[R] How can I 'predict' from an nls model with a fit specified for separate groups?

[peter dalgaard]

On Apr 20, 2011, at 17:04 , Stuart Rosen wrote:

> Following an example on p 111 in 'Nonlinear Regression with R' by Ritz & Streibig, I have been fitting nls models using square brackets with the grouping variable inside. In their book is this example, in which 'state' is a factor indicating whether a treatment has been used or not:
> 
> > Puromycin.m1 <- nls(rate ~ Vm[state] *
> + conc/(K[state] + conc), data = Puromycin,
> + start = list(K = c(0.1, 0.1),
> + Vm = c(200, 200)))
> 
> What I cannot figure out is how to specify the value of the grouping variable in a 'predict' statement. In my own example, I can only seem to get the predictions for the 1st specified level of the grouping variable. I promise that I have read the documentation, and have tried a number of things, but cannot get the correct predictions.

What's the problem? If I continue your example with

plot(Puromycin$conc, predict(Puromycin.m1), col=(1:2)[Puromycin$state])

I get a black and a red set of points, distinctly different.

It's difficult to say what you did wrong in the "number of things" that you tried, but I suspect you didn't set up your newdata properly:

> myconc <- seq(0,1.2,,11)> lv <- levels(Puromycin$state)
> predict(Puromycin.m1, newdata=data.frame(conc=myconc,state=factor("untreated",levels=lv)))
 [1]   0.0000 114.6850 133.7022 141.5248 145.7897 148.4743 150.3196 151.6661
 [9] 152.6918 153.4993 154.1514
> predict(Puromycin.m1, newdata=data.frame(conc=myconc,state=factor("treated",levels=lv)))
 [1]   0.0000 138.6154 167.8413 180.5289 187.6203 192.1490 195.2916 197.6000
 [9] 199.3674 200.7641 201.8956


-- 
Peter Dalgaard
Center for Statistics, Copenhagen Business School
Solbjerg Plads 3, 2000 Frederiksberg, Denmark
Phone: (+45)38153501
Email: pd.mes at cbs.dk  Priv: PDalgd at gmail.com