Normality Tests and Normality Transformations: Explanations, Justifications, and Uses

The calculations used in many statistical tests and methods require that the inputted data be “normally distributed”. This webinar explains what it means to be “normally distributed”, how to assess normality, how to test for normality, and how to transform non-normal data into normal data, and how to justify the transformations to internal and external quality system auditors.

Why Should You Attend:

Being able to assess whether data is “normally distributed”, and to be able to "transform to normality" is critical to ensuring that a company's “valid statistical techniques” are “suitable for their intended use” (as required by the FDA). Therefore, it is critical to a company's success. Most users of statistics make the error of assuming normality, in order to simplify their statistical analyses. However, most data sets in industry are not normally distributed, and not noticing that oftentimes results in rejecting lots that should have passed, failing processes that actually met their validation criteria, or keeping products in R&D long after they should have been transferred to Manufacturing.

Such calculations include those for Student's t-Tests, ANOVA tables, F-tests, Normal Tolerance limits, and Process Capability Indices. Unless the raw data used in such calculations is “normally distributed”, the resulting conclusions may be incorrect.

Dimensional data (length, width, height) are typically normally distributed. But many other types of data sets are almost always non-normal, such as: tensile strength, burst pressure, and time or cycles to failure. Some non-normal data can be transformed into normality, in order to then allow statistical calculations to be valid when run on the transformed data.