Periodic Reporting for period 2 - TIMEISNOW (The Time is Now: Understanding Social Network Dynamics Using Relational Event Histories)
Reporting period: 2019-08-01 to 2021-01-31
To address these shortcomings, the objectives of this project are threefold. First, I will develop an innovative and advanced statistical framework for the analysis of relational event histories which builds on the novel relational event model, which has great potential but is in a preliminary stage of development. The framework will allow us to analyze relational event data and to see how social dynamics change as a continuous movie. Second, I will implement the new framework in free and user-friendly software to ensure general utilization among social scientists. Third, in collaboration with network experts in organizational sociology, sociology of education, and criminology, I will develop extensions for dynamic social processes in important applications. In sum, this project will yield a groundbreaking new methodology for testing and building theories on time-sensitive processes in social networks.
I. The development of an advanced statistical network framework.
II. The implementation of the framework in free user-friendly software (R and JASP).
III. The application of the new framework and software to real-life social network problems in organizational science, education, and criminology, and other scientific fields.
I The advanced statistical framework for relational event history analysis
* A moving window technique has been developed that allow researchers to analyze relational event history data as a continuous movie. Thereby we can learn about temporal concepts such as speed, rhythm, and pacing, and fast and delayed interaction speed.
* A Bayes factor testing procedure has been developed for quantifying the relative evidence between time-sensitive social theories, a groundbreaking aspect of the project, together with a (software) tutorial about the methodology, a Bayes factor test for change point detection in relational event histories, and a novel computational algorithm for fast computing of Bayes factors.
* A novel Bayesian regularization technique has been developed for analyzing *Big* relational event models which results in simpler solutions (with less parameters) and better predictions of event data.
* We developed a novel semi-parametric modeling technique to learn about the shape of memory decay of actors in a social network. This allows us to see whether past events are “forgotten” in a linear trend, in an exponential trend, or in different trends.
* A latent variable approach has been developed to capture unobserved heterogeneity in relational event data caused by unobserved relationships between actors.
* Fast computational algorithms have been developed for fitting relational event models in an efficient manner.
II Implementation of the framework in free user-friendly software
The software package, with working title “remverse” (combining the abbreviation of the proposed relational event model (rem) and the word “universe”) will consist of several modules for advanced network analysis. Currently, three modules are finished (‘remstats’, ‘remstimate’, and ‘bremory’) while others are under development (‘rembrandt’, ‘remulate’, ‘remify’). The package and these modules will form the core of the software program for advanced temporal social network analysis.
III Learning about dynamic social processes in real-life problems
* Integration processes of new workers in large organizations
The new statistical framework was used to get a more precise understanding about the integration process of new employers in large organizations by analyzing relational event streams of emails containing over the course of a year. The results show that “newcomers” are more active in discussing innovation than “old-timers”, and this effect diminishes in a somewhat linear trend as employees become more established in the firm.
* Communication behavior in the Enron e-mail corpus
The new framework was used for a relational event analysis of the Enron e-mail data to learn about higher-order network behavior and unobserved relationships between actors. The new model resulted in better predictions of the possible receiver actors in the network.
* Relationships between students in friendship networks
The Bayes factors testing methodology was used to find new evidence and insights about communication behavior between students in friendship networks.
* Radio communication during Apollo 13 mission to the moon
The methodology has been used to analyze the radio communication messages between various Houston stations and the flight during the famous Apollo 13 mission. We learned when and how communication structures abruptly changed as a reults of the problems encountered during this mission (“Houston we have a problem”). These insights resulted in a deeper understanding of the possible bottlenecks in communication structures during technical problems.
* Psychological networks of post-traumatic stress disorder (PTSD) symptoms
By applying the Bayes factor tests to psychological networks we learned about causal relations between PTSD symptoms in psychopathology. The results can be used to come up with possible new ways of treating patients with PTSD.
Currently, most of the new statistical methods for advanced relational event history analysis have been developed. A first step has been made towards the implementation of the methodology in freely available statistical software (R and JASP), and towards its application to real-life problems on dynamic social networks in classrooms, organizations, criminology, and other fields of research. After this project we will have a better understanding how long it takes to develop respect in the classroom, how much faster integration occurs among teams consisting of workers from Western cultures as opposed to teams combining Western and non-Western cultures, or when violent interactions between criminal gangs are likely to occur in the future. After this project we will be able to understand all temporal aspects of dynamic social relations using relational event histories.