Name: JASMYN TOGNERE
Publication date: 31/07/2025
Advisor:
Name |
Role |
|---|---|
| EDILSON ROMAIS SCHMILDT | Advisor |
Examining board:
| Name |
Role |
|---|---|
| EDILSON ROMAIS SCHMILDT | Presidente |
| EDNEY LEANDRO DA VITORIA | Examinador Interno |
| ENILTON NASCIMENTO DE SANTANA | Examinador Externo |
| SARA DOUSSEAU ARANTES | Coorientador |
Summary: The coefficient of variation is a widely used statistical measure to express experimental
precision in agricultural and forestry studies, as it allows for the comparison of relative
variability among different traits. However, the traditional classification of the coefficient
of variation, proposed by Pimentel-Gomes, presents significant limitations, particularly
due to its generalist nature and its failure to consider fundamental aspects such as the
type of variable evaluated and the size of the experimental plot. This fixed approach
may lead to misinterpretations regarding the quality of experiments, especially when
applied to different crops or specific experimental conditions. In light of these
limitations, a new classification of the coefficient of variation is proposed, which is
better adapted to the context of experiments involving eucalyptus seedlings at the
dispatch stage. Nine morphological traits were evaluated in seedlings of six eucalyptus
clones (144, 224, BA7346, CO1407, TP361, and GG100), totaling eight thousand and
two hundred units. The estimation of the optimal plot size was conducted using the
modified maximum curvature method with bootstrap simulation, a method widely
recognized for its capacity to identify the point at which the addition of experimental
units ceases to provide significant gains in precision. The values of the coefficient of
variation obtained were subsequently analyzed according to the data distribution. For
variables that presented a normal distribution, a classification based on means and
standard deviations was used; for variables with a non-normal distribution, a
methodology based on the median and pseudo-sigma was adopted. The results
showed that the behavior of the coefficient of variation as a function of the number of
plants per plot is non-linear. As the number of plants increases, the required values of
the coefficient of variation to classify the data within ranges of low variability become
more stringent, correctly reflecting the reduction in experimental error. This variation in
behavior was also influenced by the nature of the morphological trait evaluated, which
reinforces the inadequacy of fixed and generalist classifications. The new proposal
presented classification ranges with well-defined transitions between classes and
greater adherence to the variability observed in real data. Moreover, the proposed viii
methods proved to be more effective than the traditional model by considering the data
distribution, thus providing greater sensitivity in the evaluation of experimental
precision. Therefore, the proposed approach constitutes a more appropriate statistical
tool for the reality of forest nurseries, contributing to increased reliability of experiments
and to improved criteria for the evaluation of eucalyptus seedlings at the dispatch
stage.
