Karsten Lübke’s research while affiliated with FOM University of Applied Sciences for Economics and Management and other places

This study contributes to the literature by analysing the joint association of managerial overconfidence, certainty, narcissism, and the Big Five personality traits with debt ratios in the institutional setting of the German two-tier system. Moreover, it provides insights into how corporate governance quality moderates the effects of personality. The analysis relied on the chief executive officers’ (CEOs’) speeches at annual general meetings (AGMs) that were voluntarily disseminated, a novel data source. Managers’ personality traits were measured using software-aided content analysis, and their impact on the debt ratio was analysed using panel regressions. Consistent with previous studies, the debt ratios of German issuers are significantly and positively related to the proxies of managerial certainty and narcissism. However, their model inclusion contributes only marginally to explanatory power. Conversely, the coefficients of the proxies for the Big Five personality traits remained statistically non-significant. Moreover, a significantly negative relationship between debt ratios and the interaction term between a proxy for corporate governance quality and managerial certainty is observed that corresponds to the risk-mitigating impact of corporate governance.

Image analysis represents a crucial and dynamic field within the realm of data science and machine learning, enabling automated interpretation and analysis of images through statistical methods. Beyond its practical applications, such as image recognition, images offer a valuable pedagogical tool for teaching various multivariate analysis techniques, including cluster analysis, principal component analysis, and k nearest neighbors. By employing straightforward R code, images can be transformed into tidy data formats, primed for multivariate analysis. The resultant analysis outcomes can then be translated back into images, affording students the opportunity to visually comprehend the impact of techniques like cluster analysis, principal component analysis or k nearest neighbors. This approach effectively bridges the gap between abstract multivariate analysis concepts and concrete understanding, as students can visually perceive how cluster centroids or principal components reduce the complexity of the original (image) data.

Risk prediction models in credit scoring have to fulfil regulatory requirements, one of which consists in the interpretability of the model. Unfortunately, many popular modern machine learning algorithms result in models that do not satisfy this business need, whereas the research activities in the field of explainable machine learning have strongly increased in recent years. Partial dependence plots denote one of the most popular methods for model-agnostic interpretation of a feature’s effect on the model outcome, but in practice they are usually applied without answering the question of how much can actually be seen in such plots. For this purpose, in this paper a methodology is presented in order to analyse to what extent arbitrary machine learning models are explainable by partial dependence plots. The proposed framework provides both a visualisation, as well as a measure to quantify the explainability of a model on an understandable scale. A corrected version of the German credit data, one of the most popular data sets of this application domain, is used to demonstrate the proposed methodology.

Without random sampling and/or random allocation, even descriptive statistics such as simple means or proportions can be quite misleading. Therefore, causal diagrams were added to existing course materials to address this topic and to illustrate the differences between random and convenience samples and between observational and experimental studies. We assessed student understanding in different courses with a pre-/post-survey. Additionally, we asked students to evaluate the helpfulness of the diagrams for their understanding. There is a statistically discernible positive effect with 280 students from more than seven different courses on pre- to post-knowledge. Also, most of the students agreed with the statement that the causal diagrams helped in their understanding.

Statisticians and data scientists transform raw data into understanding and insight. Ideally, these insights empower people to act and make better decisions. However, data is often misleading especially when trying to draw conclusions about causality (for example, Simpson’s paradox). Therefore, developing causal thinking in undergraduate statistics and data science programs is important. However, there is very little guidance in the education literature about what topics and learning outcomes, specific to causality, are most important. In this paper, we propose a causality curriculum for undergraduate statistics and data science programs. Students should be able to think causally, which is defined as a broad pattern of thinking that enables individuals to appropriately assess claims of causality based upon statistical evidence. They should understand how the data generating process affects their conclusions and how to incorporate knowledge from subject matter experts in areas of application. Important topics in causality for the undergraduate curriculum include the potential outcomes framework and counterfactuals, measures of association versus causal effects, confounding, causal diagrams, and methods for estimating causal effects.

In the recent past, several popular failures of black box AI systems and regulatory requirements have increased the research interest in explainable and interpretable machine learning. Among the different available approaches of model explanation, partial dependence plots (PDP) represent one of the most famous methods for model-agnostic assessment of a feature’s effect on the model response. Although PDPs are commonly used and easy to apply they only provide a simplified view on the model and thus risk to be misleading. Relying on a model interpretation given by a PDP can be of dramatic consequences in an application area such as forensics where decisions may directly affect people’s life. For this reason in this paper the degree of model explainability is investigated on a popular real-world data set from the field of forensics: the glass identification database. By means of this example the paper aims to illustrate two important aspects of machine learning model development from the practical point of view in the context of forensics: (1) the importance of a proper process for model selection, hyperparameter tuning and validation as well as (2) the careful used of explainable artificial intelligence. For this purpose, the concept of explainability is extended to multiclass classification problems as e.g. given by the glass data.

Following the growing need for statistical knowledge and the ability to handle data, an increasing number of degree programs include statistics courses and lectures in their curricula. Especially in non-STEM programmes, but not exclusively, it is possible to find heterogeneous groups of students in terms of attitudes towards these topics, and therefore, deeper insights into these attitudes can help to improve curricula and teaching to better fit with the respective student cohort. In a case study, we analyse the results of a survey in which students were asked about their attitude towards statistics, based on the well-known and widely-used SATS-36 questionnaire. Our aim is to identify different attitudinal profiles and to make the heterogeneity of student groups visible. By conducting several cluster analyses, we are able to find separable student groups that can be differentiated, e.g., by their interest towards statistics, by their self-confidence towards their own abilities, or by their willingness to invest more or less effort into learning for the respective class. It turns out that using the individual items instead of the proposed SATS-36 attitudinal constructs lead to a better separation of the student clusters, as does taking gender into account in a cluster analysis of mixed-type data.