Peer grouping in Civica
Why countries are compared the way they are.
The problem
Every comparison needs a peer set. Saying that France “ranks 12th” is meaningless without specifying the ranking universe — 12th out of what? Civica’s previous peer set was the in-house structural_family taxonomy, which used regular-expression matching over CIA World Factbook prose to bin every country into one of ten buckets (parliamentary democracy, presidential republic, semi-presidential, constitutional monarchy, absolute monarchy, one-party state, military rule, theocracy, directorial republic, and a residual other category).
Two examples of how the classification broke down:
- Material outcomes.Defaulting peer grouping to government type would group Nigeria’s Human Development Index against fellow presidential democracies — the United States, France, Brazil, Indonesia, the Philippines. That is not an analytically useful peer set for human development; the relevant peers are sub-Saharan Africa lower-middle-income economies.
- Governance outcomes. A regime classification grouped 64 countries from China to Belarus to pre-2025 Saudi Arabia into a single civilian dictatorship bucket. The bucket is analytically meaningless because it conflates closed autocracies (no meaningful electoral competition) with electoral autocracies (multi-party elections that are unfair but real).
The principle: peer sets are domain-specific
No major reference institution defaults peer grouping to a single universal primitive. Our World in Data publishes development indicators against World Bank regional and income groupings; the World Bank itself groups by region (seven buckets), income (four tiers), and lending category; the International Monetary Fund and United Nations follow similar conventions.[1] [2] [3]When governance is the subject, comparative politics standardly uses regime classifications like V-Dem’s Regimes of the World or the Polity Project’s scale.[6] [8]
Civica adopts the same domain-specific architecture. Material and governance indicators get different peer-grouping primitives because they answer different questions. Constitutional form — whether a country has a king, whether it’s federal, whether it’s a republic — is preserved as descriptive metadata, not as an analytical taxonomy.
Material outcomes — World Bank region × income
For Civica Conditions, the Human Development Index, GDP-based measures, demographics, and health outcomes, the default peer set is the country’s World Bank region intersected with its World Bank income group. That gives seven regions × four income tiers = up to 28 cohorts; in practice only ~18 are populated.[5]
Why region+income? The World Bank publishes country-and-lending groups precisely because policy-relevant material comparisons generally need both axes. A high-income East Asian economy (Singapore, Japan) has structurally different material outcomes from a low-income East Asian economy (Cambodia, Lao PDR), even though both share regional context. Conversely, a high-income economy in Sub-Saharan Africa (Seychelles) faces different material conditions from a high-income economy in North America (United States), even at matched income. Civica takes the World Bank’s convention as authoritative here.[4] [5]
Governance outcomes — V-Dem Regimes of the World
For Civica Index dimensions (democratic quality, rule of law, freedoms and rights, corruption control), Pulse signals, and any governance-flavored ranking, the default peer set is the country’s tier on V-Dem’s Regimes of the World classification — one of four buckets: closed autocracy, electoral autocracy, electoral democracy, or liberal democracy.[8]
Why V-Dem RoW over Bjørnskov-Rode / CGV for the default? RoW splits autocracy along the analytically meaningful electoral / closed axis — eliminating the 64-country civilian-dictatorship blob that prompted the change — and is methodologically coherent with the Civica Index’s existing V-Dem dependency.[8] [9]
Transparency note on V-Dem dependency. The Civica Index already uses V-Dem indicators in two of its 4dimensions (Democratic Quality, Rule of Law). Using V-Dem RoW as the governance peer set is presentational only — it determines which countries appear together in a ranking, not how their scores are computed. The CI’s scoring formula is unchanged. There is no circularity, but the overlap is worth disclosing.
Optional alternate regime lens — Bjørnskov-Rode / CGV
For users who want the executive-form-of-autocracy distinction (military dictatorship vs civilian dictatorship vs royal dictatorship), Bjørnskov-Rode / CGV remains available as a user-toggleable alternate lens.[7] [10] [11] [12] CGV has six buckets: parliamentary democracy, presidential democracy, semi-presidential democracy, civilian dictatorship, military dictatorship, and royal dictatorship.
BR/CGV is preserved for two reasons. First, it’s the taxonomy used in a substantial body of comparative political economy research, and Civica should not silently make that literature harder to cite against. Second, the executive-form distinction is sometimes the analytically relevant grouping — for example, when comparing fiscal discipline across military vs civilian autocracies. We make it accessible without making it the default.
Constitutional form as metadata
“Is there a king?” “Is the country federal?” “Is it a republic or a monarchy?” These are facts about the constitutional shell of the state, not analytical claims about how it should be compared. Civica preserves them as descriptive metadata, in two forms:
- government_form_description— free-text description drawn from the CIA World Factbook’s government type field (e.g. “federal parliamentary democracy under a constitutional monarchy”).[13]
- monarchy_status — small controlled vocabulary for filterability:
none— No reigning monarch.constitutional— Reigning monarch with constitutional limits and meaningful executive role (Liechtenstein, Jordan).ceremonial— Reigning monarch with no executive role (United Kingdom, Sweden, Spain, Japan, Cambodia).elective— Monarch chosen via political or religious process (Vatican conclave; Malaysia's rotating sultanate).absolute— Reigning monarch without constitutional limits (Saudi Arabia).theocratic— Religious head conflated with head of state where this is the load-bearing classification.
