The digital revolution has profoundly transformed the landscape of research in the social and behavioral sciences, driven in part by the emergence of novel forms of data such as process data, mobile sensing streams, and digital trace records. These data sources capture behavior in real time and at fine temporal resolution, often yielding datasets in which the number of variables far exceeds the number of participants—so-called high-dimension low-sample-size (HDLSS) data. At the same time, research paradigms have increasingly shifted toward multi-disciplinary approaches. Behavior and cognition are now studied not only from a psychological perspective, but also through environmental, social, clinical, and biomolecular lenses. This convergence has created both exciting opportunities for understanding psychological processes, but they also challenge the foundations of traditional statistical and psychometric modeling.

A wide range of high-dimensional data types is now routinely encountered in educational and psychological research. These include, for example, multimodal neuroimaging data, high-frequency behavioral, and physiological data from mobile devices, textual and language-based data, and process data derived from digital environments such as eye-tracking or virtual navigation tasks. Each of these domains offers new ways to observe and measure complex human processes, but also introduces unique analytic complexities. As such data sources continue to proliferate, there is an increasing need for frameworks that can meaningfully integrate and interpret them.

Incorporating high-dimensional data into social and behavioral science research presents a shared set of challenges. Because many of these data types are relatively recent, there is often limited theoretical groundwork to guide their analysis. This creates a pressing need for methods that not only accommodate high dimensionality, but also support theory development through interpretable, explanatory models. At the same time, predictive modeling has gained prominence due to its practical relevance in areas such as treatment outcome prediction and decision-making. However, prediction in HDLSS contexts requires careful methodological consideration to avoid overfitting and to ensure generalizability. Furthermore, many of these data sources are unstructured or semi-structured and exhibit complex dependencies, such as temporal, spatial, or hierarchical structure, which must be explicitly modeled.

Another major challenge arises from the integration of multi-view, multi-modal, or multi-domain data collected on the same individuals. Combining information across disparate sources—each potentially high-dimensional in its own right—raises questions about alignment, scaling, and the preservation of meaningful structure. Addressing these challenges often requires drawing on a diverse set of analytical approaches, including transformation of raw digital data into meaningful variables, dimension reduction and latent variable techniques, variable selection methods, natural language processing, spatiotemporal modeling and functional data analysis, and machine learning methods for regression and classification.

Importantly, the development and application of methods for high-dimensional data extend beyond psychometrics and quantitative psychology. Related advances are being made in fields such as statistics, bioinformatics, computer science, and chemometrics, among others. As a result, a multi-disciplinary perspective is not only beneficial but often essential for both methodological innovation and substantive application. This special section brings together contributions that reflect this diversity, showcasing how integrative approaches can advance the analysis of complex data in social and behavioral research.

A central insight emerging from this collection is that high-dimensional data are not merely larger versions of traditional datasets. Instead, they require new ways of thinking about representation, integration, and inference. The contributions in this issue address these challenges through complementary perspectives, and all address some combination of these five overarching themes:

1. 1. modeling high-dimensional and complex response structures,

2. 2. integrating heterogeneous and multimodal data,

3. 3. analyzing process and behavioral trajectory data,

4. 4. transforming raw digital data into meaningful constructs, and

5. 5. conducting valid inference and variable selection under data complexity.

The nine articles included in this special section collectively illustrate the breadth of contemporary approaches to integrating and analyzing complex high-dimensional data in the social and behavioral sciences. Several contributions focus on intensive longitudinal and mobile data, where both the temporal structure and dimensionality of the data present unique challenges. For example, McBride & Wang (Reference McBride and Wang2026) introduce a hierarchical Gaussian process framework embedded within an item response theory model to study mental health trajectories using repeated mHealth assessments. By combining Bayesian estimation, Hilbert space approximations, and flexible Gaussian process priors, this work demonstrates how high-dimensional ordinal data collected at irregular time intervals can be modeled in a way that captures both individual and group-level dynamics while maintaining interpretability.

Complementing this focus on mobile health data, Schoedel et al. (Reference Schoedel, Sust, Sterner and Goretzko2026) examine the preprocessing of mobile-sensing data derived from smartphones, such as app usage logs and time-stamped behavioral events. Rather than proposing a single modeling framework, this manuscript emphasizes the critical role of integrative preprocessing pipelines in transforming raw, high-frequency, high-dimensional data into psychologically meaningful variables. Through a series of case studies, it highlights methodological strategies for data enrichment and aggregation, including the incorporation of contextual information and the use of machine learning techniques, thereby addressing a key but often underdeveloped stage in the analysis of complex behavioral data.

