45208

Normal and Poisson distributions for modelling data of eggs per gram of feces in goats of Anglo-Nubian breed

Laylson da Silva Borges ; Vanessa dos Santos Neri  ; Antônia Leidiana Moreira ; Wanderson Fiares de Carvalho ; José Lindenberg Rocha Sarmento ; Max Brandão de Oliveira ; Tatiana Saraiva Torres ; Luciano Silva Sena
Download
Favorite this paper

Verminosis is one of the main factors that affect livestock production. Fecal eggs count (FEC) allows quantifying the level of endoparasite infection, and is good for diagnostic reference and to guide preventive or immediate actions. Considering the nature of FEC, a more rigorous evaluation of the need to assume a distribution different of normal should be performed (e.g. Poisson distribution), and a better strategy of analysis should be set. The purpose in this study was to compare the use of normal and Poisson distributions for modelling variation of FEC in goats of Anglo-Nubian breed. Values of FEC were used in the analysis, in addition the effects of body condition score (BCS), month of birth (MB), birth season (BS) (dry or rainy), and sire (S) were estimated in goats belonging to the Center of Agrarian Sciences from the Federal University of Piau?. Information of 400 adult animals (males and females) was used. Animals were raised in a semi-intensive system, receiving supplementary (voluminous and concentrate) and in pasture feed, with ad libitum access to water. The software R (version 3.2.1) was used to perform the data analysis. FEC was fit in a generalized linear model with normal distribution with identity link function and Poisson distribution with logarithmic link function. Akaike and Bayesian criteria (AIC and BIC), and residual analyses were used for selection of the best fit. In both situations, AIC and BIC of the model with normal distribution were lower. In residual analysis, normal distribution had an average on order of 10-2, while Poisson had average -7.65. On the other hand, the standard deviation of Poisson distribution was lower than that obtained for normal distribution. All coefficients were significant in fit of Poisson, while the intercept and the sire effect were not significant for normal distribution. The normal model can be described as FEC = - 926.29*BCS + 164.46*MB - 1802.13*BS, while the Poisson model is FEC = 7.47 - 0.62*BCS + 0.08*MB + 1.16*BS - 0.05*S. Both models showed being promising, however, normal model was superior, so that it was selected. It is noteworthy, even going against the nature of the data, the modelling using normal distribution best fit the data used in the present study. Is important emphasizing that an ascertained analysis should also be performed using other distributions for each fit.