, programs for fitting distributions
> > grouped data aren't available. (Analogous to -intreg-, however, I
> > suppose you could write down the probability of observation
> > within each
> > interval, where the probability is specified using the Johnson
> > distribution, and use -ml- to estimate the parameters.)
> >
> >
> > Stephen
> > -------------------------------------------------------------
> > Professor Stephen P. Jenkins <[email protected]>
> > Institute for Social and Economic Research
> > University of Essex, Colchester CO4 3SQ, U.K.
> > Tel: +44 1206 873374. Fax: +44 1206 873151.
> > http://www.iser.essex.ac.uk
> >
> >
> > > -----Original Message-----
> > > From: [email protected]
> > > [mailto:[email protected]] On Behalf Of
> > > D.Christodoulou
> > > Sent: 04 December 2003 17:54
> > > To: [email protected]
> > > Subject: Re: st: RE: Johnson's Distributions
> > >
> > >
> > > Thanks Nick for your reply,
> > > I will look for distribution-specific code as you suggested
> > > and try and adjust it to my distribution. I already
> > > translated the distribution and the result is a quite awkward
> > > curve (dichotomous on zero with finite limits) and I have
> > > trouble in estimating it. Anyway, I will play around with
> > > alternative distribution codes and see what happens. many
> > > thanks, Dimitris
> > >
> > >
> > > Nick Cox wrote:
> > > >
> > > > Your question is in two parts. I don't know
> > > > the jargon "hard-bound", but I guess you
> > > > mean that the support is an interval with
> > > > finite minimum and maximum. The answer
> > > > to the first part presumably comes from
> > > > a text on probability distributions.
> > > >
> > > > The question underlying the second part is
> > > > is whether code is available. Unless private
> > > > code is revealed, the ML problem can
> > > > be solved more or less easily depending
> > > > on how awkward the likelihood function is.
> > > > But you might benefit from looking at code for other
> > > distributions on
> > > > SSC and replacing the distribution-specific code. See
> > for example
> > > > -betafit-, -gammafit-, -gumbelfit-.
> > > >
> > > > I can't comment on GMM.
> > > >
> > > > Nick
> > > > [email protected]
> > > >
> > > > D.Christodoulou
> > > >
> > > > > Is it possible to translate a hard-bound frequency
> > > distribution into
> > > > > a Johnson-Sb curve and estimate its shape parameters with
> > > > > either GMM or MLE with STATA?
> > > > > Any directions (to maybe other sources) and suggestions are
> > > > > very welcome.
> > > >
> > > > *
> > > > * 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/
> > >
> > > --
> > > ---------------------------------------------
> > > Dimitris Christodoulou
> > > Associate Researcher
> > > School for Business and Regional Development
> > > University of Wales, Bangor
> > > Hen Coleg
> > > LL57 2DG Bangor
> > > UK
> > > e-mail: [email protected]
> > > ---------------------------------------------
> > > *
> > > * 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/
> > >
> >
> > *
> > * 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/
>
> *
> * 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/
--
---------------------------------------------
Dimitris Christodoulou
Associate Researcher
School for Business and Regional Development
University of Wales, Bangor
Hen Coleg
LL57 2DG Bangor
UK
e-mail: [email protected]
---------------------------------------------
*
* 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/
