ARC records nine membership characteristics for 241 armed organizations active since 2000, observed in 313 group-country units. Explore how often each kind of member appears, how they cluster together, and how recruitment varies by region, ideology, and era.
Share of observations coded as having each characteristic. Each percentage counts only the observations where that characteristic could be coded (n in parentheses). Click a bar to filter everything below to groups with that characteristic.
Every characteristic is binary and refers to the group's membership at any point in its history. A code records whether a characteristic is present, not how many members share it.
Among groups coded on all nine characteristics, a few combinations recur again and again. Each column is one profile: filled dots mark characteristics that are present, and the bar shows how many groups share it. Most of the common profiles are broad and differ mainly in whether groups recruit across borders and from state institutions. Click a column to filter to those groups.
Percentage of coded observations in each column with each characteristic (full sample; the filters above do not apply here). These are raw shares. The article's models account for other ideologies, group age, regime type, conflict context, and region.
Groups formed in every era show about six of the nine characteristics on average. Which characteristics they show has changed: groups formed in 1990 or later are less likely to include women or members with varied education and more likely to include former fighters and former police. Part of this reflects the decline of leftist groups and the rise of Islamist ones, but most of the shifts remain after accounting for ideology. Full sample; filters do not apply.
The diversity score (0–9) counts how many of the nine characteristics a group exhibits. Because missing characteristics cannot add to the count, the distribution includes only observations coded on all nine. Hover over a bar to see the groups in it.
Each glyph is one group-country observation. Its nine segments are the nine characteristics, numbered clockwise from the top as in the key: full blue means yes, a short dark segment means no, and a faint stub means no data. The center shows the count of characteristics present out of those coded. Sorted from broadest to narrowest; hover over a glyph to identify it and click it to open the group's profile.
The broadest and narrowest fully coded profiles, plus one organization that shows why ARC codes each country of operation separately. Click a card to open a profile.
One row per group-country observation. Click a column header to sort and a row to open its profile. Active filters apply.
The Armed Group Composition (ARC) Data record nine binary characteristics of armed group membership: gender, age, and educational diversity; language diversity; intra- and extra-country geographic diversity; and prior experience in a state military, a nonstate armed group, or the police.
The sample is drawn from the Armed Group Dataset (Malone 2022): group-country pairs with a UCDP actor ID whose last recorded attack was in 2000 or later. The unit of analysis is the armed group as it operates within a country. Organizations active in more than one country are coded separately in each, because recruitment practices and what observers can document differ across contexts. The 313 observations correspond to 241 organizations, 37 of which operate in more than one country. Most are rebel and insurgent groups, but the sample also includes some parties with armed wings, militias, and de facto governments.
Each observation was coded independently by two trained research assistants working from academic, government, UN, and news sources. The investigators adjudicated every disagreement. 240 observations are coded on all nine characteristics.
A data release article describing the data, coding procedures, validation against existing sources, and an application to ideology is currently under review. The data files, codebook, and replication materials will be posted here when it is published.