On Tue, 2 Mar 2004, Adam Jayson Sonfield wrote:
> Dear Statalist:
>
> I am trying to run a Tobit model regression using the svy commands ...
>
> My dependent variable is an expenditures variable; most of the values are zero, but a third or so are in dollars greater than zero.
> A Tobit model seems like the right choice.
>
Are you sure that Tobit is the appropriate technique? Tobit should be
used if your data are censored at 0. In a narrow sense, that is probably
not the case for you (you don't indicate the exact nature of the
expenditure variable, so I cannot be sure).
The following is a quote from Maddala, G.S. 1992. Introduction to
Econometrics, Second Edition. New York: Macmillan (p. 341) on this
question:
Every time we have some zero observations in the sample, it is tempting to
use the Tobit model. However, it is important to understand what the
modelreally says. What we haveis a situation where y*i can, in principle,
take on negative values. However, we do not observe them because of
censoring. Thus the zero values are due to nonobservability. This is not
the case with automobile expenditures, hours worked, or wages. These
variables cannot, in principle, assume negative values. The observed zero
values are due not to censoring, but due to the decisions of individuals.
In this case the appropriate procedure would be to model the decisions
that produce the zero observations rather than use the Tobit model
mechanically.
A recent discussion about this in political science is:
Segelman, Lee and Langche Zeng. 1999. Analyzing Censored and
Sample-Selected Data with Tobit and Heckit Models. Political Analysis
8,2: 167-182.
------------------------------------------------------------------------
Richard J. Stoll, Professor of Political Science and
Associate Dean of Social Sciences * Rice University
Dept. Political Science MS 24 * P.O. Box 1892 * Houston TX 77251-1892
E-mail: [email protected] * Tel: 713-348-3362 * FAX: 713-348-5273
*
* 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/
