[R] Offset in glm poisson using R vs Exposure in Stata

Ben Bolker bbolker at gmail.com
Tue Nov 16 14:16:20 CET 2010


Columbine Caroline Waring <caquilegia <at> hotmail.com> writes:

> I am hoping to find someone who uses both R and program Stata for GLMs.
[snip]
> What I have is the code from Stata  and am trying to reproduce the same
analysis in R - my program of choice.
> 
> .         glm count md ms rf sg, family(poisson)
> exposure(effort) eform
> 
> I am lost at the point of finding the equivalent code for 'exposure'. 
> 
> Having looked at a few forums and 'googled'. I thought 'offset', used as   
offset=(log(Eff))   or the
> equivalent    +offset(log(Eff))   would produce the desired effect.
> 
> Incidentally my code was:  
  glm(Count~md+ms+rf+sg+offset(Eff),family=poisson,data=DepthHabGen)
> 
> (Making use of glm{stats})
> 

  Based on your discussion above did you mean

glm(count~md+ms+rf+sg+offset(log(Eff)),
   family=poisson,data=DepthHabGen)

 ? is this just a typo ?

according to http://www.stata.com/help.cgi?glm

 exposure(varname)         include ln(varname) in model with coefficient
                                  constrained to 1
 offset(varname)           include varname in model with coefficient
                                  constrained to 1


> However, offset does not seem to be equivalent to 'exposure' in Stata. As
coefficients and log likelhood
> estimates differ.
> 
> So I asked the following questions:
> 
> 1. Do both programs produce the same results without
> 'exposure' i.e. glm models 
> 
> Yes,  log likelihoods and coefficients are the same.
> 
> 2. How about using the unintuitive non logged " offset=Eff" ?
> 
> Coefficients and log likelihoods still differ.

  You're not doing anything obviously wrong.  Are you sure your
"effort" and "Eff" variables are the same, i.e. nothing got mangled
moving to R?

  I don't use Stata, perhaps someone else can try.  Posting a small
reproducible example would be helpful.



More information about the R-help mailing list