Can you fit a nbd model as that is an overdispersed Poisson

You may use the glmer.nb() function (from lme4) to fit a Negative Binomial model, which is an alternative to accommodate extra-Poisson variation.

Wesley Bertoli Kelvyn Jones Many thanks for your kind suggestions! I will try them.

This will give you some ideas of comparison

http://www.bristol.ac.uk/cmm/software/mlwin/mlwin-resources.html#discrete

and slides " Modelling count and proportions data: poisson, negative binomial and binomial proportions "

http://www.bristol.ac.uk/cmm/software/support/workshops/materials/drd.html

Kelvyn Jones Many thanks for your generous help! I will study these materials right now.

See also this FAQ from one of the lme4 authors:

http://bbolker.github.io/mixedmodels-misc/glmmFAQ.html#overdispersion

Daniel Lüdecke Many thanks for your kind suggestion. I am reading it.

Lme4 is rather limited by today's standard. I would go straight to Bayesian, with either brms or INLA, both of which can handle overdispersion without problem.

Thanks for your kind suggestion, Shige Shige Song

try using glmer.nb(y~x1+x2+(1/randomvariable),family=quasipoisson,data=data)

Selamawit Serka Moja Thanks for your kind suggestion, and that's what I did finally.

HI Ling-Ying Shuai I had the same problem and at the end I am also using glmer.nb(y~x1+x2+(1/randomvariable),family=quasipoisson,data=data) as suggested by Selamawit. I would like to know how did you justifiy its use. Which references did you cite?

Thanks!

glmer.nb(y~x1+x2+(1/randomvariable),family=quasipoisson,data=data) worked for me. Sissi Lozada-Gobilard did you finally got the justification for this and references? I am also trying to justify its use. Thanks

Dear Tapiwa G. Zimunya at the end I used the function mentioning in the paper just like that without any citation. You can see what I wrote here

https://www.sciencedirect.com/science/article/abs/pii/S1617138120301588

I hope it helps.

There is no inference available for the dispersion parameter theta in "glmer.nb” function. Therefore, I would try “glmer” with the “negative.binomial” family from the “MASS” package and specify the overdispersion parameter. E.g., “glmer(...,family=MASS::negative.binomial(theta=1.75))” (taken from "lme4" manual)

Tharaka S. Priyadarshana Thanks for your advices! Best wishes, Ling-Ying

Ling-Ying Shuai Since you used this 'glmer.nb(y~x1+x2+(1/randomvariable),family=quasipoisson,data=data)' to address the issue of overdispersion in your model, can you share a link to that paper with me if it was published?

thanks.

Ling-Ying Shuai I tried both commands;

1. glmer.nb(y~x1+x2+(1/randomvariable),family=quasipoisson,data=data)

2. glmer.nb(y~x1+x2+(1/randomvariable),data=data, verbose=TRUE)

I had the same output for both except that option 1 runs faster than option 2.

Here is my suggestion：

add specific names for each sample，and list the “sample name” as an extra random variable in your model.

Guoliang Li Dear colleague, many thanks for your constructive advice. I will try it. Best wishes, Ling-Ying

This paper is very helpful:

Generalized linear mixed models: a practical guide for ecology and evolution

B. M. Bolker, M. E. Brooks, C. J. Clark, S. W. Geange, J. R. Poulsen, M. H. H. Stevens, et al.

Trends in ecology & evolution 2009 Vol. 24 Issue 3 Pages 127-135