General Statistics

Statistics is a very broad field with many specialty areas and dedicated methods for typical problems within those, but with substantial commonalities.

I leverage my broad experience in applying statistics to a large variety of problems from very different applications to provide best methods along with appropriate and meaningful interpretation of results from quantitative methods. I provide support in various areas. While it is impossible to provide an exhaustive list, some examples follow below.

Problem definition

A clear understanding on the project needs and the alignment of all stakeholders on the research purpose is indispensable, along with an understanding of its limitations (i.e. what a certain study might not answer). Failing on getting this crucial step right in the beginning is unfortunately common and almost invariably leads to frustrations or even anger later on.

Aspects for careful consideration ▼

What is the most important question to answer with the research?
Is the problem measurable, and how?
What is the scope? E.g. a certain consumer population of primary interest or specific line of a production process in a given plant.
Which endpoints are critical? Which are most important, and which maybe secondary?
What does success look like?
What risks are you willing to take? Most notably the risk of ‘finding’ something just by chance, and the risk of not finding a desired effect even if true.

Design of Experiments / Study design

With a proper problem statement at hand, the next step is to design your quantitative research. Proper selection of prototypes to test, for example, can maximize the information that is obtained from the experiment, maximizing return on investment. Designs are typically most efficient when the different aspects are considered simultaneously.

What you want to think about ▼

Design of Experiments (DoE) – a powerful approach to choose prototypes to manufacture and test
Use of covariates or other relevant information
Sample size calculations
Minimization of potential bias using
- Block designs and rotation schemes (e.g. Balanced or Imbalanced Block Designs)
- Randomization
Sequential testing and interim analyses

Hypothesis testing

You are familiar with type I and type II error rates (aka as $\alpha$ and $\beta$ )? Great. You are not? No problem, I bring you up to speed. Though unpopular at times, I do not shy away from making these concepts clear as they are indispensable for data-based risk management. Navigating through uncertainty is not a question of luck, but of understanding and dealing with the respective risks.

Key considerations ▼

Define associated risks
Understand whether testing is for differences or equivalence/similarity, or maybe for non-inferiority
Appropriate models based on the experimental design
Assumptions to be made (or to be avoided)
Development of tailor-made tests as needed
Determination of confidence intervals (and making their value understood)

Statistical Modelling

Modelling is an important area of statistics with many options and facets. It is essential to avoid overfitting and minimize the complexity of the model – not least to ease interpretation – without oversimplifying it for the problem at hand. It usually takes multiple attempts and comparisons, and always needs to link back to the problem statement to decide how good is good enough. Modeling also includes many multivariate methods.

Some modeling techniques ▼

Linear, generalized linear and non-linear regression
Analysis of Variance (ANOVA) and related methods
Partition methods and regression trees
An abundance of multivariate methods (typically focusing on dimension reduction), e.g.
- Principal Component Analysis (PCA) and Biplots
- Correspondence Analysis
- Partial Least Squares methods
- Multidimensional Scaling (MDS)