arXiv subjects

Markus Pauly

Publications and source records attributed to Markus Pauly.

At least 19 recordsLinked to original sources

arXiv2026

A Guide to Estimating Conditional Average Treatment Effects in Competing Risks Settings

Conditional average treatment effects (CATEs) are central to treatment decision-making in personalized medicine. In competing risks settings, estimating CATEs from survival data allows for patient-specific assessments of treatment effectiveness for a specific event of interest while properly accounting for alternative event types. This distinction is essential in the presence of comorbidities, where competing causes of death may otherwise confound the therapeutic benefit. Focusing on right-censored survival times with binary treatment, we examine CATEs defined as covariate-conditional differences in the absolute risk for the event of interest at a fixed time. To this end, we study meta-learners which adapt machine learning algorithms for CATE estimation in competing risks scenarios. We systematically compare six meta-learners, combining Cox regression or random survival forests for risk modeling with elastic net regression or random forests for direct CATE modeling. To provide practical guidance on model selection, we evaluate their performance in multiple simulation settings, that differ in hazard complexity, treatment heterogeneity, treatment assignment, event type distribution and censoring. To facilitate applied use, we provide the R package, crsurvlearners, which implements all considered approaches.

stat.AP↗

arXiv2026

Why is Regularization Underused? An Empirical Study on Trust and Adoption of Statistical Methods

Statistical practice does not automatically follow methodological innovation. Regularization methods, widely advocated to reduce overfitting and stabilize inference, are readily available in modern software, but are not consistently used by data analysts. We investigate this implementation gap in a large-scale empirical study of trust in, and acceptance of, regularization techniques, based on $N = 606$ data analysts. Drawing on measurement frameworks from technology acceptance research, we survey practitioners and embed a randomized experiment to test whether written recommendation of regularization methods increases trust or intended use. We find no evidence of such an effect. Instead, adoption intentions are strongly associated with analysts' perceptions of ease of implementation and practical benefit, such as improved bias control or interpretability. Perceived social norms also emerge as a central driver. These results indicate that uptake of statistical methodology depends less on formal recommendations than on usability, perceived utility, and community practice.

stat.OT↗

arXiv2026

Variable selection in linear mixed model meta-regression with suspected interaction effects -- How can tree-based methods help?

Detecting interaction effects (IEs) in meta-regression is challenging, especially when few studies are available and many plausible interactions are considered. In many meta-analyses, interpretability is essential, which limits the use of complex machine learning methods. Tree-based approaches offer a potentially useful compromise, but their role in meta-regression with random effects is not yet well understood. This paper examines how traditional linear and tree-based methods can support variable selection for IEs in random effects meta-regression. We compare test-based and information-criterion-based linear selection procedures with meta-CART approaches. These include fixed effect and random effects trees and their stability-selected ensemble variants. All methods are evaluated using a real-world meta-analytic dataset and a plasmode simulation study. The data-generating process assumes linear IEs and is complemented by settings with nonlinear interactions. Our results show that under strictly linear interactions, linear selection methods perform as expected and achieve superior performance for IE detection. Tree-based methods are more conservative when the number of studies is small, but become competitive as sample size increases, particularly the stability-selected variants. When IEs deviate from strict linearity, even in simple ways, the performance of linear methods deteriorates, whereas tree-based approaches, especially stability-selected fixed effect trees, provide a more robust alternative. Overall, stability-selected random effects trees are useful complementary tools for IE detection in applied meta-regression, particularly for metric covariates. They are well suited for pre-selection and sensitivity analyses, and selection frequency patterns in tree ensembles can help reveal structural patterns in the data.

stat.OT↗

arXiv2026

A Simulation Study to Compare Inferential Properties when Modelling Ordinal Outcomes: The Case for the (Plain but Robust) Proportional Odds Model

