|Date: ||Fri, 29 Sep 2006 10:23:48 -0400|
|Reply-To: ||Wensui Liu <liuwensui@GMAIL.COM>|
|Sender: ||"SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>|
|From: ||Wensui Liu <liuwensui@GMAIL.COM>|
|Subject: ||Re: PROC TRANSREG question|
|Content-Type: ||text/plain; charset=ISO-8859-1; format=flowed|
As far as I remember, you just have 1 variable, year, on your RHS. So
pspline or bspline really doesn't make too much difference. If I were
you, I will use pspline.
another consideration is to specifiy the degree freedom explicitly.
you can approximate the spline by piecewise linear (which means df =
1) or piecewise constant (which means df = 0).
On 9/29/06, Peter Flom <Flom@ndri.org> wrote:
> >>> "Wensui Liu" <firstname.lastname@example.org> 9/28/2006 4:05 pm >>> wrote
> I think you are getting what you asked for.
> for 11 cases, you want to fit a b-spline with nknot = 3 (which means 4
> basis) and degree = 3 (which is default in SAS). So what do you
> What I would like is something so that I can recreate the cubic
> equations that apply between each pair of knots, and the location of the
> SAS Tech Support suggested that this requires use of bspline instead of
> spline, but the output from that, while more like what I wanted, is
> still not transparent (although I have not yet studied it in detail).
Senior Decision Support Analyst
Cincinnati Children Hospital Medical Center