Repetitividade e reprodutibilidade na validação de métodos para testes de germinação de sementes de espécies florestais

Abstract

In the process of method validation, the International Association for Seed Analysis quantifies repeatability and reproducibility by lot. However, this process for seed analysis is defined by a analysis of variance model with two factors, and an R&R study is recommended. Regardless if estimated by lot or by factorial analysis of variance model, repeatability and reproducibility of validated methods for testing germination of Brazilian forest species need to be quantified and evaluated, this being the purpose of this dissertation. For this, laboratories received seed lots of distinct qualities and performed germination tests according to the methods and procedures proposed. From the analysis of variance for normal seedlings with the laboratory and batch factors and from batch results, we calculated absolute and relative measures of repeatability, reproducibility and R&R, besides Mandel s h and k statistics. The absolute measures and the contribution percentage of the variances are not good tools to evaluate germination performance, because the former does not detect the high variations occurring in tests involving lots of low quality seeds and the latter is influenced by the variability between seed lots. Mandel s h and k statistics assist the process only in identifying laboratories that were not repetitive or reproductive, not allowing inferences about the quality of the method tested. The coefficient of variation is able to detect variations in germination tests based on the quality of the seed lot, being smaller in high quality lots and greater as the quality of lots decreases. Thus, the repeatability variation and reproducibility coefficients are recommended to evaluate the variability by seed lot and the coefficient of variation of R&R to evaluate the performance of the method.