[R] censReg or tobit: testing for assumptions in R?

[Terry Therneau]

> I'm thinking of applying a censored regression model to
> cross-sectional data, using either the tobit (package survival) or the
> censReg function (package censReg). The dependent variable is left and
> right-censored.
  I assume you mean "survreg" in the survival package.  It's a shame
that censored gaussian regression has earned a unique label (tobit) that
makes people think it is something so very different.


M>y hopefully not too silly question is this: I understand that
>heteroskedasticity and nonnormal errors are even more serious problems
>in a censored regression than in an ols-regression. But I'm not sure
>how to test for these assumptions in R? Is there a way to get to the
>residuals of censored regression models (given that corresponding
>functions for lm, such as rstandard, are not applicable)?

Actually it can be less of a problem, since the really large outliers
often turn into censored observations, limiting their leverage. 
For examination, there are good ideas in the article referenced
by ?residuals.survreg.

Terry Therneau