Re: st: marginal effects for user defined programs

[May Boggess <mboggess@stata.com>]

On Tuesday morning Paul asked about calculating marginal effects after
user written estimation commands.

Paul has made a good start by storing the estimation results and the 
name of his predict command. This is important since -mfx- uses predict:


> program define myreoprob, eclass
> reoprob wagecat inc female, i(id)
> estimates local cmd="myreoprob"
> estimates local predict="mypredict"
> end
I have made one slight change to Paul's program: I have changed the 
single quotes to double quotes.

Now we need to define the program -predict- will call:

  cap program drop mypredict
  program define mypredict
  syntax newvarlist [in] [if], [x]
  tempvar touse xb
  mark `touse'
  _predict double `xb'
  gen double `varlist'= norm([_cut1]_b[_cons]-`xb') if `touse'
  end

With these two programs defined we can use -mfx- after -reoprob-:

 clear
 use http://www.stata-press.com/data/r7/auto
 keep rep mpg for turn
 drop if rep<3
 replace rep=rep-3
 rename turn id
 rename for female
 rename mpg inc
 rename rep wagecat
 myreoprob wagecat inc female, i(id)
 mfx compute

Now, given that taking the derivates and differences is not too tricky
in the instance, I'm going to do that just so I can verify I have done
this correctly.

 clear
 use http://www.stata-press.com/data/r7/auto
 keep rep mpg for turn
 drop if rep<3
 replace rep=rep-3
 rename turn id
 rename for female
 rename mpg inc
 rename rep wagecat
 reoprob wagecat inc female, i(id)

 sum inc
 local minc=r(mean)
 sum female
 local mfemale=r(mean)
 local xb (_b[inc]*`minc'+_b[female]*`mfemale')

 display "  p = " norm([_cut1]_b[_cons]-`xb')
 display "dpdinc: " normd([_cut1]_b[_cons]-`xb')*(-_b[inc])

 local xb1 (_b[inc]*`minc'+_b[female]*1)
 local xb0 (_b[inc]*`minc'+_b[female]*0)
 local p1  norm([_cut1]_b[_cons]-`xb1')
 local p0  norm([_cut1]_b[_cons]-`xb0')
 display "dpdfemale: " `p1'-`p0'

The marginal effects I calculated here "by hand" should agree with those
give by -mfx-. Of course, -mfx- has the advantage of being able to also 
give the standard errors. In version 9, one can use -nlcom- instead of
-display- in my "by hand" method to get the standard errors.

 -- May
mmb@stata.com

*
*   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/