s-news
[Top] [All Lists]

Re: coefficients of cubic splines

To: Abhilash Sreekumar Nair <abhilash@iitb.ac.in>
Subject: Re: coefficients of cubic splines
From: Prof Brian Ripley <ripley@stats.ox.ac.uk>
Date: Sat, 27 Mar 2004 17:40:08 +0000 (GMT)
Cc: s-news@lists.biostat.wustl.edu
In-reply-to: <2338.10.1.11.18.1080406454.squirrel@pgmail.iitb.ac.in>
Please do read the help page: 4 is the number of degrees of freedom.
Each knot introduces just one parameter.  Although the help page does not 
actually say so AFAICS, these are B-splines, and coefficients of the 
B-spline basis -- see any good book on splines.

I think if you want to use splines, you do need a basic familiarity with 
the underlying theory (as well as the help pages of the functions you are 
using).

On Sat, 27 Mar 2004, Abhilash Sreekumar Nair wrote:

> Dear All,
> 
> I am using S-plus 6.0 for windows, I want to fit a natural cubic spline on
> a dataset. I have used the following command:
> 
> fit.nif4<-lm(y ~ns(x,4)), (where 4 stands for the number of knots, please
> do correct me if I am wrong.)
> 
> My problem is that as I go on increasing the knots the no. of coefficients
> go on increasing, (sample results are presented below):
> 
> lm(formula = y ~ ns(x, 4))
> 
> Coefficients:
> (Intercept)   ns(x, 4)1     ns(x, 4)2   ns(x, 4)3
> -0.08108135   0.08057826   0.07411367   0.1566252
> 
>  ns(x, 4)4
>  0.0940209
> 
> if the knots are specified as 5, there are 5 + 1 Intercept coefficients.
> 
> My Question: How to interpret these coefficients. In a spline fit there
> have to be a different polynomial fit for different knots. therefore how
> to interpret these coefficients.

-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595


<Prev in Thread] Current Thread [Next in Thread>