re: st: Normally distributed error term & testing normality of

[Christopher Baum <kit.baum@bc.edu>]

<>
Justina said

Maybe this was already said, but there is also the command -hettest-

That command, as -help hettest- will show, is deprecated in favor of -estat hettest- which accesses the same suite of tests. But
-help tobit postestimation- does not offer -estat hettest- as an option, for all of the good reasons that have been discussed earlier
in this thread. Unlike standard regression, normality of the error is a maintained hypothesis in the Tobit framework, as its
likelihood combines a probit and a regression. Per earlier postings, it would be much more sensible in many cases to use -glm-
for this sort of model. 

Along those lines, the command -linktest- allows you to consider whether the 'link' explicit or implicit in
the model you estimate is supported by the data. In my own work, I have found that after estimation of -tobit- and -glm- on the same sample
of data with a mass point at zero, -linktest- often clearly rejects the Tobit specification, and does not reject the -glm- form of the model. 
That might be something for Ebru to consider, rather than fruitless enquiries about the appropriateness of tests designed for a linear regression context to the Tobit.

Kit


Kit Baum   |   Boston College Economics & DIW Berlin   |   http://ideas.repec.org/e/pba1.html
                             An Introduction to Stata Programming  |   http://www.stata-press.com/books/isp.html
  An Introduction to Modern Econometrics Using Stata  |   http://www.stata-press.com/books/imeus.html


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