Re: st: Calculating predicted probabilities at a given value of one of the independent variables

[Richard Williams <richardwilliams.ndu@gmail.com>]

At 10:05 PM 10/26/2011, Vitorino, Maria Ana wrote:
> Thanks!! But the results should be the same as in your previous 
> suggestion, right? Or am I missing something?

When using the predict command, Stata will usually, if possible, 
generate predictions for the out of sample cases, i.e. the cases that 
were not used in the original estimation. So, for example, suppose 
you have complete data on all the x variables but 50 cases are 
missing on y. The estimation command will exclude the 50 missing 
cases. But a subsequent predict command will generate predicted 
values for every case, because every case has complete data on the 
x's. By including stuff like -if e(sample)-, you only generate 
predicted values for those cases used by the estimation command, i.e. 
the 50 cases that were missing on y will have missing values on the 
variables computed by predict.

Or, suppose the original estimation command included a qualifier like 
-if male==1-, i.e. you were only analyzing males. A subsequent 
predict command would also compute predicted values for females 
unless you added an -if e(sample)- qualifier.

My original coding also created the possibility that a case that was 
originally excluded from the estimation because male was missing 
would now get included in the predictions because I set xmale=0 for all cases.

Now, there are often situations where you want out of sample 
predictions. e.g. maybe those 50 cases are missing y because for them 
y is something that will happen in the near future and you want to 
predict what the value of y will be. If that is what you want, then 
stick with my original code. But if you only want predicted values 
for the cases used in the estimation, you should go with something 
like my tweaked example.

The two sets of code would produce identical results if there is no 
missing data and you are not restricting the sample selection in any 
way. But if you know you want to limit predictions to in-sample it is 
probably a good idea to just get in the habit of including -if 
e(sample)- in your coding just to be safe.

> Ana
>
>
> On Oct 26, 2011, at 10:53 PM, "Richard Williams" 
> <richardwilliams.ndu@gmail.com> wrote:
>
> > At 06:06 PM 10/26/2011, Maria Ana Vitorino wrote:
> >> Dear Statalist users,
> >>
> >> I would like to calculate some predicted probabilities after fitting a
> >> multinomial logit model (asclogit). But, the twist is that I would
> >> like to calculate these fitted values for a given level of one of the
> >> independent variables. Namely, I would like to set one of the Xs to
> >> zero when computing the predicted probability (but not in the
> >> estimation).
> >> What would be the best way of going about this? Would it make sense to
> >> do it in 2 steps? In the first step use  "predict xb" and in the
> >> second step remove the component of xb that I'm not interested and
> >> calculate the probability using "gen" and the logit formula?
> >> Or is there a better way?
> >
> > I am going to slightly tweak my earlier advice. Assuming you only 
> want in-sample predictions,
> >
> > webuse sysdsn1, clear
> > clonevar xmale = male
> > mlogit insure age i.xmale nonwhite i.site
> > predict p1 p2 p3 if e(sample)
> > replace xmale = 0 if !missing(xmale)
> > predict pp1 pp2 pp3 if e(sample)
> >
> >
> > -------------------------------------------
> > Richard Williams, Notre Dame Dept of Sociology
> > OFFICE: (574)631-6668, (574)631-6463
> > HOME:   (574)289-5227
> > EMAIL:  Richard.A.Williams.5@ND.Edu
> > WWW:    http://www.nd.edu/~rwilliam
> >
> > *
> > *   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/

-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu
WWW:    http://www.nd.edu/~rwilliam

*
*   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/