Please use this identifier to cite or link to this item:
http://acervodigital.unesp.br/handle/11449/75134
- Title:
- Multivariate models for correlated count data
- Universidade Estadual de Campinas (UNICAMP)
- University of British Columbia
- Carleton University
- Universidade Estadual Paulista (UNESP)
- 0266-4763
- 1360-0532
- In this study, we deal with the problem of overdispersion beyond extra zeros for a collection of counts that can be correlated. Poisson, negative binomial, zero-inflated Poisson and zero-inflated negative binomial distributions have been considered. First, we propose a multivariate count model in which all counts follow the same distribution and are correlated. Then we extend this model in a sense that correlated counts may follow different distributions. To accommodate correlation among counts, we have considered correlated random effects for each individual in the mean structure, thus inducing dependency among common observations to an individual. The method is applied to real data to investigate variation in food resources use in a species of marsupial in a locality of the Brazilian Cerrado biome. © 2013 Copyright Taylor and Francis Group, LLC.
- 18-Apr-2013
- Journal of Applied Statistics, v. 40, n. 7, p. 1586-1596, 2013.
- 1586-1596
- maximum likelihood
- mixed model
- mixture distribution
- multivariate count data
- negative binomial distribution
- overdispersion
- Poisson distribution
- zero-inflated data
- http://dx.doi.org/10.1080/02664763.2013.789098
- Acesso restrito
- outro
- http://repositorio.unesp.br/handle/11449/75134
There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.