Social and Political Interacting Networks (SPINS)
December 8, 2017
693 Kerr Hall, UC Davis
Session 1: 9:00 – 10:30. Networks of Events Data and Shocks
Jesse Hammond, Department of Defense Analysis Naval Postgraduate School, Monterey
Cheap Talk and Militarized Disputes
Quantitative tests of constructivist theory tend to rely on structural variables such as shared political institutions and linguistic similarity to explain international conflict and cooperation. I build on constructivist theory by measuring levels of affinity or enmity between states using verbal engagements between states: threats, promises, accusations, and so on. These are low-information events — ‘cheap talk’ — that individually say little about how states view one another. In the aggregate, however, verbal engagements serve two purposes. First, they are imperfect but useful indicators of the underlying level of trust or affinity between states. Second, over the long term these engagements socialize states toward norms of cooperation or conflict with one another. Even after controlling for a range of variables common in the conflict literature, I find that states that engage one another positively tend to avoid militarized disputes and are more likely to assist one another during domestic crises.
Brandon Kinne and Grace Mineo-Marinello, Department of Political Science UC Davis
Detecting Shocks in the Global Event Network
Global event data measure interactions between sovereign governments. These interactions range from cooperative actions (such as providing aid or policy support) to hostile actions (such as implementing sanctions or using force). We use the Integrated Crisis Early Warning System (ICEWS) dataset to measure conflictual and cooperative interactions on a weekly basis. We use these data to specify a longitudinal, multiplex network, with 72 layers measured at weekly intervals over the 1995-2017 time period, where governments comprise the nodes and cooperative/conflictual events comprise the edges. Each layer corresponds to a different type of interaction. Shocks in this network involve time periods where the patterns of cooperative and conflictual ties between governments significantly deviate from typically observed patterns. We detect shocks in a series of steps. First, for a given weekly time period, and for each layer in the multiplex, we derive a central graph for the prior four weeks, which defines the typical pattern of ties in the network. We use this central graph as a reference matrix. Second, we compare—again for each layer in the multiplex—the focal network to the reference network using an Ipsen-Mikhailov distance metric. This produces a distribution of 72 distance metrics for each time period. Third, using bootstrap resampling, we compare the distribution of distance metrics in the focal time period to the prior 51 weeks, and we derive means and confidence intervals. Fourth, we repeat this procedure over successive time periods, and we identify those periods in which patterns of event ties are statistically significantly different than in prior time periods. This approach (1) preserves information from each layer in the multiplex, (2) provides either a continuous measure of shock magnitude or a binary indicator of shock presence/absence, and (3) identifies shocks on a global, rather than node-level, scale. Slides available here.
Coffee Break: 10:30 – 10:45
Session 2: 10:45-12:15. Shocks in Social Networks: Agent-Based Models and Experimental Results
Keith Burghardt, Departments of Political Science and Computer Science, UC Davis
Are Human Subjects Smarter than Rational Agents in a Social Agent-Based Model? Comparing the Effects of Shocks on Social Cooperation in Simulated and Human Settings
In many cooperative networks, such as alliance and trade networks, abrupt and intense changes to the state of the system (which we call “shocks”), such as wars or tariff increases, can substantially change the network. We examine how such shocks affect multiplex networks via an agent-based model (ABM) that simulates network evolution prior to and following shocks. In our model, agents interact on a coupled pair of networks, and add, drop, or change ties in order to maximize their utility. At a certain time-point, some agents are “shocked” by changing (increasing or decreasing) the cost associated with tie-formation or tie-maintenance. In this ABM, we vary the number of agents that are being shocked, as well as the values of the utility function. We compare the results of the ABM to an experiment with human subjects, who interact in a setting that is equivalent to the conditions tested in the ABM. We find that, with few exceptions, human subjects follow structural patterns that are similar to those observed in the ABM. However, under certain conditions, the utility obtained by human subjects significantly exceeds the utility of agents in the ABM. We discuss the implications of these results. Slides available here.
Democratization and International Networks.
Camber Warren, Department of Defense Analysis, Naval Postgraduate School, Monterey.
Empirically Grounded Simulation of Multiplex Co-Evolution: Predicting the Effects of Regime-Type Shocks
Existing approaches to the modeling and simulation of dynamic multiplex networks (networks composed of multiple node types and edge types, observed at multiple times) are limited. Stochastic actor-oriented models (SAOMs) allow for structural influences between multiple edge types, but require agent decisions to occur through micro-steps that are unlikely to match many empirically observed data-generating processes. Temporal exponential random graph models (TERGMs) allow for more flexible structural effects between network time periods, but cannot represent the co-evolution between multiple edge types. Here, I present an approach to estimating the parameter weights governing the “utility” of multiplex edge co-evolution, using bootstrapped pseduo-likelihood estimation of dynamic edge “motifs”. This provides a fast, scalable approach to the empirical grounding of agent-level computational simulations of complex network dynamics, and to the prediction of edge formation and dissolution decisions over time. I apply this approach to analysis of the co-evolution of conflictual, cooperative, and neutral interactions between 167 states, over the period 1995-2015, measured using event metrics derived from the ICEWS dataset. The results show that regime-type homophily plays an important role in state behavior, and that regime-type “shocks” (sudden changes in governing institutions) generate systematic shifts in subsequent patterns of conflict and cooperation. By using these estimates to simulate the forward evolution of the network (e.g. using data up to 2014, to predict the configuration in 2015), the results also demonstrate that the out-of-sample edge-level predictions generated by these empirically-grounded simulations substantially exceed the accuracy of competing approaches to multiplex simulation. Slides available here.
