We are using glm() to models to counts of deaths due to rare causes using a log link and Poisson error distribution, with population as the offset. Approximate confidence intervals for the parameter estimates are easy to calculate using a standard normal deviate, but obviously when the counts of deaths are small (which is why we are using Poisson regression), these intervals are very approximate indeed. Has anyone done any work on calculating more precise likelihood ratio - based confidence intervals for parameter estimates from generalised linear models? These are also known as "profile likelihood confidence intervals", I believe. PROC GENMOD in SAS can calculate them - I am happy to email the relevant page from the SAS online documentation which describes the way in which they are calculated in more detail to anyone who is interested - but we would much rather use R for this work... Regards, Tim Churches Epidemiology and Surveillance Branch NSW Health Department Sydney, Australia -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- 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 Wed, 21 Mar 2001, Tim CHURCHES wrote:> We are using glm() to models to counts of deaths due to rare causesusing a log link and Poisson error distribution, with population as the offset. Approximate confidence intervals for the parameter estimates are easy to calculate using a standard normal deviate, but obviously when the counts of deaths are small (which is why we are using Poisson regression), these intervals are very approximate indeed. That's not so obvious: you may have many small counts and get a good nomral approximation.> Has anyone done any work on calculating more precise likelihood ratio- based confidence intervals for parameter estimates from generalised linear models? These are also known as "profile likelihood confidence intervals", I believe. PROC GENMOD in SAS can calculate them - I am happy to email the relevant page from the SAS online documentation which describes the way in which they are calculated in more detail to anyone who is interested - but we would much rather use R for this work... Look at confint in package MASS, in the VR bundle. It has a method for glm fits. Make sure you use the latest version, as some have had problems with that function on R. BTW, please wrap your lines when posting: it makes the responder's task easier and the structure of the response easier too. -- Brian D. Ripley, ripley at stats.ox.ac.uk Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/ University of Oxford, Tel: +44 1865 272861 (self) 1 South Parks Road, +44 1865 272860 (secr) Oxford OX1 3TG, UK Fax: +44 1865 272595 -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- 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 _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
Tim CHURCHES <TCHUR at doh.health.nsw.gov.au> writes:>We are using glm() to models to counts of deaths due to rare causes using a log >link and Poisson error distribution, with population as the offset. Approximate >confidence intervals for the parameter estimates are easy to calculate using a >standard normal deviate, but obviously when the counts of deaths are small >(which is why we are using Poisson regression), these intervals are very >approximate indeed. > >Has anyone done any work on calculating more precise likelihood ratio - based >confidence intervals for parameter estimates from generalised linear models? >These are also known as "profile likelihood confidence intervals", I believe. >PROC GENMOD in SAS can calculate them - I am happy to email the relevant page >from the SAS online documentation which describes the way in which they are >calculated in more detail to anyone who is interested - but we would much >rather >use R for this work...I think that confint() in the MASS library might fit the bill here. Mark -- Mark Myatt -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- 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 _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._