On Monday, Aparna wrote:
>
> I have run a random effects probit regression, and I'm trying to
> compute the marginal effects by using the command "mfx compute".The
> regression itself took almost 7-8 hours, but the computation of the
> marginal effects seems to be taking even longer.It' s been running for
> almost a day with no output.Am I doing something wrong?Is there a way of
> speeding up the process?
In the March 2004 update -mfx- had some improvements that resulted in
increased speed in many cases.
When there are many variables in model, then obtaining the standard
errors is going to be time consuming (because derivatives of the
marginal effect with respect to every coefficient in the model must be
calculated for the delta method). In the update the -tracelvl()- option
was added. This option specifies that the results be displayed as they
are computed.
Another useful option that as added in the update is the -varlist()-
which allows the user to specify which marginal effects are to be
calculated. Thus, if you are interested in the marginal effects for only
some variables, you now don't have to wait while it calculates all of
them.
One comment on using -mfx- after -xtprobit, re-: the default prediction
after -xtprobit, re- is the linear predictor xb, so it is important
to use the -predict- option on -mfx- to specify the prediction function
you would like -mfx- to differentiate (eg. pu0). Otherwise you may wait
for a long time only to discover that b is the derivative of xb and its
standard error is just the standard error of b. In other words, you will
have exactly the same output as you obtained from your regression.
There is an FAQ on what you can do to help -mfx- do its job:
http://www.stata.com/support/faqs/stat/mfx_speed.html
--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/
