Maize productivity based on a distributional regression approach

Abstract

The objective of this work was to propose the use of traditional models based on distributional regression models to analyze maize productivity. The experiment was carried out in an alpha lattice design, with three replicates and 24 blocks. Data used refer to 102 maize plants from the permanent collection of the Centro de Desenvolvimento Científico e Tecnológico para a Agricultura of the Universidade Federal de Lavras. For the maize productivity evaluation, the following explanatory variables were used: weight of 100 seed, plant height, ear height, and days to maturation. The initial analyses involved the fitting of four distributions (gamma, generalized gamma, inverse Gaussian, and generalized inverse Gaussian) to the data, in which the gamma distribution showed the best fit based on the Akaike and Bayesian information criteria (AIC and BIC). Cob height has a considerable influence on the productivity variability because as cob height increases, the productivity variability decreases, whereas the covariates weight of 100 seed and days to maturity explain the increasing average of the productivity. The residual analysis shows that the model based on gamma distribution is suitable for explaining the data and providing useful insights for agricultural research and practice.