Emotions are dynamic entities, following the ebb and flow of daily life. Dynamic patterns reflect the habitual emotional responses of an individual to the environment. Sudden shifts in the way emotions fluctuate might be informative about the psychological health of an individual. To fully understand emotions, it is therefore important to get a grip on their dynamic characteristics.In psychology, the dynamic nature of emotions has been stressed in various process models. However, sophisticated statistical modeling of dynamic emotion processes holds some complex challenges. In this dissertation, we would like to offer researchers some statistical tools and guidelines for tackling these challenges.We start from existing dynamic modeling techniques in psychology and other disciplines. Our goal is to use these dynamic models for answering substantive questions in the literature on emotion. While applying the models, we should take into account the particular data characteristics. All our applications are based on Bayesian statistics, a statistical ideology rooted on Bayes theorem that uses distributions to estimateparameters.The dissertation consists of a general introduction and three stand-alone chapters.In chapter 2, we discuss linear dynamical system theory and state space modeling. In the state space approach, separate equations are formulated for observed and latent data. This technique is commonly used in engineering, in for instance the development of software for satellites and global positioning systems. Because of its generality and flexibility, the state space approach seems to be useful for analyzing psychological data. To take into account the hierarchical data structure, we introduce a hierarchical Bayesian implementation and apply it to physiological time series.In chapter 3, transdimensional model selection is introduced. In this model selection ideology, a model indicator is formulated as a parameter that can change values, hereby switching between different models. Following the Bayesian framework, prior information about the model probabilities is defined, and posterior probabilities are estimated, taking the observed data into account. Two examples of transdimensional techniques are reversible jump MCMC and the product space method. The reason why this way of model selection never gained popularity is because of certain implementational difficulties. In the chapter, we discuss the product space method and highlight how it should be used to obtain optimal performance. The technique is illustrated with a multiple-model selection problem in modeling circadian rhythms of emotions. To emphasize the flexibility of the method, the application is completed with other data examples in psychology.In chapter 4, we investigate the issue of physiological emotion specificity. This comprehensive research domain handles about the relation between physiological and experiential/behavioral components of emotion. In other words, the domain investigates whether and how our physiological systems are affected by our experienced emotional states. We introduce a conceptual framework that toghether graps the crucial ideas that are discussed in the domain. Further, we translate the concepts into a regime-switching time series model and apply it to data. Also here, a group-level analysis is used.
Table of Contents:
2. A hierarchical state space approach to affective dynamics
3. A tutorial on Bayes factor estimation with the product space method
4. An integrative framework for physiological emotion specificity