Resumo:
This dissertation presents a critical analysis regarding the use of some statistical
methods applied to Quality in cases in which the collected data are not normally distributed.
The general problem refers to the occurrence of questionable decision takings starting from
the interpretation of data considered normally distributed. The primordial justification for
this study is the relevance of the data analysis during its collection and handling in the
industrial practice and in scientific works in the most varied levels. The main objectives of
this work are: to present some examples related to the usage of control charts and six sigma
capability index that may induce to doubtful conclusions due to the incorrect assumption of
normality; to propose critical review proceeding to the Box-Cox and/or Johnson
transformations; and, finally, to discuss the conclusions and decisions established through the
comparison among the gross data originally collected, the data previously analyzed and
processed and the transformed data. The methodological approach combines experimental
research through the analysis of simulate data generated starting from a statistical software
as well as exploratory research with case study in an industrial process of precision hole
measurement.
