Abstract: In the network approach to psychopathology, disorders are sets of causally connected symptoms. This conceptualization offers novel perspectives on the theoretical status of mental disorders: instead of cleanly separable categories that reflect central neural or psychological deficits, disorders are tightly connected regions in a symptom network. This conceptual framework suggests novel approaches to both the analysis of research data and the organization of treatment interventions, primarily through the application of network analysis: a set of techniques that offers powerful tools to study the dynamics of interconnected systems, to analyze the architecture of networks involving large numbers of entities (e.g., neurons, people, genes, variables), and to visualize connectivity structures in such networks. In the present talk, I will give an overview of the most important insights and results that have arisen from the network approach.

Abstract: In this presentation, I will present an out-of-the-box methodology applicable to longitudinal psychological data for exploratory discovery of relationships between observed measures. This framework takes the form of the well-known (multi-level) vector-autoregression model (VAR), but extends the emphasis beyond the temporal coefficients typically interpreted as a directed network. The VAR model can be seen to extend the increasingly popularly used Gaussian Graphical Model (GGM)---a network of partial correlation coefficients---to data in which cases are not independent. In addition to temporal networks, the methodology also returns a contemporaneous network and a between-subjects network, both in the form of a GGM. During this presentation, I will discuss the plausibility of assumptions made and the potential causal interpretation of contemporaneous and between-subject network structures. In addition, two R packages will be introduced for estimating these structures: graphicalVAR, which uses LASSO regularisation to estimate the temporal and contemporaneous network of a single subject, and mlVAR, which uses multi-level modeling to estimate the temporal, contemporaneous and between-subjects networks of multiple subjects. I will show empirical examples of both methods.

Abstract: The most dominant approach to model observable behaviour in psychology makes use of latent variables. That is, it is assumed that the observable behaviour is caused or governed by an unobserved attribute (e.g., `intelligence', `ability', `disorder', etc.). Even though it is a highly successful approach, the theoretical status of latent variables is often challenged. Recently, it was proposed to model observable behaviour directly using network models, instead of using latent variables. That is, it was suggested that observed variables directly influence each other (e.g., `sleep problems' --- `concentration problems'). Such models have been applied successfully to psychopathology and personality data in the last years, for instance. Since psychometric (latent variable) models are formally related to network models, it follows that both types of models show promise when applied to the same question. In this respect, the latent variable and the network model approach can be seen as being two sides of the same coin, allowing us to gain new insights in to existing problems, and new problems for existing insights. One area where latent variable and network models have been less successful is in modelling qualitative inter-individual differences of psychological processes. For example, models have difficulties explaining why specific people who experiencing the same stressor (e.g., an adverse life-event) develop a depressive episode while others do not. Here we propose that such processes can be better explained using random cluster models, where each individual comes with their own network. That is, we propose that the network structure itself is a latent variable.

Abstract: Due to technological developments (e.g., smartphones), there is an enormous increase in studies based on daily diaries, ecological momentary assessments, ambulatory assessments, and experience sampling methods. The intensive longitudinal data stemming from these studies provide us with the unique opportunity to investigate the dynamics of psychological processes as they are unfolding over time. This can be done by using single-subject time series models, or by using new multilevel models where level 1 is formed by a time series model, while at level 2 individual differences in the time series parameters are modeled. Currently, the software package Mplus is being extended with Dynamic Structural Equation Modeling (DSEM), which will allow for N=1 time series modeling, as well as its multilevel extensions. Furthermore, it will also allow for regime-switching processes. In this talk I will provide a bird?s eye view of these exciting new developments. I will briefly present the general DSEM framework and show a few applications consisting of multilevel vector autoregressive models and latent multilevel autoregressive models. Additionally, I will touch upon some of the major challenges in this rapidly developing area, including how to standardize parameters in these models to allow for meaningful comparisons among them, and whether ESM data should be considered as 2-level or 3-level data.

Abstract: Emotion dynamic research typically aims at revealing distinct information on affective functioning and regulation. Herewith, one distinguishes various elementary emotion dynamic features (EDFs), which are studied using intensive longitudinal data. Typically, each EDF is quantified separately, which seriously hampers the study of relationships between various features. We propose a Bayesian vector autoregressive model (VAR) and apply it to emotion data. The model encompasses all six emotion dynamic features central in emotion research at once, and can be applied with relatively short time series, including missing data. The model can be applied to both univariate and multivariate time series, allowing to model the relationships between emotions. Further, it may model multiple individuals jointly as well as external variables and non-Gaussian observed data, and can deal with missing data. We illustrate the usefulness of the model with an empirical example using relatively short time series of three emotions, with missing time points within the series, measured for three individuals. Finally, we demonstrate that the model can easily deal with measurements that are not equally spaced in time.