Prepare and perform PCA/OPLS

1. Evaluate ANOVA for all Factors and all Variables employing Posthoc-Tests, not t-Tests

2. Choose a Significance % margin/level  for different groups of Variables according to your professional experience/information

3. Identify/eliminate Outliers and/or unreliable datapoints or variables or factor levels

3.a) Fill in missing values with very small values / medians of similar factor levels optionally

4. Choose an appropriate* Standardisation/Normalisation/Transformation (scaling, centering --> correlation matrix instead of covariance matrix)

4.a) Perform PCA, discuss Biplot and factor loading

5. store PCAs in the datasheet

6. devise your own PCAs from intercorrelation pattern, factor loading and professional experience, make a recursive PCA using self constructed Metavariables

7. and finallay: quantify PCA results by means of ANOVA

-

do not:

• go blindly for p < 0.05 variable filters, as you might loose key information
• log transform data without prior utilisation of the procedures above and without sound reasonig
• sell a DCA/OPLS-DA for a "supervised" PCA, as they serve different purposes

optionally: Perform an NMDS (Z- Transformation, Similarity matrix) and correlate NMDS metavariables to PCA metavariables