Meta-análise sobre o consumo de matéria seca de ovinos criados no Brasil : estimação e comparação de modelos preditivos

Abstract

In this work statistical models were developed to predict the dry matter intake of sheep raised in Brazil, based on a datasheet, called "original base" which was composed of 32 published articles with an approach related to the consumption and performance of sheep, as well as diet offered to these animals. For the models’ development, randomly16 articles were randomly selected from "original base", totaling 75 observations that made up the "development base". The quality of the adjusted models was confirmed by means of the residue analysis considering the coefficient of determination (R2 ), adjusted coefficient of determination for the number of parameters of the model (R2 ajust) and the mean square error (MSE). The significance level of 5% was adopted in the hypothesis tests. From the developed models, those who presented the best evaluation were chosen. The analysis was proceed with the comparison between the models developed and those presented in the literature, with the validation through a set of 68 observations from 16 articles, called "validation base", resulting from the selection made previously, i.e., the ones were not used for the models' development. To measure the accuracy and precision of the models, it was adopted the intercept test equal to zero and slope of the line equal to one, correlation coefficient and agreement coefficient, the mean square error of precision, among other measures. For the analyzes, MINITAB Version 15 and MES version 3.1.15 software were used. For the development of predictive models of dry matter (kg / day), combinations of some of the following predictors were considered: live weight (kg), mean daily gain (kg / day), metabolic weight (kg), squared live weight (kg2 ), daily mean square gain (kg/day2 ) and squared (kg2 ) metabolic live weight. The work showed that, among the models developed, the equation described by CMS (kg / day) = - 0.1134 + 0.02494 * PV + 4.77 * GMD - 11.135 * GMD2 with a determination coefficient of 62% to estimate the consumption of dry matter in a more precise and accurate way, presenting in its validation a correlation coefficient and agreement of approximately 0.55. This conclusion also applies when compared to other models found in the literature.