A new approach to analyzing coevolving longitudinal networks in international relations
| Author | Shahryar Minhas,Peter D Hoff,Michael D Ward |
| Published date | 01 May 2016 |
| DOI | 10.1177/0022343316630783 |
| Date | 01 May 2016 |
| Subject Matter | Research Articles |
A new approach to analyzing coevolving
longitudinal networks in international
relations
Shahryar Minhas
Department of Political Science, Duke University
Peter D Hoff
Departments of Statistics & Biostatistics, University of Washington
Michael D Ward
Department of Political Science, Duke University
Abstract
Previous models of international conflict have suffered two shortfalls. They tend not to embody dynamic changes,
focusing rather on static slices of behavior over time across a single relational dimension. These models have also been
empirically evaluated in ways that assumed the independence of each country, when in reality they are searching for
the interdependence among all countries. A number of approaches are available now for analyzing relational data
such as international conflict in a network context and a number of these can even handle longitudinal relational
data, but none are developed to the point of exploring how networks can coevolve over time. We illustrate a solution
to the limitations of existing approaches and apply this novel, dynamic, network based approach to study the
dependencies among the ebb and flow of daily international interactions using a newly developed, and openly
available, database of events among nations.
Keywords
dynamic networks, event data, international crises, tensor products, time series
Motivation
What are the tools available to us for studying endogen-
ous relational data structures in international relations?
The canonical research design in the field of interna-
tional relations for the analysis of relational data remains
the directed dyadic approach. The units of analysis in
this design are Ncountries that have been paired
together to form a dataset of Kdirected dyads. This
procedure dramatically increases the size of the dataset
as every country is paired with another resulting in
NNNdyads. The size of the dataset is then fur-
ther increased by the incorporation of the dyadic frame
into a panel context in which each of the Kdyads is
observed over Tyears, resulting in hundredsof thousands
or millions of supposedly independent observations.
Though this design has remained standard for decades,
this approach has been repeatedly argued to produce
biased statistical results (Hoff, 2005; Snijders, 2011;
Cranmer, Desmarais & Menninga, 2012). The bias
results from the fact that in the context of most relational
data, the interactions between actors are interdependent
(Hoff & Ward, 2004; Cranmer & Desmarais, 2011).
More explicitly, the directed dyadic design assumes that
given the other covariates in the model, the formation of
a tie between any two countries is independent of the
formation of ties among other countries in the network.
However, if dependencies between pairs of actors exist,
Corresponding author:
michael.d.ward@duke.edu
Journal of Peace Research
2016, Vol. 53(3) 491–505
ªThe Author(s) 2016
Reprints and permission:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/0022343316630783
jpr.sagepub.com
To continue reading
Request your trial