How to analyze germination of species with empty seeds using contemporary statistical methods?

Detalhes bibliográficos
Autor(a) principal: Santana,Denise Garcia de
Data de Publicação: 2018
Outros Autores: Carvalho,Fábio Janoni, Toorop,Peter
Tipo de documento: Artigo
Idioma: eng
Título da fonte: Acta Botanica Brasilica
Texto Completo: http://old.scielo.br/scielo.php?script=sci_arttext&pid=S0102-33062018000200271
Resumo: ABSTRACT Statistical analysis is considered an important tool for scientific studies, including those on seeds. However, seed scientists and statisticians often disagree on the nature of variables addressed in germination experiments. Statisticians consider the number of germinated seeds to be a binomially distributed variable, whereas seed scientists convert it into a percentage and often analyze it as a normally distributed variable. The requirement for normal adjustment restricts the models of analysis of variance that can be used. Lack of fit requires nonparametric tests, but they are known by their inferential problems. Generalized Linear Models (GLM) can provide better fit to germination variables for any species, including Lychnophora ericoides Mart., because they allow wider probability distributions with fewer requirements. Here we suggest the use of relative germination besides absolute germination for species with seed development problems, such for L. ericoides and others from the campos rupestres. This paper introduces the most current statistical advancements and increases the possibilities for their application in seed science research.