Abnormal value test
When analyzing a data set it can happen that we are confronted with values that do not seem to be a part of the normal data distribution. We call these abnormal value points and, as usual, you should not trust your intuition in detecting if a value is abnormal or not. There are tests that will show this.
From a statistical point of view, an abnormal value is a value that does not belong to the normal data distribution. It can come from:
- A measurement error or copying error (forgetting a decimal)
- A special reason such as a part that was not washed prior to being measured.
The entirety of statistical calculations using normal distribution properties (statistical tests, capability calculations, off-tolerance % calculations) is very sensitive to the presence of abnormal values. It is therefore important to understand their origin and to eliminate them before using these calculations. We may be able to use the non-parametric statistical tests, which are much less sensitive to abnormal values.
Deux tests sont principalement utilisés :
- Dixon test: very applicable when the amount of data is low (<30)
- Grubb test: can be used in all cases.