Ecological momentary assessment
Statistical models for intensive longitudinal data, dynamic structural equation models, and latent processes in psychology.
Professor of Statistics
Øystein Sørensen
Department of Psychology, University of Oslo
My main research interests involve developing scalable and trustworthy algorithms and software for analysis of large and complex data in the cognitive sciences. I am also increasingly interested in understanding how generative AI best could be put to use both in algorithm development and as a data science companion.
My home is at the Center for Lifespan Changes in Brain and Cognition.
I am also:
- Affiliated researcher at Integreat - Norwegian Centre for Knowledge-Driven Machine Learning
- Project manager for NORA Tech Research School, which you definitely should join if you are doing a PhD in Norway.
- Senior researcher (20% position) at RBUP Øst og Sør.
Research focus
Bayesian computation, hierarchical models, intensive longitudinal data, personalized medicine.
Research
Research Themes
Computational statistics for complex models
Efficient algorithms using state-space representations, sparse matrix methods, automatic differentiation, and Bayesian computation.
Generalized additive latent and mixed models
Semiparametric methods for nonlinear trajectories, mixed response data, and multilevel psychological measurement.
Preference learning and ranking data
Bayesian models and sequential Monte Carlo methods for rankings, pairwise preferences, and partial order data.
Machine learning in clinical psychology
Explainable personalized predictions.
Publications
Selected Publications
2026
A Hybrid NUTS-Gibbs Sampler with State Space Marginalization for Estimation of Dynamic Structural Equation Models with Binomial Outcomes
This paper presents a hybrid NUTS-Gibbs sampler for dynamic structural equation models with binomial outcomes. The Gibbs step handles Pólya-Gamma latent variables from a logit link, while the NUTS step uses a Kalman filter to marginalize over latent states. arXiv preprint.
2026
A Semiparametric Nonlinear Mixed Effects Model with Penalized Splines Using Automatic Differentiation
We present an estimation procedure for nonlinear mixed-effects models in which the population trajectory is represented by penalized splines and adapted to individuals via subject-specific transformation parameters. Exact derivatives are obtained via automatic differentiation implemented through Template Model Builder. arXiv preprint.
2026
Efficient Bayesian Estimation of Dynamic Structural Equation Models via State Space Marginalization
This paper shows that the within-level part of any dynamic structural equation model can be reformulated as a linear Gaussian state space model, enabling analytical marginalization via a Kalman filter and highly efficient estimation via Hamiltonian Monte Carlo. arXiv preprint.
2025
Gaining Brain Insights by Tapping into the Black Box: Linking Structural MRI Features to Age and Cognition using Shapley-Based Interpretation Methods
This paper evaluates multiple interpretability techniques for machine learning models applied to neuroimaging data, including SHAP and SAGE. We trained XGBoost models to predict age and fluid intelligence using UK Biobank data and found that subcortical mean intensities are associated with brain aging, while fluid intelligence prediction is driven by the hippocampus and cerebellum. Published in Neuroinformatics.
2025
Modeling Cycles, Trends and Time-Varying Effects in Dynamic Structural Equation Models with Regression Splines
This paper considers rank and preference modeling for the case in which data arrive sequentially, rather than in a batch. The goal is to compute the posterior distribution incrementally in time, so that it can be quickly updated when new data arrives. To this end, we develop a sequential Monte Carlo algorithm for the Bayesian Mallows model. arXiv preprint currently under revision.
Software
Open-Source Software
BayesMallowsSMC2: Nested Sequential Monte Carlo for the Bayesian Mallows Model
This R package provides nested sequential Monte Carlo algorithms for performing sequential inference in the Bayesian Mallows model.
galamm - Generalized Additive Latent and Mixed Models
This R package contains functions for estimating generalized additive latent and mixed models.
BayesianLaterality - Predict Brain Asymmetry Based on Handedness and Dichotic Listening
This R package contains functions for estimating whether a subject is left- or right-dominant for language processing based on the results of dichotic listening experiments and information about handedness.
metagam: Meta-Analysis of Generalized Additive Models
This R package contains functions for meta analysis using generalized additive (mixed) models, by combining fits from multiple studies.
Teaching
Teaching
At the Department of Psychology, I teach quantitative methods at bachelor, master and PhD level.
I have also occasionally given courses in statistics, machine learning, and artificial intelligence, with a particular focus on its use in clinical psychology and behavioral sciences. If interested, you are welcome to get in touch.
Contact
Contact
Email: oystein.sorensen@psykologi.uio.no
Office: Department of Psychology, University of Oslo
Oslo, Norway