The recent trend in complexity theory brought many researchers to conceptualize phenomena as complex dynamical systems, self-organized and governed by strange attractors. It fundamentally renewed the traditional approaches for analyzing change that are longitudinal studies and ARIMA-like time series analysis. But psychologists who analyze data from such standpoint face enormous difficulties, partly due to a lack of an adequate mathematical background, constraining methodological requirements and limited availability of computer software.

We present in this thesis a fruitful alternative for exploring and analyzing time series data: the dynamical configural approach. Recognizing the inherently multivariate and coarse-grained nature of psychological data, we developed a coherent set of methods that graphically and statistically describe the dynamics of a system. The proposed methods integrate both the micro- (univariate) and macro- (multivariate) perspective.

Prototypical themes and questions related to the analysis of time series data are first reviewed: patterns of transitions, temporal (in)dependence, cycles, higher-order influences, phases, and complexity. Statistical methods dealing with each of these are detailed, exemplified with experimental data drawn from the author’s previous researches. Novel graphical techniques such as Karnaugh maps and evolustrips were specially developed for the visual exploration of configurations. Statistical analyses are mainly performed through entropy-based statistics, drawn from information theory. Loglinear models, as represented in graphical models, allow to answer most of the time series questions. Fruitful perspectives on the analyses of the complexity and dynamics of psychological systems are finally discussed.

Keywords: time series analysis, complex dynamical systems, configurations, binary and categorical data, information theory, graphical modelling.