Prof. Eva Ceulemans & Prof. Iven Van Mechelen
On March 17th 2016, Joke Heylen defended her thesis entitled
Modeling multilevel time-resolved emotion data
Nowadays, many research questions imply studying time-resolved data. For example, the time dynamics of emotions is a hot topic; hence, one recently has started gathering data on the intensity of different emotion components (e.g., appraisals, physiological features, subjective experience) at several time points during an emotion episode. Given these data, it is important to capture the different shapes that the time profiles may take and how these shapes depend on episodes’ characteristics, person traits, and on the type of emotion component under examination. The latter implies two major methodological challenges.
First, we need to find out which method is best suited to gain insight into these shapes. Two classic strategies are functional component analysis (based on dimension reduction) and clustering approaches (implying categorical reduction of the time profiles). Since both strategies have some drawbacks, we intend to develop extensions that combine the attractive features of both.
Second, a proper solution to the problem of time alignment is required, which pertains to differences in shift and to stretching or contracting of the time axis. Although some functional models have been developed to deal with alignment issues (e.g., shifted and warped factor analysis), these methods have to be extended to deal with the inherent multilevel structure of the data under study.
Therefore, the goal of this project is to build new clustering and dimension reduction models for multilevel time-resolved emotion data that allow for shifting and/or warping, and to develop algorithms and model selection procedures for fitting these models to empirical data.