Ordinal measurements are common outcomes in studies within psychology, as well as in the social and behavioral sciences. Choosing an appropriate regression model for analysing such data poses a difficult task. This paper aims to facilitate modeling decisions for quantitative researchers by presenting the results of an extensive simulation study on the inferential properties of common ordinal regression models: the proportional odds model, the category-specific odds model, the location-shift model, the location-scale model, and the linear model, which incorrectly treats ordinal outcomes as metric. The simulations were conducted under different data generating processes based on each of the ordinal models and varying parameter configurations within each model class. We examined the bias of parameter estimates as well as type I error rates ($\alpha$-errors) and the power of statistical parameter testing procedures corresponding to the respective models. Our findings reveal several highlights. For parameter estimates, we observed that cumulative ordinal regression models exhibited large biases in cases of large parameter values and high skewness of the outcome distribution in the true data generation process. Regarding statistical hypothesis testing, the proportional odds model and the linear model showed the most reliable results. Due to its better fit and interpretability for ordinal outcomes, we recommend the use of the proportional odds model unless there are relevant contraindications.

stat.ME↗

arXiv2026

Inference for competing risks based on area between curves statistics

In competing risks models, cumulative incidence functions are commonly compared to infer differences between groups. Many existing inference methods, however, struggle when these functions cross during the time frame of interest. To address this problem, we investigate a test statistic based on the area between cumulative incidence functions. As the corresponding limiting distribution depends on quantities that are typically unknown, we propose a wild bootstrap approach to obtain a feasible and asymptotically valid two-sample test. The finite sample performance of the proposed method, in comparison with existing methods, is examined in an extensive simulation study.

stat.ME↗

arXiv2025

A Detailed Study on LLM Biases Concerning Corporate Social Responsibility and Green Supply Chains

Organizations increasingly use Large Language Models (LLMs) to improve supply chain processes and reduce environmental impacts. However, LLMs have been shown to reproduce biases regarding the prioritization of sustainable business strategies. Thus, it is important to identify underlying training data biases that LLMs pertain regarding the importance and role of sustainable business and supply chain practices. This study investigates how different LLMs respond to validated surveys about the role of ethics and responsibility for businesses, and the importance of sustainable practices and relations with suppliers and customers. Using standardized questionnaires, we systematically analyze responses generated by state-of-the-art LLMs to identify variations. We further evaluate whether differences are augmented by four organizational culture types, thereby evaluating the practical relevance of identified biases. The findings reveal significant systematic differences between models and demonstrate that organizational culture prompts substantially modify LLM responses. The study holds important implications for LLM-assisted decision-making in sustainability contexts.

cs.CY↗

arXiv2025

Permutation Tests Based on the Copula-Graphic Estimator and Their Use for Survival Tree Construction

Survival trees are popular alternatives to Cox or Aalen regression models that offer both modelling flexibility and graphical interpretability. This paper introduces a new algorithm for survival trees that relaxes the assumption of independent censoring. To this end, we use the copula-graphic estimator to estimate survival functions. This allows us to flexibly specify shape and strength of the dependence of survival and censoring times within survival trees. For splitting, we present a permutation test for the null hypothesis of equal survival. Our test statistic consists of the integrated absolute distance of the group's copula-graphic estimators. A first simulation study shows a good type I error and power behavior of the new test. We thereby asses simulation settings of various group sizes, censoring percentages and grades of dependence generated by Clayton and Frank copulas. Using this test as splitting criterion, a second simulation study studies the performance of the resulting trees and compares it with that of the usual logrank-based tree. Lastly, the tree algorithm is applied to real-world clinical trial data.

