Thanks to Nick Cox for additional comments on c(s) and mspline that have
made me see the light. He was absolutely right when he said
> > . -c(s)- has not been removed, but has a new identity
> > as -twoway mspline-.
I got so used to thinking of c(s) as a connect style that I completely
forgot that it did cross medians, until Nick pointed out that
> However, -c(s) since first introduction in
> Stata was based on a preliminary computation
> of cross-medians, as was documented, and there was the same
> -bands()- option to tune its operation.
He is right again. I went back to my copy of the version 2 manual (which
I keep in a glass case just for situations like this) and there it was.
Nick is a walking encyclopedia, but you all knew that.
> If there is a distinction between what -twoway mspline- does
> and what -c(s)- did, then it is pretty subtle.
I agree; they are the same thing. The only subtle difference I could
find (after RTFM) is in the defaults. The c(s) default of 200 bands is
such that in many cases it behaves more as a connect style than a
smoother. The mspline default, max(10, round(10*log10(n)) for n obs,
will generally result in more smoothing.
But back to sersets:
The command that motivated my original post was serset create_cspline,
which appears to do just the spline interpolation part. There's a
separate serset create_xmedians that does the cross medians (with 200
bands by default!), so clearly these two steps are now separated in the
plumbing, which is a good thing. It seems likely that
Stata 8's mspline =uses= serset create_xmedians + serset create_cspline
I believe that serset create_cspline has more general applicability; it
could provide the basis for a general "curvy" connect style, a feature
that would be useful quite apart from the problem of scatterplot
smoothing, connecting things other than cross medians. The tool is
there. If only we could plot sersets.
Germ�n Rodr�guez
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
