Re: st: heteroskedasticity correction with xttobit

[Maarten Buis <maartenlbuis@gmail.com>]

--- On Thu, May 17, 2012 at 9:47 AM, Austin Nichols wrote:
>> vce(boot) does not "correct" for heteroskedasticity since
>> xttobit assumes i.i.d. normal errors independent of normal random effects.
>> I.e. you are working with untenable assumptions, in all likelihood.
>> What is the outcome variable and what is the explanatory variable of interest?
>> What kind of censoring is involved?
>> Often there are preferable alternatives to tobit models.

--- On Mon, May 21, 2012 at 12:06 AM, David Youngberg wrote:
> The outcome variable is a constructed variable--patent quality--which
> ranges from 0 to 1000 (lower and upper bounds, respectively). The
> variable of interest is the labor mobility of scientists and engineers
> (proportion who changed jobs in the last year).
>
> The editor of the journal I submitted to specifically requested using
> tobit so I'm inclined to stick with that, if possible.

The bottom line of Austin's comment is that it is not possible;
normality/Gaussianity and homoscedasticity are such important parts of
the Tobit model, that you cannot meaningfully "adjust" for deviations
from these assumptions. Remember that techniques like bootstrap or
robust standard errors only "adjust" the variance covariance matrix,
but they need to assume that the point estimates are correct. In
models like Tobit deviations from Gausianity and homoscedasticity not
only feed into the variance covariance matrix but also into the
point-estimates themselves.

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany


http://www.maartenbuis.nl
--------------------------
*
*   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/