[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: RE: Re: MVPROBIT
> -----Original Message-----
> From: firstname.lastname@example.org
> [mailto:email@example.com] On Behalf Of
> Joseph Capuno
> Sent: 24 November 2003 01:08
> To: firstname.lastname@example.org
> Subject: st: Re: MVPROBIT
> Hi to all!
> I am using MVPROBIT (from ssc) to estimate a
> trivariate probit regresion of the model Pr(Membership
> status=1, Participation=1, Exposure to info materials=1).
> After running mvprobit, I did "mfx compute" to obtain the
> marginal effects, which stata supplies after some time. I'd
> appreciate any help on the following:
> (1) Is it legit to use "mfx compute" after MVPROBIT?
> (2) When "mfx compute" reports that certain
> explanatory variables have "no effect" under the
> column dy/dx, does it mean that the variables have no
> statistically valid effects on the dependent variables or
> that mfx cannot compute the desired estimates?
> (3) Is there a way I can compute for the marginal
> effects for different combinations of outcomes in the
> dependent variable, say Pr(Membership status=1,
> Participate=1, Exposure to info materials=0) or
> Pr(Membership status=1, Participate=1|Exposure to info
> materials=1). [Am looking for a similar procedure to Stata's
> probit whihc allows "mfx compute, predict(pr01)].
I am co-author of -mvprobit- (with Lorenzo Cappellari, who is not a
Statalist member). To be honest, I have no idea whether -mfx compute-
will produce valid results after a user-written -ml- program like
-mvprobit-. Perhaps StataCorp could clarify.
Be aware that computation of marginal effects using analytical formulae
(rather than numerical derivatives as in -mfx-) is difficult for
multivariate probit models. See the Statalist archive for my recent
posting about this in the context of -biprobit-.
One alternative to calculating marginal effects is to calculate compare
predicted probabilities for different covariate combinations. Using
prediction program -mvppred- after -mvprobit-, you can derive predicted
probabilities of all successes, and of all failures. (These are the
only 2 outcome combinations that were feasible to program, given that
the # outcome variables is not known in advance.) A conditional
probability such as you cite in (3) cannot currently be computed using
Professor Stephen P. Jenkins <email@example.com>
Institute for Social and Economic Research
University of Essex, Colchester CO4 3SQ, U.K.
Tel: +44 1206 873374. Fax: +44 1206 873151.
* For searches and help try: