< Back to previous page


Regression approaches for modeling affect dynamics: Leveraging on context and serial dependence

The field of affect dynamics studies how affect processes evolve over time. By collecting many repeated measurements of such processes over time, rich time series of intensive longitudinal data (ILD) become available. These data are analyzed in order to quantify features which could shed light on an individual’s affective functioning. To do so requires the formulation of a model, and the use of statistics to estimate model parameters. An interesting starting point is the autoregressive (AR) model. It allows for some key features typically hypothesized to be present in time series of affect variables, such as serial dependence and regulatory behavior towards an emotional homebase. 

The idea that we additionally need to include context in our models of affect dynamics has gained traction in recent years. Contextual influences can be thought to perturb the affective system, such as daily hassles might do, without altering the dynamics in any fundamental way. However, other contextual influences (e.g. major events) could alter the dynamics themselves. In such cases, the parameters of the models can be allowed to vary over time. A popular way of doing so is by tying changes in parameters to changes in a time-varying covariate, reflecting contextual information. However, little is known about the factors influencing how accurately such effects can be estimated, how to design studies in order to estimate the effects accurately, and how to appropriately handle both context and serial dependence. 

The dissertation is divided into 6 chapters. Chapter 1 situates the dissertation research in the context of the broader state of the art. Chapter 2 provides an overview on the analysis of psychological ILD by means of AR modeling. We focus on challenges faced by applied researchers, and how the literature has proposed to handle these challenges. In Chapter 3, we study the patterns of covariance implied under AR models with linear contemporaneous or lagged contextual covariate effects. We show that the temporal behavior such covariates have implications for estimation accuracy, and consequently study-design topics such as sample size planning. In Chapter 4, we present a critical evaluation of typically employed approaches to accommodating serial dependence in regression models. We show how these approaches are related, and how researchers can make informed decisions on how to accommodate the serial dependence. In Chapter 5 we study how to modify experience sampling designs to leverage on moments of increased affective variability. Sampling during and following emotional episodes can have beneficial implications for the accuracy of the least squares AR effect estimator, but a number of important characteristics of such a design will influence how this plays out. Chapter 6 contains a discussion and provides directions for future research. 

Date:1 Oct 2019 →  Today
Keywords:Psychological dynamics, Time series analysis
Disciplines:Mathematical psychology
Project type:PhD project