< Back to previous page

Project

Unraveling behavioral sequences in micro-coded parent-child interactions: A dynamic network approach to binary time series data

Family interaction processes are considered to be the engine of child development. To study the moment-to-moment dynamics of exchanges between parents and their children, the parent-child interactions are often videotaped and coded in short time intervals (e.g., a few seconds) for the absence or presence of multiple target behaviors, yielding binary time series data. On top of the relative frequencies, also the co-occurrence and sequential dependency of the behaviors reveal important aspects of the interaction patterns and should best be unraveled in the analysis. Given the obvious complexity of the data (i.e., a multitude of behaviors tracked across time) new data-analytic methods are needed. The key idea of this dissertation is to transform the binary time series into an easy-to-grasp network. The network nodes and their size visualize the different behaviors under study and their relative frequencies. The network edges (i.e., links between the nodes) depict the co-occurrence or the sequential dependencies between the behaviors.

This dissertation consists of five Chapters. In Chapter 1, we introduce the main network idea. In this chapter, we quantify the strength of the behavioral dependencies by means of the Jaccard similarity index. We demonstrate that this approach yields novel insights into affective family interactions by comparing the co-occurrence and sequential patterns of families with a depressed to those with a non-depressed adolescent. In Chapter 2 we propose a corrected version of the Jaccard index. The resulting index indicates whether a behavioral sequence is shown more often than would be expected by chance, given the relative frequency of the behaviors. We used the new index to reveal which moment-to-moment interaction dynamics of middle childhood aged children and their mothers can be related to self-reported styles of attachment. To account not only for the relative frequency, but also for the serial dependency of behaviors from one moment to the next, we compare in Chapter 3 two significance tests for Jaccard values, one model-based approach and one permutation-based technique. In Chapter 4 we showcase the wide applicability of the approach by applying it to longitudinal binary symptom data in which patients reported on the weekly presence of depressive symptoms over two years. In Chapter 5 we focus on the interrater reliability of the obtained micro-codes, by investigating which interrater agreement measure is best suitable for binary time series data.

Date:1 Oct 2016  →  7 Oct 2020
Keywords:Dynamic Network Analysis, Binary Time Series, Boolean Regression
Disciplines:Applied psychology
Project type:PhD project