Investigating the impact of edge weight selection on the pig trade network topology

Abstract

Traceability of animal movements and robust surveillance are crucial for prevention and control of animal diseases. While network analysis has emerged as a powerful tool for identifying higher-risk holdings through centrality metrics, its effectiveness depends on two methodological choices: (1) edge-weighting schemes (movement frequency vs. animal volume) and (2) centrality metric selection. This study investigates how alternative edge-weighting approaches (frequency vs. volume) influence network topology and node centrality rankings in a pig movement network.

Using 2021 pig movement data from Upper Austria (5,766 holdings; 92,914 movements), we: (1) quantify how edge-weighting schemes (frequency vs. volume) affect network topology and community structure, and (2) evaluate node ranking robustness across three centrality metrics (strength, betweenness, closeness) against epidemic simulation rankings. Our analysis reveals distinct edge weight distributions: frequency-based network exhibited a bimodal pattern, while volume-based was more uniform. We observed strong positive correlations ($\tau$ $>$ 0.42–0.84; $p<0.001$) in node rankings across all centrality metrics (strength, closeness, betweenness), with consistent patterns observed both: (i) between frequency- and volume-weighted networks, and (ii) within each network representation. Strength centrality exhibited the highest correlation with the simulation-based rankings, particularly for the top 5% highest-ranked nodes ($\tau b$ $=$ 0.51 for frequency-based and $\tau b$ $=$ 0.5 for volume-based). These findings highlight that strength centrality provides a computationally efficient and field-practical alternative to epidemic simulations for identifying high-risk holdings. This enables resource-efficient, data-driven surveillance while maintaining epidemiological relevance.