Matthijs ---
The line that ends
_b[x1*x2]*x1*x2_bar
should probably end
_b[x1x2]*x*x2_bar
and note legal variable names should be used inside brackets...
but probably the model should be
y=a.x1+b.x1^2+c.x2+d.x1*x2+e.x1^2*x2+r
i.e. it's a bit odd for x2 to be interacted with only one of the x1
terms, though it may or may not make a big difference for
predictions...
clear all
sysuse nlsw88
ren wage y
ren tenure x1
ren grade x2
g x1x2=x1*x2
g x1s=x1^2
g x1sx2=x1s*x2
reg y x1 x2 x1x2 x1s
su x2 if e(sample), meanonly
loc m=r(mean)
loc f "_b[_cons]+_b[x1]*x+_b[x1s]*x^2+"
loc f "`f'_b[x2]*`m'+_b[x1x2]*x*`m'"
su x1 if e(sample), meanonly
loc r2=r(max)
loc r1=r(min)
sc y x1,mc(gs12)||function `f', ra(`r1' `r2') clp(l) name(a)
reg y x1 x2 x1x2 x1s x1sx2
su x2 if e(sample), meanonly
loc f "_b[_cons]+_b[x1]*x+_b[x1s]*x^2+"
loc f "`f'_b[x2]*`m'+_b[x1x2]*x*`m'+_b[x1sx2]*x^2*`m'"
su x1 if e(sample), meanonly
loc r2=r(max)
loc r1=r(min)
sc y x1,mc(gs12)||function `f', ra(`r1' `r2') clp(l) name(b)
It's easier to -preserve-, -replace- x2 with its mean, and -predict-
and -graph- btw.
On Wed, Sep 9, 2009 at 7:15 AM, Isabelle Deltour<[email protected]> wrote:
> Hi Matthijs:
>
> Have you tried
> graph export name_of_your_graph.wmf
> and then copy and paste into a word document?
>
> Isabelle
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of Matthijs De
> Zwaan
> Sent: Wednesday, September 09, 2009 12:34 PM
> To: [email protected]
> Subject: st: graphing results
>
> Dear Stata-listers,
> I am trying to graph the results of my regression. I have estimated a
> model with quadratic terms and an interaction, as in y=x1 + x1^2 + x2
> + x1*x2. I am trying to plot my results in a graph of y versus x1.
> Using -adjust- to keep covariates doesn't do what I need, since it
> also sets x1*x2 to the mean of the interaction, rather than x1*(mean
> x2).
>
> I can get what I want by plotting it using the -graph twoway function-
> command. My current code looks like:
> graph twoway scatter y x1 ///
> || function y = _b[_cons] + _b[x1]*x = _b[x1^2]*x^2 + _b[x2]*x2_bar +
> _b[x1*x2]*x1*x2_bar ,
>
> where x2_bar is the mean of x2 (in the estimation sample).
>
> However, when imported in a document, the final graph looks coarse:
> more like a like a step-function than a smooth line, even when using
> 2500 points to graph the function instead of the default 300. Is there
> a way to get a graph that produces the effect of x1 on y controlling
> for covariates, but that looks better than what I have now?
>
> Thanks for helping!
> Matthijs
*
* 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/
