[R] Incidence Function Model in R help

[David Winsemius]

On Jul 4, 2009, at 6:15 PM, sjkimble wrote:

>
> R Help:
>
> Another question:
>
> My data are:
>
>    x.crd   y.crd           A p
> 1  361762 1034052       3.414 0
> 2  361763 1034071      94.169 0
> 3  370325 1027277     127.642 1
> 4  370416 1027166     127.961 0
> 5  370471 1027148    1804.846 1
> 6  369050 1031312    1790.493 0
> 7  370799 1026557     103.994 0
> 8  370908 1026354     199.103 0
> 9  361562 1034311    2047.637 0
> 10 365437 1022188 1622678.961 0
> 11 347047 1025334      21.169 0
> 12 349186 1024441  408556.801 0
>
> and the code, complete with error message, is:
>
>> amphimedon_compressa<-read.table(file.choose(),header=T)
>> attach(amphimedon_compressa)
>> d<-dist(cbind(x.crd,y.crd))
>> alpha<-1
>> edis<-as.matrix(exp(-alpha*d))
>> edis<-sweep(edis,2,A,"*")
>> S<-rowSums(edis[,p>0])
>> mod<-glm(p~offset(2*log(S))+log(A),family=binomial)
> Error: NA/NaN/Inf in foreign function call (arg 4)
>>
>
> R Help tells me that argument 4 for glm is "weights	 an optional  
> vector of
> ‘prior weights’ to be used in the fitting process. Should be NULL or a
> numeric vector." I am unsure why this optional vector needs to  
> employed and
> how to correct the NA/NaN/Inf error if in fact it is necessary.

It is not the reason for your error; weights do not need to be  
employed.  When I look at the object S, I see about half of them are  
zeroes. The "expectation" of a binomial model is for the LHS of the  
formulat to be either a 1/0 vector or something of the form  
cbind(events, non_events). You have satisfied that expectation with p  
but there are only 2 such cases. It seems unreasonable to my thinking  
to expect that a logistic regression model can deliver sensible output  
from only 2 events. And that holds doubly (or perhaps infinitely?)   
true when you are starting out with half of your covariates equal to  
log(0) = -Inf.

-- 
DW


>
> Details:
> Reference: Incidence Function Model in R, J Oksanen, 2004, available  
> at:
> cc.oulu.fi/~jarioksa/opetus/openmeta/metafit.pdf
> OS: Mac OS 10.5.7
> R version: 2.9.0
> GUI: 1.28 Tiger build 32-bit
> seesionInfo output:
> Error in system.file(package = pkg, lib.loc = lib.loc) :
>  'package' must be of length 1
> In addition: Warning messages:
> 1: In if (pkg == "base") pkgpath <- file.path(.Library, "base") else  
> if
> ((envname <- paste("package:",  :
>  the condition has length > 1 and only the first element will be used
> 2: In if ((envname <- paste("package:", pkg, sep = "")) %in%  
> search()) { :
>  the condition has length > 1 and only the first element will be used
> 3: In if (file.access(file.path(lib, pkg), 5) == 0L) { :
>  the condition has length > 1 and only the first element will be used
>
> Thanks,
> Steve Kimble
> Department of Forestry and Natural Resources
> Purdue University
> West Lafayette, IN 47907

David Winsemius, MD
Heritage Laboratories
West Hartford, CT