Search

In agronomic design of experiments, researchers frequently assume linear associations between explanatory and response variables, also they assume a gaussian distribution or transform it. Inherent variability of the experiment units is not considered. Besides, it is frequent to analyze the time as a factor of a factorial design when repeated measures are taken through time, missing the opportunity to analyze the plant growth. New statistical models present the opportunity to analyze the raw data without transforming, they also can consider the variability of the experiment units as random components and they can identify the relationship between explanatory and response variables as linear or nonlinear. In cut roses production, one of the main quality factors is stem length, but is not clear how temperature, relative humidity and light affect it. The aim of this research was to analyze new statistical models to better understanding the impact of environmental variables on stem length of two rose cultivars (Samurai and Blush) during two peak production periods. The models used were: generalized linear models (GLM), generalized linear mixed models (GLMM), generalized additive models (GAM), generalized additive mixed models (GAMM), vector generalized lineal models (VGLM) and repeated measures model. The results showed that GAM model with a gamma distribution and identity link function was the best model with minimum BIC value and minimum error variance. It identified a nonlinear effect of relative humidity and linear effects of heat and light. The best conditions to produce plants between 50 and 70 cm, were 650 to 830 heat units and 82.5 to 85% of relative humidity. GAMM identified the period of April to May with cultivar Blush as the best conditions to produced roses. Repeated measures analysis confirmed that plants initiated high (low), remained so through time. [ABSTRACT FROM AUTHOR]