O'Reilly logo

The R Book by Michael J. Crawley

Stay ahead with the world's most comprehensive technology and business learning platform.

With Safari, you learn the way you learn best. Get unlimited access to videos, live online training, learning paths, books, tutorials, and more.

Start Free Trial

No credit card required

Piecewise Regression

This kind of regression fits different functions over different ranges of the explanatory variable. For example, it might fit different linear regressions to the left- and right-hand halves of a scatterplot. Two important questions arise in piecewise regression:

  • how many segments to divide the line into;
  • where to position the break points on the x axis.

Suppose we want to do the simplest piecewise regression, using just two linear segments. Where do we break up the x values? A simple, pragmatic view is to divide the x values at the point where the piecewise regression best fits the response variable. Let's take an example using a linear model where the response is the log of a count (the number of species recorded) and the explanatory variable is the log of the size of the area searched for the species:

data<-read.table("c:\\temp\\sasilwood.txt",header=T)
attach(data)
names(data)

[1] "Species" "Area"

A quick scatterplot suggests that the relationship between log(Species) and log (Area) is not linear:

plot(log(Species)~log(Area),pch=16)

The slope appears to be shallower at small scales than at large. The overall regression highlights this at the model-checking stage:

model1<-lm(log(Species)~log(Area))
plot(log(Area),resid(model1))

The residuals are very strongly U-shaped (this plot should look like the sky at night).

If we are to use piecewise regression, then we need to work out how many straight-line segments to use and where to put the breaks. Visual ...

With Safari, you learn the way you learn best. Get unlimited access to videos, live online training, learning paths, books, interactive tutorials, and more.

Start Free Trial

No credit card required