Modelo de dispersão Hiper-Poisson para variáveis discretas observáveis e não observáveis
Resumen
Poisson distribution is widely used to model count data, however it has the disadvantage the
assumption that the data must have equal mean and variance, which is not always true, since
in many situations the phenomenon of overdispersion (variance greater than average) or under-
dispersion (variance lower than average) is common. Thus, we work with the hyper-Poisson
distribution, which may accomodate data with overdispersion or underdispersion. The hyper-
Poisson model is investigated here in two distinct scenarios, first modeling observable random
variables in counting problems, and secondly representing an unobservable (latent) variable used
in survival analysis models. In the first scenario, we take a classic approach for the estimation of
the parameters of the hyper-Poisson distribution and we developed the usual likelihood ratio test,
together with the gradient test to test the model dispersion parameter. In the survival analysis,
we propose a new cure rate model induced by frailty discrete with hyper-Poisson probability
distribution, since it is important to choose a distribution that takes into account the dispersion
of risk factors. For this new model we developed inferential procedures from the classical and
bayesian perspectives. All the models worked were analyzed through simulation studies and
applied to real data sets.
Colecciones
El ítem tiene asociados los siguientes ficheros de licencia: