Hi I am working on control of dog-bite related rabies in India. In this regard, I want to construct a mixed model that could predict the grouping behaviour of free-roaming dogs as solitary ( found singly), in pairs, in triads or in the groups of four or more dogs. As basically it is a count data, and chances of sighting no dog are not accounted, this response variable follows a zero-truncated Poisson distribution. The predictors are a mix of numerical ( resight probability, temperature, humidity and wind velocity of the day of the survey) and categorical ( gender, age, body condition score and if sighted in the proximity of garbage dumps) variables. The data was collected by the survey of free-roaming dogs over 7 survey occasions in the manner of capture-recapture data ( only here it was sight-resight). As many individuals were sighted more than once during the survey, and their measures are repeated, mixed models with random effect were thought to be the way to account for the clustering. I modelled the data on the glmmTMB package (the intercepts, however, did not differ much when the model was constructed using VGAM -vglm function). I seek to resolve some queries I have in this regard: 1. Is the glmmTMB package appropriate to model this kind of data? 2. How to test goodness of fit of the model? Help is greatly appreciated thanks Harish
Help with glmmTMB mixed models
2 messages · Harish Tiwari, Ben Bolker
I think I'd recommend an ordinal response (e.g. using the clmm function from the ordinal package). Other than being a positive integer-valued values, I don't think group size really matches the mechanism of a truncated Poisson very well. I'm not sure how to test goodness-of-fit for CLMM. If you use clmm2 instead of clmm, you'll be able to get predicted values from the model, which you examine to get an intuitive idea of how well the model is fitting.
On 2018-06-04 08:42 PM, Harish Tiwari wrote:
Hi I am working on control of dog-bite related rabies in India. In this regard, I want to construct a mixed model that could predict the grouping behaviour of free-roaming dogs as solitary ( found singly), in pairs, in triads or in the groups of four or more dogs. As basically it is a count data, and chances of sighting no dog are not accounted, this response variable follows a zero-truncated Poisson distribution. The predictors are a mix of numerical ( resight probability, temperature, humidity and wind velocity of the day of the survey) and categorical ( gender, age, body condition score and if sighted in the proximity of garbage dumps) variables. The data was collected by the survey of free-roaming dogs over 7 survey occasions in the manner of capture-recapture data ( only here it was sight-resight). As many individuals were sighted more than once during the survey, and their measures are repeated, mixed models with random effect were thought to be the way to account for the clustering. I modelled the data on the glmmTMB package (the intercepts, however, did not differ much when the model was constructed using VGAM -vglm function). I seek to resolve some queries I have in this regard: 1. Is the glmmTMB package appropriate to model this kind of data? 2. How to test goodness of fit of the model? Help is greatly appreciated thanks Harish [[alternative HTML version deleted]]
_______________________________________________ R-sig-mixed-models at r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models