John asked: How can I generate accurate predicted probabilities w/ PRCHANGE
if my covariate of interest is "involved" in an interaction term?
The short answer is that you cant use the prchange command because it only
works with one variable at a time. Instead, you could use multiple calls to
prvalue which allows you to change more than one variable.
Using your model: oprobit y dummy intval intxn
To examine the predicted probabilities for dummy=1 then dummy=0 you have to
also change the values of intxn. To do this, you could use:
.prvalue, x(dummy=0 intxn=0) rest(mean) save
.sum intval
.global meanint=r(mean)
.prvalue, x(dummy=1 intxn=$meanint) rest(mean) dif
the first line estimates predicted prob holding dummy at 0, the interval at
its mean, and the interaction at zero. The second line calculates the mean
of intval--which is not the mean of the interaction but rather like you said
[1*(sample mean of interval variable)]. The third line saves it as a macro
variable named meanint. The last line estimates predicted prob holding
dummy at 1 and the interaction at mean of intval. Dif and save compare
predicted probabilities between the saved mode and the current settings.
Hope this helps.
----------------------------------------------------------
David Huffer, Senior Statistician
Court Services and Offender Supervision Agency
633 Indiana Avenue, NW, Suite 844
Washington, DC 20004-2902
Voice: W(202) 220-5701 M(240) 426-3474
Fax: W(202) 220-5316 H(810) 963-2560
Email: [email protected], [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/
