Reliability of relational event model estimates under sampling: How to fit a relational event model to 360 million dyadic events

Jürgen Lerner,Alessandro Lomi

doi:10.1017/nws.2019.57

Jürgen Lerner, Alessandro Lomi

Open Access

https://doi.org/10.1017/nws.2019.57

Copy DOI

Abstract

AbstractWe assess the reliability of relational event model (REM) parameters estimated under two sampling schemes: (1) uniform sampling from the observed events and (2) case–control sampling which samples nonevents, or null dyads (“controls”), from a suitably defined risk set. We experimentally determine the variability of estimated parameters as a function of the number of sampled events and controls per event, respectively. Results suggest that REMs can be reliably fitted to networks with more than 12 million nodes connected by more than 360 million dyadic events by analyzing a sample of some tens of thousands of events and a small number of controls per event. Using the data that we collected on the Wikipedia editing network, we illustrate how network effects commonly included in empirical studies based on REMs need widely different sample sizes to be reliably estimated. For our analysis we use an open-source software which implements the two sampling schemes, allowing analysts to fit and analyze REMs to the same or other data that may be collected in different empirical settings, varying sample parameters or model specification.

Highlights

IntroductionExamples include communication networks (Monge & Contractor, 2003), teams (Leenders et al, 2016), online peerproduction (Lerner & Tirole, 2001; Lerner & Lomi, 2017), international relations (Lerner et al, 2013), and discourse networks (Leifeld, 2016; Brandenberger, 2019)
Relational event networks arise naturally from directed social interaction
The experimental results we present suggest that for some of the most common network effects typically included in empirical model specifications (Lomi et al, 2014), variation caused by sampling is much larger than the estimated standard errors of model parameters, even for relatively large sample sizes

Summary

Introduction

Examples include communication networks (Monge & Contractor, 2003), teams (Leenders et al, 2016), online peerproduction (Lerner & Tirole, 2001; Lerner & Lomi, 2017), international relations (Lerner et al, 2013), and discourse networks (Leifeld, 2016; Brandenberger, 2019). In all these cases, social interaction reflects, and at the same time shapes, relations among social actors and affects individual sentiments—such as trust, esteem, like, or dislike—and their contextual behavioral expressions (Stadtfeld & Block, 2017). It holds that (ui, vi) ∈ Rti. (Note that Butts (2008) denoted by “support set” what we call “risk set”; we adopt the latter term, using the symbol R, since it is commonly applied in survival analysis.)

Methods

Results

Conclusion