Why and How to Model Trends and Time-Varying Experimental Effects Using Mixed-Effects Models: A Simulation Study and a Tutorial
Abstract
Data from repeated-measures experiments (e.g., response time data, electrophysiological data) can show systematic changes in outcomes over time, so-called (temporal) trends. When analyzing experimental effects between conditions, different trends per condition can cause the experimental effect to vary with time, for example, to decline over the course of the experiment. Previous research has shown that ignoring time-varying effects in an analysis can influence results and distort conclusions, whereas explicit modeling yields more informative results and valuable insights into substantive theories about the underlying (psychological) processes. This article provides a systematic introduction to modeling trends and time-varying experimental effects in experimental data, accounting for fixed trends (across participants) as well as random (person-specific individual) trends. We consider different modeling approaches: Linear or polynomial growth curve models, generalized additive mixed models (GAMMs) and the recently proposed random-effects-expectation-maximization (RE-EM) trees. We first provide an overview of these different modeling approaches and their potential merits. We then compare their performance and investigate each models data requirements in two simulation studies based on the characteristics of realistic data from a cognitive response time experiment. We derive specific recommendations for their application from our results, provide a tutorial on how these models can be applied in the open source software R, and explain how their results can be interpreted along with presenting annotated R-code. We conclude with a recommended workflow on how a combination of the investigated methods can help researchers better understand time-varying effects.
Related
- Occurrence, Severity and Temporal Trends of Enzootic Pneumonia-like Lung Lesions in Hungarian Slaughter Pigs, 2018-2022: Insights from Zero-Inflated Mixed-Effects Models.
- An experimental study on HRM combined with multiplex PCR for the identification of single and mixed diatom species
- High-power pulsed-laser ablation of aluminum alloy: Simulation model and experimental validation
- INFORME: coupling information-theoretic experimental design with nonlinear mixed-effects modeling for efficient observation scheduling
- Design, simulation and experimental study on elliptical-groove spiral intermittent precision fertilization device for maize intermittent banding fertilization
BibTeX
@misc{keintzel2026model,
title = {{Why and How to Model Trends and Time-Varying Experimental Effects Using Mixed-Effects Models: A Simulation Study and a Tutorial}},
author = {Salome Li Keintzel and Anna Nikolei and Florian Scharf},
howpublished = {PsyArXiv},
year = {2026},
month = sep,
doi = {10.31234/osf.io/qzut8_v1},
url = {https://osf.io/preprints/psyarxiv/qzut8_v1/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Why and How to Model Trends and Time-Varying Experimental Effects Using Mixed-Effects Models: A Simulation Study and a Tutorial
AU - Salome Li Keintzel
AU - Anna Nikolei
AU - Florian Scharf
PY - 2026
DA - 2026/09/05
PB - PsyArXiv
DO - 10.31234/osf.io/qzut8_v1
UR - https://osf.io/preprints/psyarxiv/qzut8_v1/
AB - Data from repeated-measures experiments (e.g., response time data, electrophysiological data) can show systematic changes in outcomes over time, so-called (temporal) trends. When analyzing experimental effects between conditions, different trends per condition can cause the experimental effect to vary with time, for example, to decline over the course of the experiment. Previous research has shown that ignoring time-varying effects in an analysis can influence results and distort conclusions, whereas explicit modeling yields more informative results and valuable insights into substantive theories about the underlying (psychological) processes. This article provides a systematic introduction to modeling trends and time-varying experimental effects in experimental data, accounting for fixed trends (across participants) as well as random (person-specific individual) trends. We consider different modeling approaches: Linear or polynomial growth curve models, generalized additive mixed models (GAMMs) and the recently proposed random-effects-expectation-maximization (RE-EM) trees. We first provide an overview of these different modeling approaches and their potential merits. We then compare their performance and investigate each models data requirements in two simulation studies based on the characteristics of realistic data from a cognitive response time experiment. We derive specific recommendations for their application from our results, provide a tutorial on how these models can be applied in the open source software R, and explain how their results can be interpreted along with presenting annotated R-code. We conclude with a recommended workflow on how a combination of the investigated methods can help researchers better understand time-varying effects.
ER -