Data sources and interpretation

Package-related transformations and explanations

SCAD repeats events when they span multiple locations. For country-level use, contentiousR retains one row per positive event ID before aggregating. Its documented -99, -88, and -77 death-count values are represented as missing rather than as negative deaths.

UCDP/PRIO expands conflicts with multiple location codes before aggregation. ucdp_prio_incidence marks any active conflict in a country-year, while ucdp_prio_onset is one only when start_date2 falls in that year. The VPP file has no equivalent episode-start field; ucdp_vpp_onset is retained only as a backwards-compatible alias of ucdp_vpp_incidence.

Mass Mobilization reports participant size through a mixture of exact numbers, ranges, inequalities, and qualitative strings. mm_participants contains the maximum of exact numeric reports. mm_participants_reported preserves all reported strings so users can apply a substantively appropriate recoding.

Legacy campaign and episode sources can contain several rows for one country-year. add_conflict(aggregate = TRUE) reports their count and takes a missing-safe maximum of numeric fields. Use aggregate = FALSE or call conflict_data() directly when campaign-level observations are required.

Unmatched panel rows remain NA, because absence from a source is not always evidence of zero events.

MEC contains 2,734 reformist and maximalist contentious episodes worldwide. contentiousR preserves its episode-level rows and all published variables, using the source’s byear (beginning year) as year. Fourteen left-censored episodes have a mec_bdate before 1955 but a source byear of 1955. With add_conflict(aggregate = TRUE), MEC follows the same explicit aggregation policy as the other campaign and episode sources; use aggregate = FALSE to retain individual episodes.

The bundled UN WPP extract has three rows for some China country-years: a combined mainland+Hong Kong+Macao+Taiwan figure, a mainland-only figure, and a Hong Kong-only figure, all of which matched COW code 710 during country-name conversion upstream. load_population_data(dataset = "wpp") resolves this by taking the median pop value within each cow-year group (returned as wpp_pop), which is consistently the mainland-only figure; every other country-year has a single row, so the median is a no-op there.

NMC’s documented -9 missing-data sentinel is recoded to NA in nmc_tpop and nmc_upop.

Fariss, Anders, Markowitz, and Barnum’s replication files contain one row per country-year for each underlying source indicator (e.g. Bairoch, Penn World Table, World Bank) plus one "latent_*" row holding the model’s combined estimate. load_gdp_data(dataset = "fariss") and load_population_data(dataset = "fariss") keep only the "latent_gdp" / "latent_gdppc" / "latent_pop" rows. Their absolute units are not independently verified in this package; consult Fariss et al. (2022) before using fariss_gdp, fariss_gdppc, or fariss_pop outside of relative/comparative analysis.

The SIPRI workbook lists a small number of rows (e.g. “Africa”, “NATO”) that are regional or organizational aggregates rather than countries; these have no matching COW/Gleditsch-Ward code and are dropped.

Barnum, Fariss, Markowitz, and Morales’s military expenditure replication file contains one row per country-year for each of 24 underlying source indicators, each in its own original currency/unit, with no single combined value on a real monetary scale (unlike the Fariss GDP/population files, the paper does not publish the shared latent trait directly because “the average by itself does not have a real or direct monetary unit”). Instead, load_military_expenditure_data(dataset = "barnum") reports the model’s posterior estimate of two specific indicators: barnum_sipri ("milex_con_2022_sipri") and barnum_nmc ("milex_con_2017_nmc"), which extend each one’s coverage back to 1816 using information from all 24 sources. Both are in constant (inflation-adjusted) dollars, so barnum_nmc is not directly comparable to nmc_milex from dataset = "nmc", which NMC documents in current-year dollars; their ratio varies smoothly over time (around 2.4 for the United States in 1990, falling toward 1 by the mid-2010s) rather than reflecting disagreement between the sources. barnum_milburden (milexgdp) is a unit-free ratio of spending to GDP and so is published directly.

Licensing

The package’s MIT license applies to the software, not to third-party data. Gapminder and UCDP identify their bundled data as CC BY 4.0; NAVCO 2.1 and MEC are CC0; UN WPP is CC BY 3.0 IGO. Other source files, including NMC, SIPRI, and the Fariss et al. and Barnum et al. estimates, retain their creators’ terms — SIPRI in particular is free to use with attribution but requires separately negotiating a royalty for commercial use. See inst/COPYRIGHTS and verify the applicable terms before redistributing a package build containing those files.