Re: st: Grouped Tobit

[Austin Nichols <austinnichols@gmail.com>]

Gordon Hughes <G.A.Hughes@ed.ac.uk>:
With strong assumptions on normality...
. help intreg
after some work coding top and bottom of ranges.

On Thu, Nov 10, 2011 at 4:26 AM, Gordon Hughes <G.A.Hughes@ed.ac.uk> wrote:
> For those who are unfamiliar with the terminology, the term "grouped tobit"
> is used in econometrics to refer to the analysis of responses that are
> nominally cardinal - i.e. money values - but which have been recorded by
> range rather than value: e.g. 1 = 0-5, 2 = 5-10, 3 = 10-20, etc.  The tobit
> reference arises by assuming that the ranges are censored either at the
> bottom or the top end.
>
> To answer the original question: I am not aware of a user-written procedure
> that implements grouped tobit.  If you really want to do that, then you
> should probably use Limdep for the purpose as it has a richer range of
> econometric procedures for models with limited and dependent variables.
>
> You should also consider whether a grouped tobit specification is really an
> improvement on, say, an ordered probit model, which is available in Stata
> via -oprobit- and other routines.  The theory is that grouped tobit is more
> efficient because you can make use of knowledge of the upper and lower
> values of each category/range.  However, in practical terms I am sceptical
> that most datasets with valuation ranges really embed consistent and
> reliable upper and lower values for each range.  In many circumstances it
> may be more reasonable to assume that answers are ordinal rather than
> cardinal.
>
> Regards
>
> Gordon Hughes
> g.a.hughes@ed.ac.uk

*
*   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/