Sort et al. (Reference Sort, Le Brusquet and Tenenhaus2026) address the analysis of multidimensional longitudinal data structured as tensors, which are increasingly common in both psychometric and biomedical applications. Building on the CANDECOMP/PARAFAC (CANonical DECOMPosition / PARAllel FACtor analysis) decomposition, the authors propose a latent functional extension that incorporates smooth functional structure along the temporal mode while accounting for stochastic variability through a probabilistic framework. This approach enables the representation of high-dimensional tensor data in a lower-dimensional space, facilitating interpretation and analysis even under sparse and irregular sampling conditions. The application to neurocognitive data from the Alzheimer’s Disease Neuroimaging Initiative illustrates the utility of combining tensor decomposition, functional data analysis, and latent variable modeling in complex real-world settings.

In a complementary application of functional data analysis, Zhang et al. (Reference Zhang, Shuster, Morehouse, Mednick, Yu, Shen and Simon2026) explore high-dimensional behavioral data generated in a virtual environment, specifically 3D path data from a navigation task implemented in Minecraft. The authors develop a methodology to transform raw spatial trajectories into functional representations, enabling the application of functional clustering and regression techniques. This work highlights how spatiotemporal modeling and functional data analysis can be used to uncover meaningful patterns in complex movement data, linking navigation strategies to cognitive outcomes such as memory retention and spatial learning.

Fang et al. (Reference Fang, Wang, Chen, Zhang, Liu and Zhong2026) address the growing importance of process data in educational and psychological assessment, where rich, temporally ordered information is collected alongside traditional outcome measures using an application to a spatial rotation task. They propose a two-component data analytic framework that integrates structured assessment data with unstructured eye-tracking sequences. The first component uses a time-aware Long Short-Term Memory (LSTM) autoencoder to extract latent features from fixation locations, durations, and elapsed times, thereby capturing both spatial and temporal dependencies in high-dimensional process data. The second component applies statistical techniques—including clustering, contingency analysis, and mixed-effects modeling—to interpret these features and relate them to testing and learning behaviors.

Helwig et al. (Reference Helwig, Chen, Guy and Lyford-Pike2026) focus on the challenges of modeling high-dimensional multilevel data with complex response formats, introducing a regularized multilevel multinomial regression framework for select-all-that-apply (SATA) responses. Their approach combines penalization techniques with hierarchical modeling to accommodate both high-dimensional predictors and clustered data structures arising from repeated measurements. Using facial expression perception data, the authors integrate statistical shape analysis and dimension reduction to construct interpretable predictors, while regularization enables stable estimation in a high-dimensional setting. This work provides a compelling example of multilevel high-dimensional predictive modeling, demonstrating how flexible regression frameworks can be extended to modern data contexts without sacrificing interpretability.

Li and Biesanz (Reference Li and Biesanz2026) extend classical psychometric theory into high-dimensional settings through the development of the Double Machine Learning Lens Model (DML-LM). Building on the traditional lens model of perceptual accuracy, their framework incorporates machine learning techniques to handle large sets of predictors such as text embeddings. In an application to social class perception from thousands of essays, the authors systematically compare a wide range of modeling pipelines and show that regularized linear models applied to dimension-reduced embeddings provide strong predictive performance while retaining interpretability. This work highlights how integrating machine learning with established theoretical frameworks can advance both predictive modeling and psychological theory.

Ackerman et al. (Reference Ackerman, Zhang, Hannig, Prothero and Marron2025) address the integration of multiple high-dimensional data blocks, proposing the DIVAS (Data Integration via Analysis of Subspaces) framework. DIVAS decomposes variation into fully shared, partially shared, and individual components, enabling a nuanced understanding of relationships across heterogeneous data sources. Applied to data from the Human Connectome Project, the method jointly analyzes structural and functional brain connectivity, cognition, substance use, and genetic information, revealing interpretable patterns of shared variation across modalities. In addition, the authors develop new inferential tools based on Jackstraw procedures. Like the functional PARAFAC approach of Sort et al. (Reference Sort, Le Brusquet and Tenenhaus2026), this work emphasizes dimension reduction as a central tool for managing high-dimensional data, but extends it to the context of multi-block data integration.

Bacci et al. (Reference Bacci, Dreassi, Grilli and Rampichini2026) focus on the problem of variable selection and inference in high-dimensional data with missing values and categorical predictors, extending the knockoffs framework to more realistic data settings commonly encountered in social science research. Their approach combines multiple imputation with knockoff-based variable selection, enabling control of false discoveries while accommodating complex predictor structures. The motivating application involves large-scale educational assessment data with a multilevel structure of students nested within schools. This contribution complements the work of Helwig et al. (Reference Helwig, Chen, Guy and Lyford-Pike2026) by providing a second example of high-dimensional modeling in multilevel contexts, but with a distinct focus on valid statistical inference and feature selection rather than predictive modeling alone. These two articles illustrate different strategies for addressing the challenges of high-dimensional data in hierarchical settings.

Together, these contributions underscore the diversity of application domains—ranging from mental health and mobile sensing to neuroimaging and virtual environments—and demonstrate the central role of advanced statistical and machine learning methods in extracting insight from high-dimensional data.