stat.ME↗

arXiv2025

Confidence Intervals for Random Forest Permutation Importance with Missing Data

Random Forests are renowned for their predictive accuracy, but valid inference, particularly about permutation-based feature importances, remains challenging. Existing methods, such as the confidence intervals (CIs) from Ishwaran et al. (2019), are promising but assume complete feature observation. However, real-world data often contains missing values. In this paper, we investigate how common imputation techniques affect the validity of Random Forest permutation-importance CIs when data are incomplete. Through an extensive simulation and real-world benchmark study, we compare state-of-the-art imputation methods across various missing-data mechanisms and missing rates. Our results show that single-imputation strategies lead to low CI coverage. As a remedy, we adapt Rubin's rule to aggregate feature-importance estimates and their variances over several imputed datasets and account for imputation uncertainty. Our numerical results indicate that the adjusted CIs achieve better nominal coverage.

stat.ME↗

arXiv2025

Explainable anomaly detection for sound spectrograms using pooling statistics with quantile differences

Anomaly detection is the task of identifying rarely occurring (i.e. anormal or anomalous) samples that differ from almost all other samples in a dataset. As the patterns of anormal samples are usually not known a priori, this task is highly challenging. Consequently, anomaly detection lies between semi- and unsupervised learning. The detection of anomalies in sound data, often called 'ASD' (Anomalous Sound Detection), is a sub-field that deals with the identification of new and yet unknown effects in acoustic recordings. It is of great importance for various applications in Industry 4.0. Here, vibrational or acoustic data are typically obtained from standard sensor signals used for predictive maintenance. Examples cover machine condition monitoring or quality assurance to track the state of components or products. However, the use of intelligent algorithms remains a controversial topic. Management generally aims for cost-reduction and automation, while quality and maintenance experts emphasize the need for human expertise and comprehensible solutions. In this work, we present an anomaly detection approach specifically designed for spectrograms. The approach is based on statistical evaluations and is theoretically motivated. In addition, it features intrinsic explainability, making it particularly suitable for applications in industrial settings. Thus, this algorithm is of relevance for applications in which black-box algorithms are unwanted or unsuitable.

stat.AP↗

arXiv2025

Multivariate and Multiple Contrast Testing in General Covariate-adjusted Factorial Designs

Evaluating intervention effects on multiple outcomes is a central research goal in a wide range of quantitative sciences. It is thereby common to compare interventions among each other and with a control across several, potentially highly correlated, outcome variables. In this context, researchers are interested in identifying effects at both, the global level (across all outcome variables) and the local level (for specific variables). At the same time, potential confounding must be accounted for. This leads to the need for powerful multiple contrast testing procedures (MCTPs) capable of handling multivariate outcomes and covariates. Given this background, we propose an extension of MCTPs within a semiparametric MANCOVA framework that allows applicability beyond multivariate normality, homoscedasticity, or non-singular covariance structures. We illustrate our approach by analysing multivariate psychological intervention data, evaluating joint physiological and psychological constructs such as heart rate variability.

stat.ME↗

arXiv2025

Choosing a Model, Shaping a Future: Comparing LLM Perspectives on Sustainability and its Relationship with AI

As organizations increasingly rely on AI systems for decision support in sustainability contexts, it becomes critical to understand the inherent biases and perspectives embedded in Large Language Models (LLMs). This study systematically investigates how five state-of-the-art LLMs -- Claude, DeepSeek, GPT, LLaMA, and Mistral - conceptualize sustainability and its relationship with AI. We administered validated, psychometric sustainability-related questionnaires - each 100 times per model -- to capture response patterns and variability. Our findings revealed significant inter-model differences: For example, GPT exhibited skepticism about the compatibility of AI and sustainability, whereas LLaMA demonstrated extreme techno-optimism with perfect scores for several Sustainable Development Goals (SDGs). Models also diverged in attributing institutional responsibility for AI and sustainability integration, a results that holds implications for technology governance approaches. Our results demonstrate that model selection could substantially influence organizational sustainability strategies, highlighting the need for awareness of model-specific biases when deploying LLMs for sustainability-related decision-making.

cs.CY↗

arXiv2025

A Cautionary Tale About "Neutrally" Informative AI Tools Ahead of the 2025 Federal Elections in Germany

