Re: A question or two re: splines

Thread View: sci.math.stat

2 messages

[T] Back to Threads [V] Tree View [G] Group Articles

2 total messages Started by J E H Shaw Tue, 06 Oct 1992 16:46

#644

Author: J E H Shaw
Date: Tue, 06 Oct 1992 16:46

19 lines
1410 bytes

Sorry - my e-mailed reply bounced, so I'm posting this (reply to
how to fit splines with a fixed number of free knots).

  I looked (briefly!) at this a long time ago.  With free knots,
  the equations are horribly ill-conditioned.  I think your best
  chance of fitting a spline in practice is iterative ad-hockery:
  fix the knots & fit the spline, find the `least important'
  knot (e.g. where the n'th deriv. changes least or the knot whose
  omission changes the likelihood least) & try it somewhere else
  (e.g. in the middle of the section with largest n'th deriv.)
  and repeat.  Or something like that.  Good luck.
                -- Ewart Shaw

I'll try to dig up whatever references I found on the problem.
-- 
J.E.H.Shaw,  Department of Statistics,  |  JANET:  strgh@uk.ac.warwick.cu
             University of Warwick,     |  BITNET: strgh%uk.ac.warwick.cu@UKACRL
             Coventry  CV4 7AL,  U.K.   |  PHONE:  +44 203 523069
Sarah (20 months) on first seeing juggling:  `Balls skipping in the sky'.

Re: A question or two re: splines

#454

Author: Robert Strawderman
Date: Wed, 07 Oct 1992 15:59

24 lines
1784 bytes

The way you suggest is the one that I am currently using - I had a feeling that the
procedure I posed would not be well-behaved - you've confirmed my suspicions.

In regard to the "knot-selection" procedure you posed, there are two such references
which come to mind:  Pat Smith's paper (a NASA tech report) on using B-splines in 
regression, and a paper by Kooperberg and Stone which appeared recently in one
of the stat journals (Comp Stat and Data Analysis, I believe).

Thanks for your reply...

-------------------------------------------------------------------------------
*******************************************************************************
\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\

Rob Strawderman				Email:  rob@biostat.med.umich.edu
Department of Biostatistics  		Office:	(313) 936 - 1002
University of Michigan

              "Let no man lack for the thirst of real ale...."

///////////////////////////////////////////////////////////////////////////////
*******************************************************************************
-------------------------------------------------------------------------------

Thread Navigation

This is a paginated view of messages in the thread with full content displayed inline.

Messages are displayed in chronological order, with the original post highlighted in green.

Use pagination controls to navigate through all messages in large threads.

Back to All Threads