Even Bergseng <[email protected]>:
I think not. Predicting at some X=x is a way of
characterizing/interpreting the model's estimated coefficients;
selecting on the explanatory variables X in creating a subsample to
predict over (or to compute means of X at which to predict) is
unproblematic, and relatively easy to interpret for a simple selection
rule, e.g. X=(female, college). Selecting on y seems like a bad idea
for all kinds of reasons.
On Dec 20, 2007 10:31 AM, Even Bergseng <[email protected]> wrote:
> I want to use -mfx- to compute elasiticities after -tobit/xttobit-. My interpretation is that the commands
>
> . mfx compute, eyex predict(ystar(0,.))
> . mfx compute, eyex predict(e(0,.))
>
> should produce respectively unconditional (all observations) and conditional (non-limit observations) elasticities. Am I correct?
>
> If so: Both elasticities are interpreted at the mean of all observations. Would it be logical to use mean values for the non-limit sample to produce the conditional elasticities? (i.e. add the option at(avg) to mfx where avg is means for the non-limit sample).
>
*
* 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/
