Mirka Henninger, University of Basel

In my talk, I highlight key opportunities to develop novel methodological tools and gain insights into psychological processes by working at the intersection of psychometrics, multilevel and longitudinal modeling, and machine learning.

In the first part of the talk, I focus on the measurement of intensive longitudinal data (ILD). ILD are widely used to study affect dynamics, including intraindividual variability and temporal dependencies (inertia). However, because these data typically rely on self-reports, they may be systematically biased by stable response styles, such as extreme response style (ERS). Drawing on IRTree and location-scale modeling approaches, we examined the association between ERS and affect dynamics using data from a controlled experiment (N = 1,389) and multiple experience sampling studies conducted in daily life (total N = 1,254). Across datasets, higher ERS was consistently associated with greater intraindividual affect variability, but not with moment-to-moment affective inertia. These findings suggest that commonly used indicators of affective variability may partially reflect systematic response behavior rather than exclusively genuine affect dynamics. Accounting for response styles is therefore crucial when modeling and interpreting affect dynamics in ILD.

In the second part, I turn to the analysis of multilevel or longitudinal data using decision trees from machine learning. Decision trees are flexible and powerful tools for capturing complex interactions and nonlinear effects. Recent extensions aim at adapting these methods to multilevel data by incorporating random effects structures. Through simulation studies, we demonstrate that existing multilevel tree approaches exhibit inflated type-1 error rates for cluster-level predictors. Moreover, we show that predictor-specific model predictions are substantially distorted, leading to false positive conclusions about variable importance and model predictions.

I conclude by outlining future directions for integrating psychometric modeling and machine learning into measurement and analysis of multilevel and longitudinal data.

About the Speaker

Mirka Henninger is an assistant professor for Statistics & Data Science at Faculty of Psychology of the University of Basel, Switzerland. Her background is in psychology, and her research is grounded at the intersection of psychometrics, machine learning, and multilevel modeling. The overarching goal of her research is to strengthen confidence in psychological science by ensuring that empirical conclusions are based on methods that are technically sound, interpretable, and theoretically informative.