Relational event models for longitudinal network data with an application to interhospital patient transfers

Fast meta-analytic approximations for relational event models: applications to data streams and multilevel data

Large relational-event history data stemming from large networks are becoming increasingly available due to recent technological developments (e.g. digital communication, online databases, etc). This opens many new doors to learn about complex interaction behavior between actors in temporal social networks. The relational event model has become the gold standard for relational event history analysis. Currently, however, the main bottleneck to fit relational events models is of computational nature in the form of memory storage limitations and computational complexity. Relational event models are therefore mainly used for relatively small data sets while larger, more interesting datasets, including multilevel data structures and relational event data streams, cannot be analyzed on standard desktop computers. This paper addresses this problem by developing approximation algorithms based on meta-analysis methods that can fit relational event models significantly faster while avoiding the computational issues. In particular, meta-analytic approximations are proposed for analyzing streams of relational event data, multilevel relational event data and potentially combinations thereof. The accuracy and the statistical properties of the methods are assessed using numerical simulations. Furthermore, real-world data are used to illustrate the potential of the methodology to study social interaction behavior in an organizational network and interaction behavior among political actors. The algorithms are implemented in the publicly available R package 'remx'.

Nodal Heterogeneity can Induce Ghost Triadic Effects in Relational Event Models

Temporal network data is often encoded as time-stamped interaction events between senders and receivers, such as co-authoring scientific articles or communication via email. A number of relational event frameworks have been proposed to address specific issues raised by complex temporal dependencies. These models attempt to quantify how individual behaviour, endogenous and exogenous factors, as well as interactions with other individuals modify the network dynamics over time. It is often of interest to determine whether changes in the network can be attributed to endogenous mechanisms reflecting natural relational tendencies, such as reciprocity or triadic effects. The propensity to form or receive ties can also, at least partially, be related to actor attributes. Nodal heterogeneity in the network is often modelled by including actor-specific or dyadic covariates. However, comprehensively capturing all personality traits is difficult in practice, if not impossible. A failure to account for heterogeneity may confound the substantive effect of key variables of interest. This work shows that failing to account for node level sender and receiver effects can induce ghost triadic effects. We propose a random-effect extension of the relational event model to deal with these problems. We show that it is often effective over more traditional approaches, such as in-degree and out-degree statistics. These results that the violation of the hierarchy principle due to insufficient information about nodal heterogeneity can be resolved by including random effects in the relational event model as a standard.

From Ties to Events in the Analysis of Interorganizational Exchange Relations

Relational event models expand the analytical possibilities of existing statistical models for interorganizational networks by: (i) making efficient use of information contained in the sequential ordering of observed events connecting sending and receiving units; (ii) accounting for the intensity of the relation between exchange partners, and (iii) distinguishing between short- and long-term network effects. We introduce a recently developed relational event model (REM) for the analysis of continuously observed interorganizational exchange relations. The combination of efficient sampling algorithms and sender-based stratification makes the models that we present particularly useful for the analysis of very large samples of relational event data generated by interaction among heterogeneous actors. We demonstrate the empirical value of event-oriented network models in two different settings for interorganizational exchange relations-that is, high-frequency overnight transactions among European banks and patient-sharing relations within a community of Italian hospitals. We focus on patterns of direct and generalized reciprocity while accounting for more complex forms of dependence present in the data. Empirical results suggest that distinguishing between degree- and intensity-based network effects, and between short- and long-term effects is crucial to our understanding of the dynamics of interorganizational dependence and exchange relations. We discuss the general implications of these results for the analysis of social interaction data routinely collected in organizational research to examine the evolutionary dynamics of social networks within and between organizations.

Discovering trends of social interaction behavior over time: An introduction to relational event modeling : Trends of social interaction

Real-life social interactions occur in continuous time and are driven by complex mechanisms. Each interaction is not only affected by the characteristics of individuals or the environmental context but also by the history of interactions. The relational event framework provides a flexible approach to studying the mechanisms that drive how a sequence of social interactions evolves over time. This paper presents an introduction of this new statistical framework and two of its extensions for psychological researchers. The relational event framework is illustrated with an exemplary study on social interactions between freshmen students at the start of their new studies. We show how the framework can be used to study: (a) which predictors are important drivers of social interactions between freshmen students who start interacting at zero acquaintance; (b) how the effects of predictors change over time as acquaintance increases; and (c) the dynamics between the different settings in which students interact. Findings show that patterns of interaction developed early in the freshmen student network and remained relatively stable over time. Furthermore, clusters of interacting students formed quickly, and predominantly within a specific setting for interaction. Extraversion predicted rates of social interaction, and this effect was particularly pronounced on the weekends. These results illustrate how the relational event framework and its extensions can lead to new insights on social interactions and how they are affected both by the interacting individuals and the dynamic social environment.

Influence of Hospital Characteristics on Hospital Transfer Destinations for Patients With Stroke

BACKGROUND

Patients with stroke are frequently transferred between hospitals. This may have implications on the quality of care received by patients; however, it is not well understood how the characteristics of sending and receiving hospitals affect the likelihood of a transfer event. Our objective was to identify hospital characteristics associated with sending and receiving patients with stroke.

METHODS

Using a comprehensive statewide administrative dataset, including all 78 Massachusetts hospitals, we identified all transfers of patients with ischemic stroke between October 2007 and September 2015 for this observational study. Hospital variables included reputation (US News and World Report ranking), capability (stroke center status, annual stroke volume, and trauma center designation), and institutional affiliation. We included network variables to control for the structure of hospital-to-hospital transfers. We used relational event modeling to account for complex temporal and relational dependencies associated with transfers. This method decomposes a series of patient transfers into a sequence of decisions characterized by transfer initiations and destinations, modeling them using a discrete-choice framework.

RESULTS

Among 73 114 ischemic stroke admissions there were 7189 (9.8%) transfers during the study period. After accounting for travel time between hospitals and structural network characteristics, factors associated with increased likelihood of being a receiving hospital (in descending order of relative effect size) included shared hospital affiliation (5.8× higher), teaching hospital status (4.2× higher), stroke center status (4.3× and 3.8× higher when of the same or higher status), and hospitals of the same or higher reputational ranking (1.5× higher).

CONCLUSIONS

After accounting for distance and structural network characteristics, in descending order of importance, shared hospital affiliation, hospital capabilities, and hospital reputation were important factor in determining transfer destination of patients with stroke. This study provides a starting point for future research exploring how relational coordination between hospitals may ensure optimized allocation of patients with stroke for maximal patient benefit.

Multiple clocks in network evolution

Relational event models shift the analytical focus away from network ties defined in terms of transitions between mutually exclusive states of connectivity, to bonding processes emerging from observable flows linking senders and receivers of action. In this framework, the possibility to connect social mechanisms of theoretical interest to sequences of observed relational events depends on the relative speed at which these mechanisms operate. Building on established non-parametric methods in survival analysis, in this paper we introduce a new approach to the analysis of the internal time distribution of relational mechanisms of broad theoretical interest in research on the evolutionary dynamics of social and other kinds of networks. We propose general algorithms that may be adopted to study the time structure of theoretically relevant network mechanisms. We illustrate the practical value of our proposal in an analysis of a large sample of high-frequency financial transactions observed over a period of 11 years. We show how the internal time structure of the social mechanisms that control flows of market transactions is sensitive to institutional change in transaction regimes induced by successive financial crises. The results we report invite reflection on a new notion of network “structure” incorporating change as one of its constitutive elements. The study suggests a number of conjectures that provide broad conceptual bases for the development of testable hypotheses about the forces that shape the evolutionary dynamics of network structure.

Bayesian dynamic network actor models with application to South Korean COVID-19 patient movement data

Motivated by the ongoing COVID-19 pandemic, this article introduces Bayesian dynamic network actor models for the analysis of infected individuals’ movements in South Korea during the first three months of 2020. The relational event data modelling framework makes use of network statistics capturing the structure of movement events from and to several country’s municipalities. The fully probabilistic Bayesian approach allows to quantify the uncertainty associated to the relational tendencies explaining where and when movement events are established and where they are directed. The observed patient movements’ patterns at an early stage of the pandemic can provide interesting insights about the spread of the disease in the Asian country.

The Third Man: hierarchy formation in Wikipedia

Wikipedia articles are written by teams of independent volunteers in the absence of formal hierarchical organizational structures. How is coordination achieved under such conditions of extreme decentralization? Building on studies on the organization of dominance relations in animal and human societies, we theorize that coordination in Wikipedia is made possible by an emergent hierarchical order sustained by self-organizing sequences of text editing events. We propose a new method to turn the editing history of Wikipedia pages into an evolving multiplex network resulting from three types of interaction events: dyadic undo, dyadic redo, and third-party based edit events. We develop new relational event models for signed networks that specify how the probability of observing various types of edit events depends on their embeddedness in sequences of past edit events. Using a random sample of page histories comprising 12,719 revisions produced by 7,657 unique users, we examine the relation between theoretically defined sequences of text editing events, and the emergence of linear dominance hierarchies that regulate production relations within Wikipedia. We find evidence that dyadic interaction gives rise to systematic extra-dyadic dependence structures that are partially consistent with a hierarchical interpretation of the Wikipedia editing network. We support and complement the statistical analysis of multiplex event networks with data visualizations that provide qualitative validation of our main results.