On Thursday, Friedrich Huebler wrote:
> I ran a bivariate probit regression and then tried to calculate
> marginal effects with the command -mfx, predict(pmarg1)-. This
> resulted in the following message:
>
> warning: derivative missing; try rescaling variable age
>
Friedrich wne t on to say that rescaling age didn't help his problem,
but after using the -tracelvl()- option, he experimented by rescaling a
different variable, and that did solve his problem.
So, why did -mfx- suggest rescaling age? Well, it is difficult for
-mfx- to tell where the real source of the problem lies. If you are
trying to differentiate dy/d(age) with respect to the coefficient of,
say, weight, is it age that is the problem or weight? So, the error
message is just -mfx-'s estimate of the variable most likely to be
causing the trouble.
However, it is heartening that by using -tracelvl()- Friedrich was able
to find the source of the problem and cure it. This type of detective
work is exactly the reason why the options -tracelvl()- and
-diagnostics()- were added in the March 2004 update.
If Friedrich is still curious about why -mfx- chose age for his
particualr dataset and model, I'd be happy to take a look at it. He can
send it to me through Stata Technical Support at [email protected].
In answer to Friedrich's second question, SE and Intercooled use the
same code, so you will get the same behaviour in both flavors.
--May
[email protected]
*
* 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/
