CombiStats: Robust regression
Outliers are a problematic feature of some bio-assays. It is in general not recommended to exclude values from the analysis only because of their magnitude without any experimental explanation for the occurrence of such values. However, sometimes no explanation can be found, whereas inclusion would really flaw the results. In such cases, it may be recommendable to use robust regression.
Here is example 5.1.3 from the European Pharmacopoeia with 1 value modified to make it appear an outlier. Three ways to analyse this assay are shown:
- Inclusion and applying normal regression
- Exclusion and applying normal regression
- Inclusion and applying robust regression
Robust regression can be achieved by using a weighting function of w=h, where h stands for Huber's robust weight with winsorisation factor k=1.5.