Statistics in times of AI

This year's DAGStat2025 conference will feature a variety of scientific sections, reflecting key areas of current research in statistics. Below is the list of sections that will be covered during the conference.

• 10 Biometry and Epidemiology
• 11: Causal Inference
• 12: Design of Experiments and Clinical Trials
• 13: Nonclinical Statistics
• 14: Statistical Methods in Epidemiology
• 15: Statistics in the Pharmaceutical and Medical Device Industry
• 16: Survival and Event History Analysis
• 20 Data Science, Machine Learning, and AI
• 21: Artificial Intelligence and Machine Learning
• 22: Clustering and Classification
• 23: Deep Learning + Statistics
• 24: Text Mining, NLP and Content Analysis
• 25: Time Series and Statistical Forecasting
• 26: Trustworthy Data Science
• 27: Visualisation and Exploratory Data Analysis
• 30 Statistical Applications
• 31: Bioinformatics and Systems Biology
• 32: Empirical Economics and Applied Econometrics
• 33: Marketing and E-Commerce
• 34: Statistics in Agriculture and Ecology, Environmental Statistics
• 35: Statistics in Finance
• 40 Statistical Theory and Methods
• 41: Advanced Regression Modelling
• 42: Bayesian Statistics
• 43: Mathematical Statistics
• 44: Meta-Analysis
• 45: Robust and Nonparametric Statistics
• 46: Testing and Scaling
• 50 Survey, Social Sciences, and Educational Statistics
• 51: Official and Survey Statistics
• 52: Regional and Urban Statistics
• 53: Statistical Literacy and Statistical Education
• 54: Statistics in Social, Behavioral and Educational Sciences
• 55: Structural Equation Modelling and Latent Variables
• 56: Survey Methodology
• 60 Special Topics
• 61: Extreme Values and Rare Events
• 62: Network Analysis
• 63: Spatial and Spatio-temporal Statistics
• 64: Statistical Software
• 65: Synthetic Data, Georeferencing & Disclosure Control
• 90 by invitation only
• 91: Young Statisticians
• 92: STRATOS

Related Links

• details on the keynote speakers and invited speakers
• abstract submission

Keynote Speakers

Invited Speakers

Related Links

• list of the conference sections
• abstract submission

Monday is Tutorial Day. There are Morning Tutorials, Afternoon Tutorials and Full Day Tutorials:

Morning Tutorials

• Target trial emulation and causal inference for time-dependent treatments
  
  Vanessa Didelez, Malte Braitmaier and Bianca Kollhorst (BIPS Bremen, Germany)

  Using routinely collected data (or real-world data) to answer questions about (long-term) effectiveness or safety of drugs or preventive measures bears sources of potential bias, besides confounding. These are termed “self-inflicted” biases because they can be minimised by careful study designs. Target trial emulation (TTE) is an increasingly popular principle to facilitate the formulation of the causal estimand and determine a suitable designs. However, lack of familiarity with statistical techniques and software may deter researchers from actually using TTE. Moreover, typical analyses are often limited to ITT-type emulation even if the intended use of drugs or treatments is sustained or adaptive. This course therefore aims to address and demonstrate practical aspects and implementation of TTE specifically for time-dependent treatments. In particular, we will cover techniques of “sequential target trials”, “artificial censoring & weighting” and “cloning” to avoid, e.g., immortal time or prevalent user and similar types of bias.

  Some basic knowledge of methods for causal inference will be helpful but is not strictly required. Exercises will be "pen and paper" (or their electronic equivalents).

• Social Media and Statistics - How Do They Fit Together?
  
  Alexander Schacht (The Effective Statistician, Sanevidence GmbH, Germany)

  In today's digital age, social media platforms, especially LinkedIn, offer unprecedented opportunities for statisticians and researchers to enhance their visibility, engage with a broader audience, and foster a community centered around their work. However, leveraging these platforms effectively requires more than just self-promotion; it involves a strategic approach to content creation and networking that emphasizes value and collaboration.This workshop, "Social Media and Statistics - How Do They Fit Together?" aims to bridge the gap between the statistical community and the potential of social media. Participants will learn how to use social media not just as a tool for sharing research but as a platform to build a supportive network, receive valuable feedback, and contribute to the broader scientific community.

  Participants should bring a laptop and smartphone for interactive exercises.

• Generalized pairwise comparisons: A practical guide to the design and analysis of patient-centric trials
  
  Johan Verbeeck (U Hasselt, Belgium), Brice Ozenne (U Copenhagen, Denmark)

  When assessing the effects of a treatment in clinical trials, often several clinical meaningful endpoints are simultaneously of interest. Yet, standard methods of analysis for multiple endpoints are limited in a number of ways. The generalized pairwise comparisons (GPC) methods form a very flexible class of non-parametric techniques for prioritized endpoints, which overcome all of these limitations. Major advantages are interpretability, the individual-level patient-centric analysis, and good small sample properties.

