Diferentes métodos estadísticos para el análisis de variables discretas. Una aplicación en las ciencias agrícolas y técnicas

the objective of this article was to evaluate three different statistical methods to conduct analyses of discrete variables. the information came from an experiment developed at the camilo cienfuegos genetics enterprise in the pinar del río province in 2007-2008 related to the ct-115 forrage production. a complete randomized design was used with three treatments and ten repetitions. the variables analysed were: number of stems, number of sprouts, total number of leaves/stem, total number of leaves/sprout, number of dried leaves/stem and number of dried leaves/sprouts. the parametric variance analysis and its homologous non-parametric, kruskal-wallis test and the generalized lineal model were taken into account. the theoretical assumptions of the variance analyses to the test error normality were verified. the shapiro wilk test, kolmogorov smirnov and the lilliefors test were used, shapiro wilk test was the most robust to detect lack of normality. for the variance homogeneity, the bartlett and levene test were used both with similar results. the variables were transformed with the square root transformation which did not improve the normal distribution adjustment to the variable number of dried leaves/sprout. the probability values maintained the same outcomes respect to ho tests for the non-parametric test, compared with its homologous parametric f from fisher test. the criteria of goodness of fit in the generalized linear model permitted evaluating the best adjustment effects. it was considered that this model is more flexible than the parametric variance analyses because the variables under study did not require the theoretical assumptions fulfilment.