| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.41 MB | Adobe PDF |
Autores
Resumo(s)
Para modelar variáveis que apresentam um grande número de zeros existem os modelos com excesso de zeros e os modelos de duas partes, que têm sido amplamente usados numa grande variedade de áreas de estudo. Estes modelos serão aplicados ao estudo, numa população Angolana, da distribuição da Schistosomose, uma parasitose superada em frequência apenas pela malária. Os resultados sugerem que o modelo GLM (Generalized Linear Models) binomial negativa, o modelo binomial negativa com excesso de zeros e o modelo de duas partes com binomial negativa são preferíveis aos modelos com distribuição de Poisson. Os resultados obtidos mostram que não existem diferenças significativas entre os modelos com binomial negativa. Escolheu-se o modelo binomial negativa com excesso de zeros pois existem duas fontes possíveis para os zeros: os indivíduos que não estão infectados (zeros verdadeiros) e os indivíduos que, apesar estarem infectados, não apresentam ovos na urina (zeros falsos). Assim, a estrutura deste modelo é mais concordante com a situação estudada.
The zero inflated and the Hurdle models are widely applied in a large variety of fields of study for modeling variables that exhibit a large number of zeros. These models are here used for studying the number of cases of Schistosomiasis observed in some regions of Angola. The Schistosomiasis is the second most frequent human parasitic disease after the malaria. The results suggest that the Generalized Linear Model Negative Binomial, the Zero Inflated Negative Binomial and the Hurdle Negative Binomial models are preferable to the Poisson models. No significant differences were found between the various negative binomial models. The Zero Inflated Negative Binomial model was chosen because the sample contains two possible sources of zeros: either the individuals did not have the infection or they were already sick, although no eggs were found in the urine. Therefore, the structure of this model is the most similar to the case being studied.
The zero inflated and the Hurdle models are widely applied in a large variety of fields of study for modeling variables that exhibit a large number of zeros. These models are here used for studying the number of cases of Schistosomiasis observed in some regions of Angola. The Schistosomiasis is the second most frequent human parasitic disease after the malaria. The results suggest that the Generalized Linear Model Negative Binomial, the Zero Inflated Negative Binomial and the Hurdle Negative Binomial models are preferable to the Poisson models. No significant differences were found between the various negative binomial models. The Zero Inflated Negative Binomial model was chosen because the sample contains two possible sources of zeros: either the individuals did not have the infection or they were already sick, although no eggs were found in the urine. Therefore, the structure of this model is the most similar to the case being studied.
Descrição
Tese de mestrado em Estatística, apresentada à Universidade de Lisboa, através da Faculdade de Ciências, 2011
Palavras-chave
Modelos com excesso de zeros Modelos de duas partes Binomial negativa Poisson Schistosomose urinária Teses de mestrado - 2011
