Dear list members, I want to fit a nonlinear mixed model using the nlme command. My dependent variable takes the form of event counts for different countries over a number of years, and hence I was going to fit a mixed effects negative binomial model. The problem, as far as I can glean from Pinheiro & Bates 2000, is that I need a model that is not normal in the errors. All the models they discuss have linear error structures. Is there a package in the R language that fits a negative binomial mixed effects model? Thank you, Jose Aleman PhD Candidate Politics Department 130 Corwin Hall Princeton, NJ 08544 609.937.0190
This is a little bit tricky (nonlinear, mixed, count data ...) Off the top of my head, without even looking at the documentation, I think your best bet for this problem would be to use the weights statement to allow the variance to be proportional to the mean (and add a normal error term for individuals) -- this would be close to equivalent to the log-Poisson model used by Elston et al. (Parasitology 2001, 122, 563-569, "Analysis of aggregation, a worked example: numbers of ticks on red grouse chicks"), and might do what you want. -- 620B Bartram Hall bolker at zoo.ufl.edu Zoology Department, University of Florida http://www.zoo.ufl.edu/bolker Box 118525 (ph) 352-392-5697 Gainesville, FL 32611-8525 (fax) 352-392-3704
Ben Bolker said the following on 2005-04-12 21:40:> This is a little bit tricky (nonlinear, mixed, count data ...) Off the > top of my head, without even looking at the documentation, I think your > best bet for this problem would be to use the weights statement to allow > the variance to be proportional to the mean (and add a normal error term > for individuals) -- this would be close to equivalent to the log-Poisson > model used by Elston et al. (Parasitology 2001, 122, 563-569, "Analysis > of aggregation, a worked example: numbers of ticks on red grouse > chicks"), and might do what you want.A recent posting http://finzi.psych.upenn.edu/R/Rhelp02a/archive/48429.html suggests that an R function for fitting the negative binomial mixed-effects model actually exists. HTH, Henric
> I *think* (but am not sure) that these guys were actually (politely)>advertising a commercial package that they're developing. But, looking >at >the web page, it seems that this module may be freely available -- >can't >tell at the moment. > Ben The Software for negative binomial mixed models will be free ie free as in you can use it without paying anything. It is built using our proprietary software. The idea is to show how our software is good for building nonlinear statstical models including those with random effects. Turning our stand alone software into somethng that can be called easily from r has been a bit of a steep learning curve for me, but we are making progress. So far we have looked at 3 models. The model in Booth et al. (easy). An overdispersed data set that turned out probably be a zero inflated poisson (faily easy but the negative binomial is only fit to be rejected for the simpler model) and what appears to be a true negative binomial (difficult but doable) and we are discussing the form of the model with the person who wishes to analyze it. A few more data sets would be useful if anyone has an application so that we can ensure the robustness of our software. Dave -- Internal Virus Database is out-of-date. Checked by AVG Anti-Virus.