Date:         Thu, 7 Dec 2006 14:39:13 -0700
Reply-To:     Tomas Nilsson <tknilsso@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Tomas Nilsson <tknilsso@GMAIL.COM>
Subject:      Re: Tobit (w/ proc qlim): computing marginal effects
In-Reply-To:  <1165519394.510867.35990@16g2000cwy.googlegroups.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed

Hello Shiling,
Thanks for the response. Yepp, the marginal statement works (funny that
I didn't find this statement when searching for the words "marginal" and
"tobit" in SAS help manual search module). I was really looking for the
marginal effects evaluated at the mean values of the covariates (e.g.
Wooldridge's example pp. 527). I will do it myself then.

All the best,

Tomas





On 12/7/06, shiling99@yahoo.com <shiling99@yahoo.com> wrote:
>
> There is an option <marginal> on OUTPUT statement. Or you can
> calculation it yourself.
>
> Attached below is an example.
>
> For details you may refere Greene's paper "Marginal Effects in the
> Censored Regression Model".
>
> HTH.
>
> data t1;
> do i = 1 to 2000;
>    x=rannor(123900);
>        y=3+3*x+rannor(123900);
>        if y>=5 then y=5;
>        else if y<=0 then y=0;
>        censor=(y=5 or y=0);
>        output;
> end;
> run;
>
> proc sort data=t1; by y;run;
>
> proc freq data=t1;
> table censor;
> run;
> ods output ParameterEstimates=para;
>
> proc qlim data=t1;
>     model y = x;
>     endogenous y ~  CENSORED(lb=0 ub=5);
>         output out=t2  MARGINAL;
> run;
>
> data _null_;
>   set para;
>   call symput(Parameter,compress(put(Estimate,best.)));
>   run;
>
> %put >>&intercept &x &_sigma<<;
>
> data t3;
> set t2;
> Ystar=&intercept + &x*x;
> IF Ystar>=5 then prob = SDF('NORMAL',5,Ystar,&_sigma);
> else if Ystar<=0 then prob = CDF('NORMAL',0,Ystar,&_sigma);
> else prob =
> CDF('NORMAL',5,Ystar,&_sigma)-CDF('NORMAL',0,Ystar,&_sigma);
> if 0<Ystar<5 then marginal=&x*prob;
> else              marginal=&x*(1-prob);
> d=round(Meff_x,0.00001)-round(marginal,0.00001)  ;
> run;
>
> proc print;
> var x y  censor Meff_x  marginal d Ystar prob ;
> run;
>
>
>
> Tomas Nilsson wrote:
> > Hello,
> > I've estimated a Tobit model and about to compute the marginal effects (
> e.g.
> > equation (16.16) in Wooldridge (p. 523). Unless my right eye betrays me
> > however, there is no "easy default option" in SAS to compute the
> marginal
> > effects in the proc qlim statement. Am I correct?
> >
> > Cheers,
> >
> > Tomas
>