Multilayer-dynamic-network-analysis

Description

Analysis of the structure and temporal evolution of a multilayer dynamic network of international interactions using ICEWS data. The analysis considers four interaction layers: Material-, Material+, Verbal-, and Verbal+.

The study includes network centrality analysis, structural network features, Matrix Autoregressive (MAR) modeling, impulse response analysis, residual diagnostics, and simulation experiments.

Project Structure

• analysis.R: Data analysis, MAR model estimation, diagnostics, impulse response analysis, and simulation experiments
• visualizations.Rmd: Static and dynamic network visualizations
• icews_tensor_processed.RData: Processed ICEWS data
• README.md: Project documentation

Methodology

1. Selection of the 25 most central countries using PageRank centrality
2. Computation of structural network features across layers and time
3. Trend and stationarity analysis joint with standardization
4. Estimation of a Matrix Autoregressive (MAR) model
5. Impulse response and residual diagnostics
6. Monte Carlo simulations to evaluate model performance

Data

The analysis uses event-level data from the Integrated Crisis Early Warning System (ICEWS), available through the Harvard Dataverse.

The raw ICEWS event files from 2004–2014 were processed by:

• Combining the yearly event files
• Classifying events into four interaction layers using CAMEO codes
• Removing self-interactions
• Aggregating events by source country, target country, interaction type, and month
• Constructing a four-dimensional tensor of interactions: Source × Target × Type × Month

The resulting tensor is stored in icews_tensor_processed.RData.

The ICEWS data are also used in:

Hoff, P. D. (2015). Multilinear tensor regression for longitudinal relational data. Annals of Applied Statistics, 9(3), 1169–1193. https://doi.org/10.1214/15-AOAS839

Reproducibility

• R: 4.4.2
• RStudio: 2023.09.1+494
• Operating system: macOS Big Sur 11.7.11
• Processor: Dual-Core Intel Core i5, 2.6 GHz
• Memory: 8 GB RAM
• Architecture: x86_64

R Packages

• dplyr 1.1.4
• expm 1.0-0
• fable 0.5.0
• feasts 0.5.0
• ggplot2 4.0.3
• ggraph 2.2.2
• igraph 2.2.1
• ndtv 0.13.4
• network 1.20.0
• networkDynamic 0.12.0
• psych 2.6.5
• scales 1.4.0
• tensorTS 1.0.2
• tidygraph 1.3.1
• timetk 2.9.1
• tsibble 1.1.6
• zoo 1.8-14