Multivariate Statistics:

Multivariate Statistics is the tool most commonly used to interpret PAT data.  And it comes in many flavours.  These techniques can all be very effective when used correctly.  There are a large number of commercially available software packages, some better suited to scientific research and development, while others are designed specifically for use in GMP environments.

The most common applications for MV statistics are:

1. Classification – e.g. material identity tests.
2. Quantification – e.g. prediction of quality critical attributes (e.g. concentration).
3. Creating Process signatures.

There are many methods for classification.  PLS is hands-down the most successful method of creating quantitative/predictive assays (also called multivariate calibrations).  For process signatures, PCA can be used to measure process variance, and chart how it evolves with time in (presumably) a characteristic manner.

Multivariate statistics is particularly widely used in the interpretation of spectroscopic data (Near Infrared spectroscopy being the prime example).

Experimental Design / Design of Experiments:

Where problems contain multiple variables, or indeed if a problem is poorly understood and there are a large number of variables that “could” be implicated, and assuming you have the luxury of being able to create samples or processing conditions at the desired set points, then Experimental Design / Design of Experiments is used to minimise the number of experiments required to produce the maximum amount of useful information. 

(Was this helpful? Could you write a better summary? Do you want to?  Get in touch. - Ed)