Methodology方法论

Panel construction面板数据构建

The published view is a province × yearbinary treatment panel covering African ADM1 regions, 2000–2024. A province-year is coded CRM_active = 1 if at least one critical-raw-material mine within its boundary had a recorded owner that year.

发布视图是一个覆盖非洲 ADM1 区域、2000–2024 年的 省 × 年 二元采矿面板数据。 若某省边界内至少有一座关键原材料矿山在该年有记录在案的所有者，则该省-年被编码为 CRM_active = 1。

Construction proceeds in three steps:

构建分为三个步骤：

1. Reshape S&P’s 25 historical-owner columns (one per year, 2024 → 2000) into long form (mine_id, year, owner); mark a mine-year as active if owner is present.
2. Spatially join each mine’s lat/lon to the Natural Earth 10 m ADM1 polygon containing it (nearest-neighbour fallback for coastal / borderline cases).
3. Aggregate to province-year: CRM_active(GID_1, year) = max(active mines that year).

1. 将标普（S&P）的 25 个历史所有者列（每年一列，2024 → 2000）重塑为长格式 (mine_id, year, owner)；若所有者存在，则将该矿山-年标记为活跃。
2. 将每座矿山的经纬度空间连接到包含它的 Natural Earth 10 m ADM1 多边形 （对沿海/边界情况采用最近邻回退）。
3. 聚合到省-年：CRM_active(GID_1, year) = max(active mines that year)。

Data sources数据来源

• S&P Global Metals & Mining (Properties database). Mine-level records: name, location (lat/lon), country, primary commodity, ownership history per year. Filtered to African properties; commercial dataset.
• Natural Earth 10 m Admin-1 States & Provinces. Public-domain sub-national boundaries used for the spatial join. naturalearthdata.com. GADM v3.6 was the original choice in the underlying R pipeline; we use Natural Earth 10 m here for network-fetch reliability and a near-equivalent province count (~908 vs ~1,000).

• S&P Global Metals & Mining（Properties 数据库）。 矿山级记录：名称、位置（经纬度）、国家、主要矿产、逐年所有权历史。 已筛选为非洲矿产；商业数据集。
• Natural Earth 10 m Admin-1 States & Provinces。 用于空间连接的公共领域次国家级边界。 naturalearthdata.com。 底层 R 流程最初选用 GADM v3.6；此处改用 Natural Earth 10 m，以获得更可靠的网络获取 和近乎等价的省份数量（约 908 对约 1,000）。

Caveats注意事项

• Ownership ≠ operation. A non-null historical-owner field indicates a mine was on the books that year, not that it was physically producing. Active vs Inactive operational status is captured separately in the per-mine actv_status field.
• Spatial-join slack. Two mines were snapped >50 km from their reported coordinates to the nearest ADM1 polygon (typically coastal points or imprecise coords).
• Coverage skew. 529 mines are concentrated in a handful of mineral-rich economies (DRC, South Africa, Zambia, Zimbabwe, Namibia together account for >65 % of records). Many African provinces register no treatment in this dataset; that means S&P has no commercially-tracked mine there, not that no extraction occurs.
• Commodity classification. Each mine has one PRIMARY_COMMODITY field. Multi-commodity operations (frequent in copper-cobalt belts) are present in the underlying COMMODITIES_LIST column but only the primary tag is exposed in the published views.

• 所有权 ≠ 运营。 非空的历史所有者字段表示该矿山当年在册，并不代表其实际在生产。运营状态（活跃与停产） 单独记录在每座矿山的 actv_status 字段中。
• 空间连接误差。 有两座矿山被从其报告坐标处吸附到最近的 ADM1 多边形，距离 >50 公里 （通常为沿海点位或坐标不精确）。
• 覆盖偏斜。 529 座矿山集中在少数矿产丰富的经济体（刚果民主共和国、南非、赞比亚、 津巴布韦、纳米比亚合计占记录的 >65%）。许多非洲省份在本数据集中没有采矿记录； 这意味着标普（S&P）在当地没有商业追踪的矿山，而非当地没有开采活动。
• 矿产分类。 每座矿山有一个 PRIMARY_COMMODITY 字段。 多矿产运营（在铜钴带常见）存在于底层 COMMODITIES_LIST 列中，但发布视图仅展示主要矿产标签。

Future expansion未来扩展

The R pipeline behind this dataset (in the source folder, not redistributed) joins many more data layers onto the same province-year skeleton. A later release of this site is expected to add:

本数据集背后的 R 流程（位于源文件夹，未对外分发）在同一省-年骨架上连接了更多数据图层。 本站后续版本预计将新增：

• UCDP Georeferenced Event Dataset (GED v25.1). Annual conflict-event counts and battle-related deaths per province.
• ACLED Armed Conflict Location & Event Data. Protests and riots per province-year.
• World Bank WGI / WDI. Six governance indices and seven economic indicators at the country-year level.
• EPR ethnic-politicization indicators.
• Spatial & temporal lags (KNN-10 and 1-year lags) for network-style analyses of cross-province spillovers.

• UCDP 地理参照事件数据集（GED v25.1）。 各省逐年冲突事件计数及战斗相关死亡人数。
• ACLED 武装冲突地点与事件数据。 各省-年的抗议与骚乱。
• 世界银行 WGI / WDI。 国家-年层面的六项治理指数与七项经济指标。
• EPR 族群政治化指标。
• 空间与时间滞后（KNN-10 与 1 年滞后），用于 跨省溢出效应的网络式分析。

Citation引用

If you use this dataset, please cite the underlying source paper (forthcoming) and the data source:

如果您使用本数据集，请引用其底层来源论文（即将发表）及数据来源：