  In this course, we provide an introduction to the framework of GPC, focusing on practical solutions for the design and analysis of clinical trials with examples in cardiology, oncology and rare diseases, followed by a Q&A for the remainder of the session. Interested participants will be able to apply the GPC methodology on provided datasets with the aid of a user-friendly R package on their own computer, shared through the open science framework platform.

  The course is aimed at statisticians, clinicians and trialists from academia, industry and regulatory agencies with knowledge on clinical trials, who want to learn more about the GPC statistical methodology for multiple endpoints of potentially different data types in large and small sample trials. No specific hardware knowledge is required. Participants are encouraged to bring a personal computer with R pre-installed and the BuyseTest package installed from CRAN.

• Reproducible Research in R: How to Do the Same Thing More Than Once
  
  Aaron Peikert, Maximilian Ernst, Hannes Diemerling (Formal Methods Group at Max Planck Institut for Human Development, Germany)

  Computational reproducibility is the ability to obtain identical results from the same data with the same computer code. The high rate of irreproducible research limits the reach of results and decreases the efficiency of researchers. Reproducible research is a building block for transparent and cumulative science because it enables the originator and other researchers, on other computers and later in time, to reproduce and thus understand how results came about. Many researchers want to work reproducibly, but it is not easy. Considerable time is required to acquire the skills required for reproducible research, and the path is lined with pitfalls. This workshop gets researchers up to speed on how to create reproducible data analyses in R (and beyond). Specifically, researchers learn to automate the whole process from raw data to publishable manuscripts. This automation is possible by combining dynamic document generation (via R Markdown/Quarto), version control (via Git), workflow orchestration (via Make) and software management (via Docker). These tools and, therefore, automatic reproduction of results are available on any machine with Docker installed. The resulting workflow is, hence, highly transferable across machines and time. These core properties of reproducibility are demonstrated for any reader by automatically reproducing the manuscript online.

Afternoon Tutorials

• Enhancing your Code: Combining R and C++ via Rcpp and RcppArmadillo
  
  Erin Sprünken and Lukas Mödl (iBikE, Charité - Universitätsmedizin Berlin, Germany)

  Many researchers rely on simulation studies and complex computations demonstrate properties of their methods. While R is a widely used language for statistical computing, its interpreted nature can limit performance for intensive and repetitive computations. On the other hand, C++, a compiled language, offers faster execution but requires more programming expertise. . This tutorial aims to provide basic knowledge to C++ programming and demonstrates how to integrate C++ code into R using the R packages Rcpp and RcppArmadillo. By leveraging C++'s low-level efficiency, researchers can substantially accelerate their computations. Content: 1. Basics of C++; 2. The Armadillo-Library; 3. Importing C++ Code into R; 4. Practical Examples; 5. Parallelization of C++ Code.

  Participants are required to bring their own notebooks with R, Rtools, and the R-Packages "Rcpp"" and "RcppArmadillo" installed. Participants should have a working knowledge of R programming concepts, such as loops, conditional statements, and user-defined functions.

• Nonparametrics: Some basics and new developments - common misunderstandings, pitfalls, and surprising results
  
  Edgar Brunner (University Medical Center Göttingen, Germany) und Arne Bathke (U Salzburg, Austria)

  What does it actually mean when the hypothesis of a nonparametric test is being rejected? Which alternatives do the different nonparametric and related parametric hypothesis tests really detect? When using inferential methods, it is important to understand the meaning of the so-called consistency region of a test. Misinterpretations might easily happen if this concept is not considered, in particular if the underlying statistical model has restrictions.

  It is the aim of this tutorial to explain the basics, as well as surprising and paradoxical results obtained from different nonparametric procedures – offered in several R-packages – by means of many examples, and to demonstrate possible misinterpretations and misuses of them. We consider (among others) the tests of Mann-Whitney, Kruskal-Wallis, Akritas-Arnold-Brunner, Brunner-Konietschke-Pauly-Puri, Munzel-Hothorn, Jonckheere-Terpstra, Hettmansperger-Norton, van-Eletren, sign-/Wilcoxon-signed-ranks, Friedman, Kepner-Robinson, and Akritas-Brunner.

  Participants are expected to have a solid understanding of statistical inference, including a working knowledge of basic nonparametric, rank-based methods.