In this study, we examine the reliability of AI-based Voting Advice Applications (VAAs) and large language models (LLMs) in providing objective political information. Our analysis is based upon a comparison with party responses to 38 statements of the Wahl-O-Mat, a well-established German online tool that helps inform voters by comparing their views with political party positions. For the LLMs, we identify significant biases. They exhibit a strong alignment (over 75% on average) with left-wing parties and a substantially lower alignment with center-right (smaller 50%) and right-wing parties (around 30%). Furthermore, for the VAAs, intended to objectively inform voters, we found substantial deviations from the parties' stated positions in Wahl-O-Mat: While one VAA deviated in 25% of cases, another VAA showed deviations in more than 50% of cases. For the latter, we even observed that simple prompt injections led to severe hallucinations, including false claims such as non-existent connections between political parties and right-wing extremist ties.

cs.LG↗

arXiv2024

Which Imputation Fits Which Feature Selection Method? A Survey-Based Simulation Study

Tree-based learning methods such as Random Forest and XGBoost are still the gold-standard prediction methods for tabular data. Feature importance measures are usually considered for feature selection as well as to assess the effect of features on the outcome variables in the model. This also applies to survey data, which are frequently encountered in the social sciences and official statistics. These types of datasets often present the challenge of missing values. The typical solution is to impute the missing data before applying the learning method. However, given the large number of possible imputation methods available, the question arises as to which should be chosen to achieve the 'best' reflection of feature importance and feature selection in subsequent analyses. In the present paper, we investigate this question in a survey-based simulation study for eight state-of-the art imputation methods and three learners. The imputation methods comprise listwise deletion, three MICE options, four \texttt{missRanger} options as well as the recently proposed mixGBoost imputation approach. As learners, we consider the two most common tree-based methods, Random Forest and XGBoost, and an interpretable linear model with regularization.

stat.AP↗

arXiv2024

A Central Limit Theorem for the permutation importance measure

Random Forests have become a widely used tool in machine learning since their introduction in 2001, known for their strong performance in classification and regression tasks. One key feature of Random Forests is the Random Forest Permutation Importance Measure (RFPIM), an internal, non-parametric measure of variable importance. While widely used, theoretical work on RFPIM is sparse, and most research has focused on empirical findings. However, recent progress has been made, such as establishing consistency of the RFPIM, although a mathematical analysis of its asymptotic distribution is still missing. In this paper, we provide a formal proof of a Central Limit Theorem for RFPIM using U-Statistics theory. Our approach deviates from the conventional Random Forest model by assuming a random number of trees and imposing conditions on the regression functions and error terms, which must be bounded and additive, respectively. Our result aims at improving the theoretical understanding of RFPIM rather than conducting comprehensive hypothesis testing. However, our contributions provide a solid foundation and demonstrate the potential for future work to extend to practical applications which we also highlight with a small simulation study.

math.ST↗

arXiv2024

Quadratic Form based Multiple Contrast Tests for Comparison of Group Means

Comparing the mean vectors across different groups is a cornerstone in the realm of multivariate statistics, with quadratic forms commonly serving as test statistics. However, when the overall hypothesis is rejected, identifying specific vector components or determining the groups among which differences exist requires additional investigations. Conversely, employing multiple contrast tests (MCT) allows conclusions about which components or groups contribute to these differences. However, they come with a trade-off, as MCT lose some benefits inherent to quadratic forms. In this paper, we combine both approaches to get a quadratic form based multiple contrast test that leverages the advantages of both. To understand its theoretical properties, we investigate its asymptotic distribution in a semiparametric model. We thereby focus on two common quadratic forms - the Wald-type statistic and the Anova-type statistic - although our findings are applicable to any quadratic form. Furthermore, we employ Monte-Carlo and resampling techniques to enhance the test's performance in small sample scenarios. Through an extensive simulation study, we assess the performance of our proposed tests against existing alternatives, highlighting their advantages.