12:15 – 1:00. Lunch
Session 3: 1:00 – 2:30. Imbalances in Multiplex Networks: Political, International, and Animal Behavior
Zeev Maoz, Department of Political Science, UC Davis
Organisms interact in multiple settings. Some of these settings involve positive ties, some involve strictly negative ties, and still other settings involve ties that may be either positive or negative. Extensive research across a wide array of scientific disciplines explored the presence and consequences of imbalanced relations—e.g., the friend of my friend is my enemy, the enemy of my enemy is my enemy. A relational imbalance is defined as a contradictory set of paths connecting a pair of nodes (or more formally a relational cycle with a negative sign). There has been considerable work on imbalance in network science since the pioneering work of Harary (1956) and his associates. However, little research has been done on imbalances in multiplexes. We develop a method to measure relational imbalances in multiplex networks and compare it to a null model that estimates the degree of imbalance that obtains by chance. We illustrate our method on international (alliance and conflict) and domestic (Congressional roll call) political networks, as well as on a multiplex network of behaviors of monkeys. We examine some of the correlates of such imbalances, and discuss the implications of this work for network science and substantive social and political settings. Slides available here.
Martón Pósfai, Department of Computer Science, UC Davis
Rhesus Macaques — Societal Collapse in a Multiplex Social Network
Rhesus macaques live in cohesive hierarchically-structured groups, their social organization is regulated by a multiplex network of affiliative and adversarial interactions. A notable property of macaque societies is that they can become unstable: the hierarchical organization may collapse, culminating in large-scale fighting, dissolution of social order and disbanding of entire groups. Here we analyze social networks collected from 5 groups of rhesus macaques housed at California National Primate Research Center. Each study consisted of a baseline observation period, followed by an experimental perturbation and further observation. Although it was never the goal to destabilize groups, in 2 out of 6 cases the social organization collapsed, offering an opportunity to study drastic transitions in social systems. Analyzing the dataset we reveal key features of the multiplex social network, group fight structure and heterogeneity of populations. We extend mathematical models of hierarchy formation and maintenance to include these important features, allowing us to identify key features of social organization and possible mechanisms leading to instability.
Coffee Break: 2:30 – 2:45.
Session 4. 2:45 – 4:15. Co-Evolution of Multiplex Networks
Haochen Wu, Department of Computer Science, UC Davis
Quantifying Correlated Structural Evolution in Multiplex Networks
Many natural, engineered, and societal systems consist of several differing types of interactions, and such interactions are intrinsically captured by a multiplex network. Crucial to understanding the behavior of the entire system is the relationship between layers in such a network, both statically and dynamically. In contrast to well-studied node-node relationship in such networks, interactions among layers were not systematically checked before and there is an increasing demand for measures quantifying such interactions. We develop a method of applying information theoretic measures, which is well-known to be suited for quantifying interaction, dependency, and influence, but was not applied to multiplex networks before due to their non-trivial connectivity patterns, to the layers of a multiplex network. We find in many data that there are usually complex relationships between layers that may worth further study. For example, while undergoing evolution some layers may have temporal precedence on others, which can provide enhanced predictability over the process. Our method aides in the identification of the layer which has the most predictive power over other layers in the network, and therefore may justify further study. This paradigm also introduces many opportunities to explore how other information measures can help in understanding network evolution.
Weiran Cai, Department of Computer Science UC Davis.
Hierarchy and Evolution of the Hidden World-Wide Military Cooperation Network
Human civilization has been undermined by hidden powers. The culprits are both numerous non-state military groups and the supporters behind them. International studies have stretched out from standard direct warfare to these indirect confrontations, which tend to have an even more enduring influence on interstate relations. While causes and consequences of such obscure collusive relations have been questioned mostly on the state level, a global view of the hierarchy of the entire network and its evolutionary path remain unclear. We employed the latest hierarchy detection algorithms to study this bipartite state-nonstate armed group (NAG) network and traced its evolution over the period between 1946 and 2010. We discovered that the network has developed into an increasingly robust architecture, which is analogous to that of ecological mutualistic networks. The dynamical partitioning of the temporal slices of the network also showed the consistency in the roles of states and NAGs as inter-modular hubs or peripheral nodes that are beset only in single modules. This hierarchical view of this bipartite cooperation network is expected to provide guidance for the destruction of critical links. In another thread, we also tried to understand the cause of the formation of such hierarchy in a general context. We proposed a stochastic model based on the optimisation of payoffs of participating units.
Break: 4:15 – 4:20.
Session 5: 4:20-5:00. Shocks on the Web.
George Barnett and Carlos Algara, Departments of Communication and Political Science, UC Davis
The Evolution of the International News Website Network
This research examines the evolution of the international worldwide web (WWW) and its response to shocks (extreme events), which may be represented as a two-mode network. Daily data were collected from Alexa.com on the use of the world’s 500 most frequently visited websites by 118 countries over the 17-month period September 1, 2015 to January 30, 2017, to create a longitudinal two-mode network (countries and websites). From these data, the countries use of the 44 most frequently visited news websites were extracted because it was thought that these sites would be more volatile in response to events. While certain events, such as the November 2015 Paris terrorist attacks and the outcome of the Brexit vote (June 2016), had an impact on the structure of the network, it remained remarkably stable over time. Reasons for its stability will be discussed. We plan to use these data to examine the coevolution of the international news network with the event network data being collected by Brandon Kinne. Slides available here.
5:00-5:15. Concluding Comments.
For questions regarding this workshop and/or to access workshop content, please contact Jaime Jackson (email@example.com) or Zeev Maoz (firstname.lastname@example.org).