• Variable selection and prediction modelling for high-dimensional genomic data
  
  Christian Staerk (TU Dortmund, Germany), Hannah Klinkhammer (IMBIE, U Bonn, Germany)

  High-dimensional biomedical data with many genetic, environmental and clinical variables require scalable and interpretable statistical learning methods for variable selection and prediction. In particular, polygenic risk scores are derived from large cohort studies that include many participants and genetic variants to quantify genetic predispositions for traits and diseases. In the first part of this tutorial, we provide an overview of variable selection and prediction methods for high-dimensional data (large p), including the lasso and its variants, information criteria and boosting. We will also discuss challenges of data-driven variable selection in biomedical research. In the second part, we explore scalable methods for modelling individual-level data from large cohort studies like the UK Biobank, consisting of hundreds of thousands of genetic variants and individuals (large p and n). Specifically, we present frameworks for applying regularized regression via the lasso and statistical boosting for polygenic risk score modelling on large-scale genomic datasets. Our tutorial will include practical exercises in R.

  Participants are asked to bring their laptops with an R installation and ideally also have some prior experience with R.

Full Day Tutorials

• Distributional Regression – Models and Applications
  
  Nadja Klein (Karlsruhe Institute of Technology, Germany) and Lucas Kock (NU Singapur)

  This tutorial, structured into two three-hour blocks, provides an interactive exploration of distributional regression, building on the foundational concepts of generalized linear and additive models. Participants will obtain a review of various distributional regression models and their applications, highlighting the advantages of modeling entire response distributions over traditional mean regression. The session will include hands-on exercises using R, with a focus Bayesian Additive Models for Location Scale and Shape and distributional regression for univariate responses and its extensions to multivariate responses. Through practical exercises and real-world illustrations, participants will gain insights into when and how to apply these models effectively. By the end of the workshop, attendees will have a solid understanding of distributional regression principles and practical skills in model building, estimation, and interpretation.

  Participants should bring their laptop with a running and recent version of R installed. Participants are assumed to be familiar with basic R syntax and have ideally experience with conducting linear or additive regression models in R.

• An introduction to estimands and estimand-aligned estimation
  
  Tim Friede (U Göttingen, Germany), Tobias Mütze (Novartis), Vivian Lanius (Bayer) and Norbert Benda (BfArM, Germany)

  Defining the scientific questions of interest in a clinical trial is crucial to align its design, conduct, analysis, and interpretation. Difficulties in expressing clearly the treatment effects to be estimated led to an addendum to the International Council of Harmonization (ICH) E9 guideline on statistical principles for clinical trials. The ICH E9(R1) addendum on estimands and sensitivity analysis in clinical trials describes a structured framework for precisely defining treatment effect(s) of interest, i.e. the estimand(s), prior to choosing statistical methods aligned with the estimand for the estimation.

• From Theory to Practice: Vine Copula Models
  
  Claudia Czado, Ariane Hanebeck, Ferdinand Buchner (TU München, Germany) and Özge Şahin (TU Delft, The Netherlands),

  As the availability of complex multivariate data continues to grow, understanding dependence structures becomes increasingly important. Copulas play a key role in this exploration. However, standard copula models often fall short when capturing flexible dependence patterns and tail behaviors. To address these limitations, (simplified) vine copula models (vine-copula.org) were developed, offering enhanced flexibility and the ability to model asymmetric tail dependence. In this tutorial, we’ll start with the basics of copulas and vine copulas. We’ll then demonstrate how vine copulas build flexible multivariate regression models, focusing on univariate and bivariate responses. Practical application sessions will use R libraries like rvinecopulib and vinereg to explore real-world scenarios such as probabilistic weather forecasting and financial stress testing. By the end, participants will understand both the theory and practice, equipping them to apply vine copula-based methods in their work.

  Participants are expected to have a basic knowledge of R and RStudio and an interest in statistical modeling and data analysis. Please ensure that you bring a laptop with the latest versions of R and RStudio installed.

• Introduction to Machine Learning with R and mlr3
  
  Bernd Bischl (LMU München, Germany), Michel Lang (UA Ruhr, Germany) and Marvin Wright (BIPS Bremen, Germany)

  Supervised Learning, Performance Evaluation, Hyperparameter Tuning, Benchmarking, Pipelines

• Bayesian Data Analysis
  
  Javier Enrique Aguilar, Luna Fazio (TU Dortmund, Germany) and Maximilian Scholz (U Stuttgart, Germany)

  This workshop provides an introduction to the contemporary Bayesian workflow: participants will learn the core concepts that motivate the use of Bayesian inference, the computational tooling that enables one to work within this framework, and a set of practical guidelines to verify that the resulting models provide reliable inferences. Specific statistical methods that will be covered include basic linear regression, generalized linear models, hierarchical models, and prior specification. We will also discuss visualization and communication of results. Additional specialized topics may be covered based on participant interest and time availability

  Participants should be familiar with basic probability concepts (joint, conditional and marginal probabilities, expectation of a random variable and probability distributions). Prior experience with linear regression will be beneficial. To follow along with the exercises, participants should bring a personal laptop with a recent version of R (4.2 or higher) installed. Ideally, the rmarkdown, tidyverse, rstan and brms packages should also be installed beforehand.

