Instructor: David Goretzko

Abstract: This workshop introduces the basics of supervised machine learning (ML) focusing on its application in psychological research and assessment. After a general introduction clarifying key concepts, we will discuss a) different resampling strategies to ensure the valid performance evaluation of complex modeling pipelines, b) tree-based supervised learning algorithms as state-of-the-art approaches for tabular data, c) regularization to avoid overfitting and boost generalizability, d) hyperparameter-tuning to optimize the performance of ML algorithms, and e) how to set up benchmarks experiments for model selection. During the course, participants will apply the discussed concepts and modeling strategies through a hands-on exercise in R using the mlr3 ecosystem. The exercise will focus on evaluating various algorithms in a benchmark experiment and optimizing the predictive performance through hyperparameter optimization. Please bring your own laptops and make sure to install all necessary software before the workshop (R, RStudio, mlr3verse – see also https://mlr3book.mlr-org.com/).

Date and time: March 21, 2027, full day from 10:00 to 17:00 

Instructors: Gidon Frischkorn, Philipp Musfeld, Ruhi Bhanap

Abstract: Cognitive measurement models translate observed behaviour (e.g., accuracy, response times, or confidence ratings) into parameters that describe latent cognitive processes (e.g., sensitivity, response bias, or the probability of a memory retrieval). Compared with analysing raw behavioural measures, these models make theoretical assumptions explicit and can separate processes that might otherwise be confounded in a single summary statistic. In practice, fitting such models often requires custom code and specialised statistical knowledge. The bmm package removes these barriers: using common formula syntax in R, it provides a unified interface for estimating a range of cognitive measurement models in a Bayesian framework. This full-day workshop gives a hands-on introduction to the package. It covers the logic of measurement models and how to interpret their parameters, interactive exploration of how parameters shape the observed data, and fitting models to data in R. Participants will learn to test hypotheses about model parameters and to compare the conclusions with those obtained from traditional analyses of the behavioural data. Participants should be comfortable importing and manipulating data in R; some familiarity with Bayesian linear models is helpful but not required. Participants should bring their own laptop.

Date and time: March 21, 2027, full day from 10:00 to 17:00 

Instructor: Daniel Heck

Abstract: Multinomial processing tree (MPT) models provide a powerful tool for disentangling different cognitive processes contributing to the same observable responses. In many fields of psychology such as memory, decision making, and social cognition, MPT models have been successfully used to test hypotheses concerning specific processes via separate model parameters. First, the workshop covers the logic of MPT modeling, including identifiability, parameter estimation, and goodness of fit. Second, participants will develop a new MPT model together, starting from a research question and an experimental paradigm. Third, the workshop introduces hierarchical Bayesian MPT models, which account for differences between participants by assuming a group-level distribution of the individual parameters. Finally, participants will practice fitting hierarchical MPT models with the R package TreeBUGS, which includes model fitting, convergence checks, model fit, and interpretation of parameters. 

Prerequisites are: Solid background in statistics (master's level),  basic skills in using R for data analysis, notebook with the necessary software installed (R, RStudio, JAGS, TreeBUGS, multiTree; instructions will be sent before the workshop).

Date and time: March 21, 2027, time TBA

Instructor: Stefan Radev

Abstract: What if we could build statistical models around our psychological theories, rather than adapting our theories to the limitations of computational methods? After more than a decade of research, amortized Bayesian inference now opens up exciting possibilities for modeling virtually any psychological process, allowing our questions and creativity to guide model development. In this hands-on workshop, participants will learn how to develop, fit, and evaluate models of behavior using the BayesFlow framework. First, we will cover the basic principles of modern simulation-based inference and the software ecosystem. Through practical examples, we will explore how to translate increasingly complex psychological models into simulators and run in silico experiments; we will then train neural networks to quickly estimate parameters from data, test predictions, and refine the models. We will also introduce agentic AI workflows that close the loop between model development, model fitting, and model validation. Using a principled Bayesian workflow as a verification blueprint, we will customize our own agents to enable open-ended, model-based science grounded in domain knowledge and decision-theoretic criteria.

Date and time: March 21, 2027, full day from 10:00 to 17:00

Instructors: Nathan J. Evans, Amir Hosein Hadian Rasanan, Niek Stevenson, and Lukas Schumacher

Abstract: Continuous judgments are central to experimental psychology, ranging from perceptual estimation to probability judgments and economic valuation. Many traditional theoretical accounts of judgments focus primarily on response distributions without explicitly explaining response times. Yet response times provide additional information about the underlying cognitive processes, helping distinguish mechanisms that produce similar judgment distributions. Evidence accumulation models offer an established framework for jointly explaining responses and response times. These models describe judgment formation as the accumulation of noisy evidence until a stopping criterion is reached. Although originally developed for discrete choices, researchers have recently extended them to continuous judgments on wrapped circular, bounded, and two-dimensional response scales. This full-day workshop introduces the theoretical and computational foundations of evidence accumulation models of continuous judgment. We examine two complementary timescales: how evidence accumulates within an individual judgment and how cognitive processes change across successive judgments. Four sessions combine theoretical introductions with practical demonstrations and hands-on examples. Participants will first learn the principles of cognitive modeling and multidimensional diffusion models. They will then use JEAM in Python to specify models, simulate judgments and response times, estimate parameters using maximum likelihood, and translate psychological hypotheses into computational models. A session using EMC2 in R will introduce hierarchical Bayesian estimation, convergence assessment, and posterior predictive checks. Finally, participants will use SuperStats in Python to investigate changes in cognitive parameters across trials and learn how simulation-based inference can estimate models whose likelihoods are unavailable for direct evaluation or computationally intractable.

Date and time: March 21, 2027, full day from 10:00 to 17:00

Instructor: Tanja C. Roembke

Abstract: Words and pseudowords are everywhere in cognitive psychology. They appear in studies of memory, language, and reading, as well as across many other experimental contexts (e.g., as cover tasks). Generations of psycholinguistic researchers have investigated how characteristics of words and pseudowords influence human cognition, cumulating in a multitude of findings of interest not just to language researchers, but also to the broader field of psychology and cognitive science. In recent years, we, the TRUST network, collated a list of recommendations for research into word and pseudoword processing into a collection of consensus-based guidelines (https://trust-guidelines-v01.readthedocs.io/). We went on to apply these recommendations in a large-scale multi-lab study, across more than 20 research sites, on word and pseudoword processing (the German Lexicon Project [GLP]; gewonn.github.io/trust/glp/). This yielded an open dataset with lexical decision data from over 2,400 participants and 77,500 items to date. Together, the guidelines and GLP form valuable resources for understanding words and pseudowords, and how they are processed. In this three-hour workshop, we introduce the TRUST Guidelines and the GLP from two complementary perspectives: how to build a large-scale, collaborative research resource and how to use the GLP for one's own research. We discuss the motivation for developing the guidelines, the consensus-building process, and key methodological decisions involved in implementing the GLP. In hands-on sessions, participants explore the available data files and learn how to use GLP data to address novel research questions through virtual experiments. We also demonstrate how to select and control experimental stimuli by filtering and matching on relevant lexical characteristics, including applications beyond word processing research. We conclude with lessons learned from organizing large-scale, multi-lab collaborations using open science practices.

Date and time: March 21, 2027, half day from 13:00 to 17:00

Instructor: Dirk Wulff

Abstract: TBA

Date and time: TBA