I've downloaded R for windows (9.0.1) and it is great! I've
converted all my lecture notes for my GLM course to run on R (they are
available on my web page below). I must admit I particularly like the
default contrast options, which are identical to GLIM. Also I like the
gl function - very useful! I have a couple of questions/bugs:
1. predict.glm doesn't work, but predict.lm does - actually all I am
after is the equivalent of the GLIM
$extract %vl
command, which I want to get from
vl<-predict.glm(glm,se.fit=T)$se.fit^2
A bit clumsy - is there a better way? I'm using it to calculate
'deleted' residuals:
resid(glm)/(1-vl/summary(glm)$dispersion)
2. What are you actually using as an estimator of the dispersion
parameter? Is it dev/df, or is it sum((Y-mu)^2/V(mu))/df?
3. It would be very nice to be able to fix the dispersion parameter in
advance, before using summary(glm). You might know that the data is
exponential, in which case you want to fix the gamma dispersion
parameter to 1, or you might have a previous value from another
analysis, or you might want to cope with extra-binomial/poisson
variation by using dev/df. Is there an easy way of doing this? In GLIM,
you just say $cal %sc= ...
4. I don't think AIC is stricly correct. My understanding of AIC is that
it is the log likelihood maximised over all the parameters, including
the dispersion parameter. Now the problem is that the mle of the
dispersion parameter for the gamma family (and the exponential family in
general) is really awkward - I think it has to be found by interative
methods, using the derivative of the gamma function - I see from a print
of Gamma that you are using dev/n, but I don't think this is the mle....
5. As a matter of interest, section 9.1 of my notes defines the
poisson-expoential family, with a variance function mu^(3/2), useful for
continuous data with exact zeros. I coded up the family by modifying the
poisson family - seems to work, giving the same values as GLIM, but
again I had to leave AIC blank because it's too awkward to estimate the
dispersion parameter....
Keith Worsley
Department of Mathematics and Statistics
McGill University office: BH 1232
805 ouest, rue Sherbrooke tel: (514)-398-3842
Montreal fax: (514)-398-3899
Quebec e-mail: worsley at math.mcgill.ca
Canada H3A 2K6 web: http://www.math.mcgill.ca/keith
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
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 Mon, 31 Jan 2000, Keith Worsley wrote:> I've downloaded R for windows (9.0.1) and it is great! I've > converted all my lecture notes for my GLM course to run on R (they are > available on my web page below). I must admit I particularly like the > default contrast options, which are identical to GLIM. Also I like the > gl function - very useful! I have a couple of questions/bugs: > > 1. predict.glm doesn't work, but predict.lm does - actually all I am > after is the equivalent of the GLIMMore precisely, predict.glm(,se.fit=T) doesn't work. On line 36, change from type=type to type=ifelse(type=="link","response",type)> > $extract %vl > > command, which I want to get from > > vl<-predict.glm(glm,se.fit=T)$se.fit^2 > > A bit clumsy - is there a better way? I'm using it to calculate > 'deleted' residuals: > > resid(glm)/(1-vl/summary(glm)$dispersion)I think rstudent() does what you want.> 2. What are you actually using as an estimator of the dispersion > parameter? Is it dev/df, or is it sum((Y-mu)^2/V(mu))/df?It's the latter. Look at summary.glm for the code> 3. It would be very nice to be able to fix the dispersion parameter in > advance, before using summary(glm). You might know that the data is > exponential, in which case you want to fix the gamma dispersion > parameter to 1, or you might have a previous value from another > analysis, or you might want to cope with extra-binomial/poisson > variation by using dev/df. Is there an easy way of doing this? In GLIM, > you just say $cal %sc= ...summary.glm() has a dispersion= option. -thomas Thomas Lumley Assistant Professor, Biostatistics University of Washington, Seattle -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- 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 _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._
> 4. I don't think AIC is stricly correct. My understanding of AIC is that > it is the log likelihood maximised over all the parameters, including > the dispersion parameter. Now the problem is that the mle of the > dispersion parameter for the gamma family (and the exponential family in > general) is really awkward - I think it has to be found by interative > methods, using the derivative of the gamma function - I see from a print > of Gamma that you are using dev/n, but I don't think this is the mle....Yes you are right. I chose this only as a reasonable approximation for the gamma and inverse Gaussian distributions. If you want the exact AIC, you can get it from the gnlr function in my gnlm library. Jim -.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.- 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 _._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._