# RE: [R] nls segmented model with unknown joint points

From: <Bill.Venables_at_csiro.au>
Date: Sun 17 Apr 2005 - 13:36:32 EST
```y ~ b41*(x - 1) + b42*(x^2 - 1) +
ifelse((a41 - x) >= 0, b43*(a41 - x)^2, 0) +
ifelse((a42 - x) >= 0, b44*(a42 - x)^2, 0)

```

This is a direct translation from your funny foreign-looking code below that probably makes it clear what's going on. A more swish R form might be

y ~ b41*(x - 1) + b42*(x^2 - 1) +

b43*pmax(a41 - x, 0)^2 + b44*pmax(a42 - x, 0)^2

You mention nlm, too. Here you would use a function rather than a formula, but the idea is the same.

V.

Hello,

I am interested in fitting a segmented model with unknown joint points in nls and perhaps eventually in nlme. I can fit this model in sas (see below, joint points to be estimated are a41 and a41), but am unsure how to specify this in the nlm function. I would really appreciate any suggestions or example code. Thanks a lot. -andy

proc nlin data=Stems.Trees;
params b41=-3 b42=1.5 b43=-1.5 b44=50 a41=0.75 a42=0.1;

term1 = (b41*(x - 1) + b42*(x**2 -1));

if (a41 - x) >= 0 then

term2 = (b43*(a41 - x)**2);
else

term2 = 0;

if (a42 - x) >=0 then

term3 = (b44*(a42 - x)**2);
else

term3 = 0;

model y = term1+term2+term3;
run;

R-help@stat.math.ethz.ch mailing list