Re: [R] Negative binomial regression for count data,

From: Seyed Reza Jafarzadeh <srjafar_at_yahoo.com>
Date: Sat 12 Mar 2005 - 10:06:59 EST


Dear list,
I would like to know:  

  1. After I have used the R code (http://pscl.stanford.edu/zeroinfl.r) to fit a zero-inflated negative binomial model, what criteria I should follow to compare and select the best model (models with different predictors)?
  2. How can I compare the model I get from question 1 (zero-inflated negative binomial) to other models like glm family models or a logistic regression to see which fits the data better?

Thanks,
Reza  

vito muggeo <vmuggeo@dssm.unipa.it> wrote: Hi,
I do not know the article. Notice that an excess of zeroes can lead to (spurious) overdispersion in data, therefore you should decide whether assuming a zip ( zero excess coming from a "mixture") or a negBin (zero execess due to overdispersion) model. Of course some likelihood based criteria (eg AIC) could be of help for you

However it is possible (at least in principle) to account for both extra zeros and overdispersion as well. S. Jackman has written code to fit zip or zinb regression models http://pscl.stanford.edu/zeroinfl.r You should modify the code if you want to assume different "linear predictors" in the logit (zero vs non/zero) and count part

Hope this helps,
vito

Seyed Reza Jafarzadeh wrote:
> Dear list,
>
> I would like to fit a negative binomial regression model as described in "Byers AL, Allore H, Gill TM, Peduzzi PN., Application of negative binomial modeling for discrete outcomes: a case study in aging research. J Clin Epidemiol. 2003 Jun;56(6):559-64" to my data in which the response is count data. There are also 10 predictors that are count data, and I have also 3 categorical and 2 continious predictors. There is overdispersion in the distribution of the response variable (mean=2.8, variance=28) as well as in predictors, and there are a lot of zero's (zero-inflated).
> The authors of that paper used PROC GENMOD in SAS 8.1. I wonder which of the following packages and tests to use in R to acheive such model for my analysis. Is there any tutorial available?
>
>
> anova.negbin
> Likelihood Ratio Tests for Negative Binomial GLMs
> glm.convert
> Change a Negative Binomial fit to a GLM fit
> glm.nb
> Fit a Negative Binomial Generalized Linear Model
> negative.binomial
> Family function for Negative Binomial GLMs
> rnegbin
> Simulate Negative Binomial Variates
> theta.md
> Estimate theta of the Negative Binomial
> gam.neg.bin
> GAMs with the negative binomial distribution
> dnb2
> Density for negative binomial, used in mmlcr
> theta.mmmod
> Estimate theta of the Negative Binomial by Moments
> NegBinomial
> The Negative Binomial Distribution
> ezinb
> The expected value of the censored zero-inflated negative binomial model
>
>
>
> Thanks,
>
> Reza
>
> [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help@stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
>

-- 
====================================
Vito M.R. Muggeo
Dip.to Sc Statist e Matem `Vianelli'
UniversitÓ di Palermo
viale delle Scienze, edificio 13
90121 Palermo - ITALY
tel: 091 6626240
fax: 091 485726/485612
====================================

	[[alternative HTML version deleted]]

______________________________________________
R-help@stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
Received on Mon Mar 14 09:57:41 2005

This archive was generated by hypermail 2.1.8 : Fri 03 Mar 2006 - 03:30:44 EST