Critical Materials Atlas
Method · network robustness

Is China’s network centrality just a truncation artifact?

The Network page draws each material’s trade graph from the top-6 partners per side. A smaller graph mechanically concentrates centrality — so does the “China is the broker” finding hold once we stop truncating? Here every 2024 graph is rebuilt at top-6, top-10, top-20, and full (every reporting country), and China’s centrality re-measured at each.

The Network page draws each graph from only the top-6 trade partners per side. Does China’s central position hold when I stop cropping the network? I rebuild every 2024 graph with more and more of it and re-measure.
How it’s tested

For all 32 materials I rebuild the 2024 trade graph at four cuts — top-6, top-10, top-20, and the full uncapped graph (every reporting country) — and recompute China’s share of total trade flow, its rank as a broker (betweenness), and how much of the network fragments if it is removed. If China’s centrality barely moves from top-6 to full, the finding is real, not an artifact of the crop.

Computed by build_network_sensitivity.py.

China’s throughput share as the graph stops being truncated

Average across 32 materials. If the line is roughly flat, truncation isn’t manufacturing China’s centrality.

Per material — top-6 vs full uncapped graph (2024)

China’s throughput share and betweenness rank at the truncated and full graphs; “frag” is the share of trade routes that break if China is removed.

MaterialCN share top6CN share fullCN bet-rank top6→fullfrag top6→full

Full graph = every reporting country (no per-side cap). Betweenness on inverse-value distances. Fragility = fraction of reachable ordered country-pairs lost when China is removed; it falls on the full graph because real alternative routes exist that the cap hides — so the Network page’s fragility figures are an upper bound. Source: BACI HS17 2024 → network_sensitivity.json.