R help - Jul 2008 - sem & testing multiple hypotheses with BIC

I'm coming from the AMOS world and am wondering if there is a simple
way to do multiple hypothesis testing in the manner of BIC analyses in
AMOS using the sem package in R.  I've read the documentation, but
don't see anything in there except for basic BIC scores.  Perhaps
someone has devised a simple way to compare the relative likelihood of
all possible path-fittings within a specified set of paths?

Dear Donald,

My apologies for the slow response: I was out of town and taking a break
from r-help when you posted your message.

The sem() function does not automatically fit alternative models. If you're
considering removing all possible combinations of a subset of paths, then
you should be able to write a program to generate the models, using, e.g.,
the combn() function. If you go this route, you may be able to speed up the
computations by using start values for each model selected from the
estimated coefficients for the largest model. For a model named mod,
summary(mod)$BIC will give you the BIC.

I hope this helps,
 John

--------original message ---------
Donald Braman dbraman at law.gwu.edu 
Tue Jul 15 22:20:44 CEST 2008 
I'm coming from the AMOS world and am wondering if there is a simple
way to do multiple hypothesis testing in the manner of BIC analyses in
AMOS using the sem package in R.  I've read the documentation, but
don't see anything in there except for basic BIC scores.  Perhaps
someone has devised a simple way to compare the relative likelihood of
all possible path-fittings within a specified set of paths?

Not that I know of.  However, remember, the RAM format is just a matrix. 
It's fairly simple to write some code to scan through and make changes to
models.  For example, here's something to delete a specified path.  It
should be fairly simple to run through a set of paths, delete them
piecewise, fit the model, and then get a BIC (or other information
criterion) to create a BIC table.  Then again, those wouldn't be a priori
models, and might violate some of the spirit of using information
criteria...

delete.model.element<-function (delete.text, old.model,
sort.by="path") {

	#determine what sort of model element we'll be evaluating
	col.index<-0
	if(sort.by=="path"){col.index<-1}
	if(sort.by=="variable"){col.index<-1}
	if(sort.by=="coefficient"){col.index<-2}
	if(col.index==0) stop("Cannot delete a ", sort.by)

	#blank vectors to start with
	col1<-vector()
	col2<-vector()
	col3<-vector()

	#purge spaces from delete.text
	delete.text<-strip.white(delete.text)
	
	for (line.num in 1:length(old.model[,1])){
		#purge spaces from the relevant line, just in case this is a path
		old.model[line.num, col.index]<-strip.white(old.model[line.num,
col.index])
		
		#see if this is a line to be deleted
		if(!(delete.text %p.in% old.model[line.num, col.index])){
			col1<-c(col1,old.model[line.num,1])
			col2<-c(col2,old.model[line.num,2])
			col3<-c(col3,old.model[line.num,3])
		}
	}
	
	#now turn the columns into a RAM model
	ram<-cbind(col1, col2, col3)
	class(ram)<-"mod"
	return(ram)

}


Donald Braman-2 wrote:
> 
> I'm coming from the AMOS world and am wondering if there is a simple
> way to do multiple hypothesis testing in the manner of BIC analyses in
> AMOS using the sem package in R.  I've read the documentation, but
> don't see anything in there except for basic BIC scores.  Perhaps
> someone has devised a simple way to compare the relative likelihood of
> all possible path-fittings within a specified set of paths?
> 
> 
> 

-- 
View this message in context:
http://www.nabble.com/sem---testing-multiple-hypotheses-with-BIC-tp18474096p19005157.html
Sent from the R help mailing list archive at Nabble.com.