Skip to main content
eScholarship
Open Access Publications from the University of California

A Statistical Approach to Detecting Patterns in Behavioral Event Sequences

  • Author(s): Heins, Kevin Andrew
  • Advisor(s): Stern, Hal S
  • et al.
Creative Commons Attribution 4.0 International Public License
Abstract

The identification of recurring patterns within a sequence of events is an important task in behavior research. In this thesis, we develop a probabilistic framework for identifying such patterns from behavioral data. This framework allows us to distinguish between events that belong to a pattern and events that occur as part of background or unstructured behavior. Stochastic processes are introduced to describe the incidence of both background events and events that belong to recurring patterns. The behavioral events are modeled together using a competing risks framework combining the stochastic processes. We develop an inference procedure to detect the sequences present in observed data. The motivation for this work comes from a large scale longitudinal study to assess the impact of fragmented and unpredictable maternal behavior on emotional and cognitive development of children. We describe our results on both simulated data and the maternal data.

We also consider extensions to our model. We develop a model to study population level behavior, allowing us to compare separate populations as well as pool information across individuals. To perform this analysis, we describe how to extend both our model and inference procedure to a hierarchical setting. We describe results for both simulated data and the maternal data.

Finally, we explore the distributional assumptions inherent in our model. We consider a variety of parametric forms for our model, as well as a nonparametric approach that is both flexible and computationally efficient. In addition to improved model fit, these approaches allow us to better describe background behaviors, such as behaviors that occur in bursts or with strict regularity. We also explore the effect of distributional assumptions on both simulated and real data.

Main Content
Current View