Re: [R] Interactions in a nls model

Tue, 08 Feb 2011 18:22:44 -0800 

Bert,
Thanks for the input. I was hoping for an easy answer, but as life is, there usually isn't one. I will find a statistician here on campus that might be able to help. Just so you know, the data is remote sensing data that is an average of 9 measurements on one day. However, the data set is taken from 3 different fields over a period of 8 years. I found that a non-linear fit is better than a linear or 2nd order polynomial. However, my conundrum is that I want to tell if the fit using just corn is different from the fit using just soybean. Again, thanks for your help. 

Tony

On 2/8/2011 5:15 PM, Bert Gunter wrote:

Nonlinear models are an entirely different kettle of fish then linear
models. You need to specify exactly how the different crops affect the
parameters in your growth model. I suggest you consult a local
statistician for help (this is not an R question).

Incidentally, depending on the nature of your data -- does it consist
of repeated measures of individual plants, for example? -- you may
need nlme or nlmer (or other special packages) to fit the data and do
inference -- nls makes independence assumptions that aren't
appropriate for such data. Your local statistician can help here, too,
if needed. But be warned, inference is inherently difficult in such
situations.


-- Bert

On Tue, Feb 8, 2011 at 1:12 PM, Anthony Lawrence Nguy-Robertson
<[email protected]>  wrote:

   

I am interested in testing two similar nls models to determine if the lines
are statistically different when fitted with two different data sets; one
corn, another soybean. I know I can do this in linear models by testing for
interactions. See Introductory Statistics with R by Dallgaard p212-218 for
an example. I have two different data sets I am comparing to lai. ci.re
should have very little difference between corn and soybean, while ci.gr,
there should be a difference. If I use the simplistic form described in
Dallgaard (see example in code below) it does not work correctly. What do I
need to do to test for this interaction? Thank you!

My data is located here: ftp://snrftp.unl.edu/Outgoing/example/

Here is my example code:

###load data###
data<- read.csv("example.csv", header=TRUE)
eq<- function(x,a,b,c) {a+b*exp(x^c)}

##create non-linear models##
nls.ci.re<- nls(lai~eq(ci.re,a,b,c),data=data, start=c(a=1,b=1,c=1))
nls.ci.gr<- nls(lai~eq(ci.gr,a,b,c),data=data, start=c(a=1,b=1,c=1))

#create non-linear models for corn#
nls.ci.re.corn<- nls(lai~eq(ci.re,a,b,c),data=subset(data,crop=="corn"),
start=c(a=1,b=1,c=1))
nls.ci.gr.corn<- nls(lai~eq(ci.gr,a,b,c),data=subset(data,crop=="corn"),
start=c(a=1,b=1,c=1))

#create non-linear models for soybean#
nls.ci.re.soybean<-
nls(lai~eq(ci.re,a,b,c),data=subset(data,crop=="soybean"),
start=c(a=1,b=1,c=1))
nls.ci.gr.soybean<-
nls(lai~eq(ci.gr,a,b,c),data=subset(data,crop=="soybean"),
start=c(a=1,b=1,c=1))

#test interactions according to Introductory Statistics with R by Dalgaard
p. 213#
corn.soybean.interactions.ci.re.a<-
    abs((summary(nls.ci.re.corn)$coeff[1,1]-
    summary(nls.ci.re.soybean)$coeff[1,1])/
    sqrt(summary(nls.ci.re.corn)$coeff[1,2]^2+
    summary(nls.ci.re.soybean)$coeff[1,2]^2))

corn.soybean.interactions.ci.gr.a<-
    abs((summary(nls.ci.gr.corn)$coeff[1,1]-
    summary(nls.ci.gr.soybean)$coeff[1,1])/
    sqrt(summary(nls.ci.gr.corn)$coeff[1,2]^2+
    summary(nls.ci.gr.soybean)$coeff[1,2]^2))

corn.soybean.interactions.ci.re.a.p.value<-
pt(corn.soybean.interactions.ci.re.a,df=summary(nls.ci.re.corn)$df[2],lower.tail=FALSE)
corn.soybean.interactions.ci.gr.a.p.value<-
pt(corn.soybean.interactions.ci.gr.a,df=summary(nls.ci.gr.corn)$df[2],lower.tail=FALSE)

______________________________________________
[email protected] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.


     




   


--
Anthony L. Nguy-Robertson
Doctoral Student
University of Nebraska-Lincoln
School of Natural Resources
Center for Advanced Land Management Information Technologies (CALMIT)
223 Hardin Hall
3310 Holdrege Street
Lincoln, NE  68583
Office: 402-472-2565
Fax: 402-472-2946
Email: [email protected]

______________________________________________
[email protected] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.






















					Reply via email to