Skip to content

nlme function examples for dose-respone

3 messages · Christof Kluß, Bert Gunter

Christof Kluß
# 
Hi,

I want to fit nonlinear dose-response curves, as "fun(X,a,b,c)", for
each of our 5 trail locations. Our data basis is something like

location plot year dose response

For each location there are 4 plots as repetitions (over 3 years). So
the interactions "location*year" and "location*plot" should be random
effects.

There are some examples in "Mixed-Effects Models in S and S-PLUS"
(Pinheiro and Bates), but I do not see how they can help me for my
model. Of course I can start with something like

mod <- nlme(response ~ fun(dose,a,b,c)
            , fixed = list(a ~ 1, b ~ 1, c ~ 1)
            , random = list(a ~ 1, b ~ 1, c ~ 1)
            , groups = ~location
            , data=dat
            , start= ... )

But that is not what I want. How do you describe that you want one fit
for each of the five locations and that "location*year" and
"location*plot" or something similar are random effects?

Do you have some other examples that fit better to this problem setting?
I welcome any tips.

thx
Christof
Bert Gunter
# 
1. Post on R-sig-mixed-models instead. Much more expertise and relevance there.

2. I would forget about mixed effects and treat the locations as
fixed. With only 5, you don't have enough information  to estimate the
variance component with any precision anyway.

3. Feel free to ignore (2) and defer to the experts at (1).

Cheers,
Bert
On Tue, Sep 25, 2012 at 8:52 AM, Christof Klu? <ckluss at email.uni-kiel.de> wrote:

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Christof Kluß
    

    

      
      #
    

  


  

      
Hi,

thx Bert, I forward the question. I forgot a detail. For each (location,
plot, year) combination there are 6 (dose, response) pairs.

$ plot: Factor w/ 4 levels "1","2","3","4":
$ location : Factor w/ 5 levels "loc1","loc2",..:
$ year: Factor w/ 3 levels "2009","2010",..:
$ dose: num  27.3 32.7 57.2 33 183.1 ...
$ response: num  54.2 64.9 74.3 62 92.2 ...

So I can fit each (location, plot, year) with 6 points and make a bottom
up approach. But I would be glad to have one overall model, like

mod <- nlme(response ~ fun(dose,a,b,c)
             , fixed = list(a ~ 1, b ~ 1, c ~ 1)
             , random = list(a ~ 1, b ~ 1, c ~ 1)
             , groups = ~location
             , data=dat
             , start= ... )

and "location*plot" and "location*year" as random and for each location
one best fitted curve. But unfortunately I did not know how to formulate
this in nlme.

Christof


Am 25-09-2012 18:13, schrieb Bert Gunter: