# Re: st: "mfx, predict at(...)" evaluated at CONDITIONAL means?

 From "Martin Weiss" To statalist@hsphsun2.harvard.edu Subject Re: st: "mfx, predict at(...)" evaluated at CONDITIONAL means? Date Sun, 15 Mar 2009 21:09:10 +0100

```<>

Couple of things:

You cannot use locals in the -mfx- if you define permanent vars in the -egen- statement.

I played around with the -if- qualifier as well, but that restricts the sample of which the mean is taken to the cases where -mpg- equals thirty while you want to adjust for mpg equal to thirty...

Eva`s solution is most elegant, and I think she wins our little contest - hands down.

HTH
Martin

-------- Original-Nachricht --------
> Datum: Sun, 15 Mar 2009 20:57:43 +0100
> Von: Christoph Carl Basten <cc.basten@googlemail.com>
> An: statalist@hsphsun2.harvard.edu
> Betreff: Re: st: "mfx, predict at(...)" evaluated at CONDITIONAL means?

> Hi Martin and Eva,
>
> thank you very much for both your proposals, both of which seem to be
> working well.
>
> As for Eva's: I earlier considered to do the whole probit only with if
> mpg==30, and rejected this because I wanted to use the information
> on all units, but adding the if... only at the mfx stage seems to
> circumvent this problem.
>
> But Martin's suggestion made me wonder why the following approach does
> not also work (it doesn't, I just don't see why):
>
> . sysuse auto, clear
> . egen condmean_price = mean(price) if mpg==30
> . egen condmean_weight = mean(weight) if mpg==30
> . qui probit foreign mpg weight price
> . mfx, at(mpg=30 price=`condmean_price' weight=`condmean_weight')
>
> Stata then tells me: 'weight' found where number expected;  (and I get
> the same error when putting "loc" in front of the two egen commands).
> In Martin's solution, no such problem seemed to exist. Could you tell me
> what did the trick there?
>
> Many thanks and best regards,
> Christoph
>
>
>
>
>
> Eva Poen wrote:
> > Christoph,
> >
> > I haven't had a closer look, but I think there is a very simple
> > solution for your problem: you can simply use an if condition with
> > -mfx-. Here is an example using the auto data:
> >
> > ****
> > sysuse auto
> > probit foreign mpg rep78 weight
> >
> > * marginal effects at conditional means of mpg and weight if rep78==3:
> > mfx compute if rep78==3
> >
> > * check if the last column really contains conditional means:
> > sum mpg weight if e(sample) & rep78==3
> > ****
> >
> > I quickly checked Martin's solution, and it comes to the same results.
> >
> > Hope this helps,
> > Eva
> >
> >
> >
> > 2009/3/15 Christoph Carl Basten <cc.basten@googlemail.com>:
> >
> >> Hi,
> >>
> >> when estimating the marginal effects (using e.g. "mfx") after a Probit
> >> regression, I'd like to set one value (age) to a representative value,
> say
> >> 50 years, and have all other variables at the /conditional/ mean for
> those
> >> aged 50.
> >>
> >> From what I understand, if I specify only   "mfx, predict at(age=50),
> Stata
> >> will set all other variables to the /population/ means, so the only way
> to
> >> have them set at the means conditional on age=50 would be to compute
> all
> >> these conditional means by hand and then specifying them manually in
> the
> >> atlist.
> >>
> >> But is there any option to tell Stata that I want age=50 and e.th. else
> at
> >> the conditional means?
> >>
> >> Thanks for your consideration and best regards,
> >> Christoph
> >>
> >>
> >>
> >
> > *
> > *   For searches and help try:
> > *   http://www.stata.com/help.cgi?search
> > *   http://www.stata.com/support/statalist/faq
> > *   http://www.ats.ucla.edu/stat/stata/
> >
> >
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/

--
Martin Weiss
Hackersteigle 3
72076 Tübingen
00497071/793535
00491784597218

Psssst! Schon vom neuen GMX MultiMessenger gehört? Der kann`s mit allen: http://www.gmx.net/de/go/multimessenger01
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```