As part of this year’s conference, we are pleased to feature a special exhibition on Emil Julius Gumbel (1891-1966).

Gumbel was a German statistician known for his political activism and pacifism. His work on extreme values and the Gumbel distribution named in his honor continue to be influential in many fields today. Combining his statistical knowledge and political advocacy, he published the book “Vier Jahre politischer Mord (Four years of political murder)” in 1922, showing the judiciary system of the Weimar Republic to be heavily biased in favor of right-wing extremists.

His continued outspoken engagement against fascism led to him being forced out of his tenure at the University of Heidelberg in 1932 after the “Gumbel riots” instigated by the National Socialist German Student League. After Jewish-born Gumbel moved to Paris with his family, his works were subject to Nazi book burnings. In 1940 he escaped the German invasion by fleeing to the US, where he received a professorship at Columbia University in 1953.

Gumbel’s legacy was honored with a Stolperstein, a plaque commemorating the victims of Nazi persecution, in front of his home in Heidelberg in 2021.

In times of increased political division and resurgence of anti-democratic sentiment, it is ever more important to keep the legacy of anti-fascist activists alive. At Humboldt-Universität zu Berlin and Charité - Universitätsmedizin Berlin, we are committed to the values of democracy and respectful coexistence. We are against exclusion and hatred.

The exhibition will be open from 24 March - 14 May 2025. We’re looking forward to sharing this very special experience with you and encourage all conference attendees to join us at:

The scientific committee consists of representatives of the member societies of the German Consortium in Statistics:

• Andrä, Diana (VDSt)
• Behr, Sigrid (DGEpi)
• Beißbarth, Tim (GMDS)
• Binder, Karin (Association for the Promotion of Stochastics Teaching in Schools)
• Boulesteix, Anne-Laure (IBS-DR)
• Brandt, Holger (Methods and Evaluation of DGP)
• Brenzel, Hanna (Destatis)
• Elff, Martin (Methods Section of DVPW)
• Friedrich, Sarah (Secretary General of DAGStat)
• Greven, Sonja (Organization Committee)
• Holling, Heinz (Methods and Evaluation of DGP)
• Holzmann, Hajo (Econometric Committee of the Vereins für Socialpolitik)
• Huscher, Dörte (Organization Committee)
• Ickstadt, Katja (Chairwoman of the DAGStat)
• Kauermann, Göran (GDS)
• Kestler, Hans A. (GfKl)
• Kirch, Claudia (FG Stochastik)
• Kneib, Thomas (Deputy Chairman of the DAGStat)
• Knoth, Sven (DeEnbis)
• Konietschke, Frank (Organization Committee)
• Kopp-Schneider, Anette (IBS-DR)
• Leitgöb, Heinz (Section Methods of Empirical Social Sciences of the DGS)
• Mathes, Tim (Methodology section of the EbM-Netzwerk)
• Münnich, Ralf (DStatG)
• Paetz, Friederike (GfKl)
• Pauly, Markus (FG Stochastik)
• Wilhelm, Adalbert (GfKl)

Sessions will take place in the following buildings of the Humboldt-Universität zu Berlin:

• Main Building UDL6: Unter den Linden 6, 10117 Berlin
• University Building at Hegelplatz: Dorotheenstraße 24, 10117 Berlin
• Department of Library and Information Science: Dorotheenstraße 26, 10117 Berlin

Conference Fees^*

Note: Speakers are asked to register by 31 January 2025

Conference Tool

A Welcome Reception will take place at Thaersaal, Invalidenstraße 42, 10115 Berlin. Please make reservations when registering for the conference.

The conference dinner will take place at the Eventlocation Wasserwerk - on Thursday, March 27, 2025. Please make reservations when registering for the conference.

A Guided Tour is planned "from students for students" in two batches

• on Tuesday, March 25, 2025.
• on Thursday, March 27, 2025.

Berlin, a city rich in history and culture, offers a vibrant tapestry of art and architecture. From its iconic landmarks to hidden gems, Berlin’s streets tell stories of its tumultuous past and dynamic present. Explore the capital’s diverse neighborhoods, each with its own unique flavor, through a variety of guided tours that cater to all interests.

For popular touristic activities, please visit Berlin's official webpage.