stat.ME↗

arXiv2024

Is GPT-4 Less Politically Biased than GPT-3.5? A Renewed Investigation of ChatGPT's Political Biases

This work investigates the political biases and personality traits of ChatGPT, specifically comparing GPT-3.5 to GPT-4. In addition, the ability of the models to emulate political viewpoints (e.g., liberal or conservative positions) is analyzed. The Political Compass Test and the Big Five Personality Test were employed 100 times for each scenario, providing statistically significant results and an insight into the results correlations. The responses were analyzed by computing averages, standard deviations, and performing significance tests to investigate differences between GPT-3.5 and GPT-4. Correlations were found for traits that have been shown to be interdependent in human studies. Both models showed a progressive and libertarian political bias, with GPT-4's biases being slightly, but negligibly, less pronounced. Specifically, on the Political Compass, GPT-3.5 scored -6.59 on the economic axis and -6.07 on the social axis, whereas GPT-4 scored -5.40 and -4.73. In contrast to GPT-3.5, GPT-4 showed a remarkable capacity to emulate assigned political viewpoints, accurately reflecting the assigned quadrant (libertarian-left, libertarian-right, authoritarian-left, authoritarian-right) in all four tested instances. On the Big Five Personality Test, GPT-3.5 showed highly pronounced Openness and Agreeableness traits (O: 85.9%, A: 84.6%). Such pronounced traits correlate with libertarian views in human studies. While GPT-4 overall exhibited less pronounced Big Five personality traits, it did show a notably higher Neuroticism score. Assigned political orientations influenced Openness, Agreeableness, and Conscientiousness, again reflecting interdependencies observed in human studies. Finally, we observed that test sequencing affected ChatGPT's responses and the observed correlations, indicating a form of contextual memory.

cs.CL↗

arXiv2024

Single CASANOVA? Not in multiple comparisons

When comparing multiple groups in clinical trials, we are not only interested in whether there is a difference between any groups but rather the location. Such research questions lead to testing multiple individual hypotheses. To control the familywise error rate (FWER), we must apply some corrections or introduce tests that control the FWER by design. In the case of time-to-event data, a Bonferroni-corrected log-rank test is commonly used. This approach has two significant drawbacks: (i) it loses power when the proportional hazards assumption is violated [1] and (ii) the correction generally leads to a lower power, especially when the test statistics are not independent [2]. We propose two new tests based on combined weighted log-rank tests. One as a simple multiple contrast test of weighted log-rank tests and one as an extension of the so-called CASANOVA test [3]. The latter was introduced for factorial designs. We propose a new multiple contrast test based on the CASANOVA approach. Our test promises to be more powerful under crossing hazards and eliminates the need for additional p-value correction. We assess the performance of our tests through extensive Monte Carlo simulation studies covering both proportional and non-proportional hazard scenarios. Finally, we apply the new and reference methods to a real-world data example. The new approaches control the FWER and show reasonable power in all scenarios. They outperform the adjusted approaches in some non-proportional settings in terms of power.

stat.ME↗

arXiv2024

AR-Sieve Bootstrap for the Random Forest and a simulation-based comparison with rangerts time series prediction

The Random Forest (RF) algorithm can be applied to a broad spectrum of problems, including time series prediction. However, neither the classical IID (Independent and Identically distributed) bootstrap nor block bootstrapping strategies (as implemented in rangerts) completely account for the nature of the Data Generating Process (DGP) while resampling the observations. We propose the combination of RF with a residual bootstrapping technique where we replace the IID bootstrap with the AR-Sieve Bootstrap (ARSB), which assumes the DGP to be an autoregressive process. To assess the new model's predictive performance, we conduct a simulation study using synthetic data generated from different types of DGPs. It turns out that ARSB provides more variation amongst the trees in the forest. Moreover, RF with ARSB shows greater accuracy compared to RF with other bootstrap strategies. However, these improvements are achieved at some efficiency costs.

stat.ML↗