Fetch one row per campaign/episode onset from several conflict datasets

Usage

fetch_campaign_events(datasets, start_year, end_year, coding_system)

Arguments

datasets: Character vector of conflict_data() dataset keys.
start_year, end_year: Integer year bounds.
coding_system: "cow" or "gw".

Value

A data frame with dataset, unit (the coding-system code), year, and region (World Bank 7-region classification, via countrycode::countrycode()).

Details

"navco1.3", "beissinger", "csra", and "mec" are already one row per campaign/episode. "navco2.1" is not: conflict_data() returns one row per campaign-year, so a single long-running campaign contributes many rows. Comparing that directly against onset-coded datasets would overstate NAVCO 2.1's event counts by roughly the average campaign duration. To keep the comparison on the same onset-level footing as the source article, "navco2.1" is filtered here to nvc2.1_ONSET == 1, conflict_data()'s own per-campaign onset flag (keyed on the numeric campaign id, not the free-text campaign name, which is not unique – e.g. three distinct "Myanmar Regime Change Campaign" entries share that name). That flag is computed over the dataset's full native coverage before start_year/end_year filtering, so campaigns already under way at start_year are not mistaken for new onsets.

countrycode::countrycode()'s "region" destination is not a clean 7-category classification: a handful of pre-1990 historical entities (Yemen Arab Republic, Yemen People's Republic, the United Arab Republic) still carry the World Bank's older "Middle East & North Africa" label, while every other country in that region carries the newer, longer "Middle East, North Africa, Afghanistan & Pakistan" label – these historical COW/GW codes do appear in campaign data covering the 1950s-80s, so left as-is they'd split one region into two bars. The newer label is recoded to the shorter, older one here (kept as the canonical form for display) so the region breakdown stays at exactly 7 categories.