RE: st: RE: Income distribution

["Nick Cox" <n.j.cox@durham.ac.uk>]

All the distributions named in these emails have computable CDFs
("analytic" is for theorists or calculus teachers). -qpfit- from SSC
contains Q-Q and P-P plotting routines for several distributions used in
income work. 

I agree with Stas' main point that discrimination between candidate
distributions is tricky if the data arrive coarsely binned. It can be
pretty difficult even with thousands of records! 

Nick 
n.j.cox@durham.ac.uk 

Stas Kolenikov

No, this is the stuff to work with individual records. With grouped
data, I
don't think you'd be able to detect differences from a gamma or a
log-normal
distribution unless you have a couple dozen income categories... which
in my
experience does not happen; you are lucky if you have 10 or more; and I
don't think it is fair to say that income data "typically" come in this
form, many surveys have questions or modules on incomes and
expenditures.

I'd say, "write your own -ml- estimator", but most of these flexible
distributions have only density functions available, while the cdfs do
not
have analytic expressions.

On Wed, Mar 3, 2010 at 3:03 PM, Martin Weiss <martin.weiss1@gmx.de>
wrote:

> findit Singh-Maddala

Sridhar Telidevara

> Does anybody know whether stata has programs to estimate the
> parameters of various distributions (like Singh-Maddala, beta-2,
> lognormal etc.) if the income data are available in the grouped form.
> Unit wise income data are not available.  This is the case for
> household income data, where typically only frequencies within
> specified classes are available. The underlying distribution of the
> continuous data is modeled by the above mentioned parametric models.

*
*   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/