ABSTRACT.
There are researchers who do not recommend data transformation arguing it causes problems in inferences and mischaracterises data sets, which can hinder interpretation. There are other researchers who consider data transformation necessary to meet the assumptions of parametric models. Perhaps the largest group of researchers who make use of data transformation are concerned with experimental accuracy, which provokes the misuse of this tool. Considering this, our paper offer a study about the most frequent situations related to data transformation and how this tool can impact ANOVA assumptions and experimental accuracy. Our database was obtained from measurements of seed physiology and seed technology. The coefficient of variation cannot be used as an indicator of data transformation. Data transformation might violate the assumptions of analysis of variance, invalidating the idea that its use will provoke fail the inferences, even if it does not improve the quality of the analysis. The decision about when to use data transformation is dichotomous, but the criteria for this decision are many. The unit (percentage, day or seedlings per day), the experimental design and the possible robustness of F-statistics to ‘small deviations’ to Normal are among the main indicators for the choice of the type of transformation.
Keywords:
assumptions; coefficient of variation; criteria for data transformation; parametric and nonparametric statistics; robustness