Skip to content

probit etc. for dose-response modeling

8 messages · Thomas Lumley, Johannes Ranke, Brian Ripley

#
Hello all

I have done some fitting of pnorm functions to dose-response data, so I could
calculate EC50 values (dose where the response is 0.5). I used the nlm function
for this, so I did not get any information about the confidence intervals of
the fitted parameters.

What would be a good way to do such a probit fit, or is there a package which I
could use?

Best regards

Johannes Ranke
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
#
On Tue, 27 Aug 2002, Johannes Ranke wrote:

            
Probit models are generalised linear models so you can use glm().

	-thomas

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
#
Hi again

I found that the nonlinear least squares method also works nicely:

	library(nls)
	model <- nls(viability ~ pnorm(-log10(conc),-EC50,slope),data=data, \
		 start=list(EC50=1.8,slope=0.8))

Is there an advantage of using glm, and how would this work in this case?

Johannes

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
#
That fits by least-squares, which is not optimal.  glm fits by
maximum-likelihood.  This can matter: the menarche data set (in MASS)
is one example.
On Wed, 28 Aug 2002, Johannes Ranke wrote:

            

  
    
#
I do not have binomial data, though - gaussian only allows link="identity" - 
is there another way? 

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
#
What do you have?

`dose-response modeling'  normally implies success/fail responses.
On Wed, 28 Aug 2002, Johannes Ranke wrote:

            

  
    
#
the "healthy status" of our cells is around 1, and if they are
inactive, they are around 0. Values > 1 and < 0 do occur.

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
#
Sounds like you need to dichotomize.

More generally, R can only fit a fully-specified statistical model.
Specifying the model is the hard part of applied statistics.
On Wed, 28 Aug 2002, Johannes Ranke wrote: