Statistics and Data Science

We conduct reserach in three areas and their intersection: psychometric modeling, machine learning methods, and multilevel and longitudinal modeling. On the one hand, we assess existing tools, test their applicability, and demonstrate opportunities and pitfalls for analyses in psychological research. On the other hand, we develop new tools and extend existing tools to offer new possibilities for data analysis and hypothesis generation. Our methodological approach are mainly simulation studies conducted in the open-source software R and illustrations of the new methods using empirical data from psychology.

Our Research in Detail

Psychometric Modeling

Psychometric modelling tackles challenges in the area of psychological testing and measurement. We examine response biases in self-report measures, e.g., in personality measurement. Such response biases can distort measurement and have implications, for instance, on educational decisions. Furthermore, we examine methods for differential item and differential step functioning. These phenomena occur when certain items are more difficult for respondents with certain background characteristics. In our research group, we examine and develop tools that support researchers to improve data quality by extending psychometric modeling approaches or identifying relevant background variables, for example using machine learning methods.

Machine Learning Methods & Interpretation Techniques

Machine learning methods, such as decision trees or random forests, are robust, yet powerful methods to capture and interpret complex dynamics and non-linear effects of predictor variables on outcomes. At the same time, many machine learning methods are so-called “black boxes” that do not allow us to see in what way the machine learning method has come to its prediction. In our research group, we assess machine learning methods and interpretability techniques for their applicability in psychological research. We test how these methods react to data structures that are typical in psychology, such as correlated predictor variables, or data with a multilevel structure. We also extend machine learning methods, for example by tools from psychometrics, to enhance interpretation.

Multilevel and Longitudinal Modeling

Many data in psychology are hierarchically clustered, for example when students are clustered in classes, or multiple measurements are clustered in persons. Multilevel modeling has become an important statistical tool to study humans in contexts, but also to distinguish intra- and interindividual psychological processes. In our research group, we extend the toolbox for the analysis of multilevel data. We assess how machine learning methods can be used as a reliable tool to analyze multilevel and longitudinal data. We also develop new psychometric approaches that help to assess data quality issues in intensive longitudinal assessments.

Nach oben