Parametric or Non-Parametric
Your manufacturing business uses statistics to measure its performance. There are a wealth of statistical tools out there and all of them have preconditions that let you know how capable the tool will be. One of the most important preconditions is that the inference or population data is normally distributed. The fact of the matter is that no distributions are exactly normally distributed except theoretical ones. However, if you are close to normally distributed within the tools that require a normal distribution are adequate.
If you’re data distribution deviates too much from normality then you start to look at non-parametric statistical tools. Non-parametric statistical tools do not require that the data is normally distributed.
It can be very difficult to determine if your probabilistic data is in fact normal. Often times there is not enough data to emphatically make that conclusion.
This leaves the manufacturer in a difficult position. It’s entirely possible that you are making an assumption whose inaccuracy invalidates business conclusions.
Lean Business recommends that you engage us or a Lean Six Sigma Black Belt.
If you are currently using a graphical technique to determine the normality of your data and that is a good start however, we recommend that you also look at the Anderson–Darling Normality Test. This is a test for normality. The next recommendation is to analyze the current test you’re using for sensitivity to the assumption of normality. Lastly we recommend that you investigate the analogous non-parametric test and analyze its relevance. If relevant, then perform that test also and compare it to your parametric test to determine the similarity of results. If the results are close, then you are in good shape. Otherwise, If the results are not close, then it is time to investigate the preconditions.
The right tool for the right job will validate the fact that you are making the right decisions.