Neither field is used as a peer-grouping primitive. They are available as filters (“show me all ceremonial monarchies”) and as searchable text on country pages. The vocabulary above is provisional — if the canonical fact-layer derivation lands a different vocabulary, this methodology adopts theirs and notes the diff in the changelog.
Minimum-n rule and fallback chain
A peer band only renders when the cohort has at least n ≥ 8 countries. Below that floor, the panel falls back to a broader grouping with an explanatory label, following these chains:
Material lens (region × income):
- region + income (the default)
- region only
- income only
- global
Governance lens (V-Dem RoW tier):
- V-Dem RoW tier (the default)
- global
The governance fallback is flatter on purpose. RoW only has four tiers, so once a country is outside its tier the next-most-meaningful grouping is “all democracies” or “all autocracies” — categories that are honestly less interpretable than a clean global comparison.
When the fallback fires, the panel displays a small note such as “n=4 in region+income; using region only (n=21).” Readers should always be able to see when a substitution has happened.
Coverage limitations — non-sovereign and ambiguous jurisdictions
Some jurisdictions lack World Bank or V-Dem coverage entirely. Civica documents the per-jurisdiction fallback explicitly rather than silently mapping these to the closest peer:
| Jurisdiction | World Bank | V-Dem RoW | Plan |
|---|---|---|---|
| Taiwan | Not listed | Covered | Use V-Dem RoW for governance; for material indicators, fall back to East Asia & Pacific + High income with a documented note. |
| Kosovo | Listed | Not in RoW pre-2008 | Both lenses available; flag RoW data as starting from inclusion year. |
| Palestine | Listed (West Bank and Gaza) | Not in RoW | World Bank for material; for governance, fall back to global with a 'no V-Dem coverage' indicator. |
| Western Sahara | Not listed | Not in RoW | Both lenses unavailable; show only government_form_description and a 'limited peer comparison available' pill. |
| Vatican City | Not listed | Not in RoW | Same as Western Sahara — descriptive metadata only. |
When neither lens applies, country pages display only the government_form_descriptionand a “limited peer comparison available” pill. This explicit unavailability state is preferred to silent miscoding.
Reference vintage
Every external classification is pinned to a specific upstream vintage. The current pinned vintages, refreshed quarterly with the rest of the Civica reference data:
- World Bank region + income — World Bank Country and Lending Groups, refreshed annually each July. Civica pulls the latest at the next quarterly cut.[5]
- V-Dem Regimes of the World — V-Dem dataset, refreshed annually each spring. The pinned-vintage label appears on every Civica surface that displays a V-Dem-derived cohort.[9]
- Bjørnskov-Rode / CGV — distributed via the Quality of Government dataset, refreshed annually each January.[12]
- government_form_description — CIA World Factbook, frozen January 2026. The Factbook is no longer actively maintained beyond that date; Civica may add a Wikidata cross-check in a future version, but for v1.0 the field is effectively static.[13]
How this methodology was decided
Civica adopted this peer-grouping architecture on 2026-05-02 after a multi-LLM deliberation panel rejected two alternatives: writing a methodology paper for the existing structural_familyheuristic, and keeping the heuristic with a disclaimer. The full audit trail — problem framing, three-option briefing, deliberation transcript, and unanimous resolution — is preserved in the planning archive under peer-grouping-resolution-v1.md and peer-grouping-deliberation-transcript.md.[14] [15]
Future maintainers, external reviewers, or readers who want to challenge the methodology should be able to see HOW Civica reached the decision, not just WHAT was decided. The audit trail is part of the deliverable.
Limitations
- V-Dem cadence. V-Dem updates annually. Intra-year regime transitions (a coup; a successful democratic transition) are not reflected in the peer-set classification until the next V-Dem release. Civica Pulse captures these events at daily cadence and presents them separately, but the peer-set tier itself lags by months.
- Non-sovereign jurisdictions. Coverage gaps (Taiwan, Kosovo, Palestine, Western Sahara, Vatican City) are documented above. Civica falls back to global comparison or marks the lens unavailable rather than silently mapping to a near-peer.
- government_form_description currency. CIA Factbook is frozen January 2026. Constitutional changes after that date are not reflected in the description until an alternative source pipeline is wired up. The field is descriptive metadata, not analytical taxonomy, so this staleness is bounded in impact.
- Single-lens defaults.Each domain has one default peer lens. Power users may want to compose lenses (“electoral democracies in Sub-Saharan Africa”) for finer-grained comparisons. Compound peer sets are not in v1.0; they are an explicit deferred enhancement.
Migration table
The full per-country mapping — old structural_familyvalues to new peer-lens fields, country by country — is published as a separate page at /civica-index/methodology/peer-grouping/migration. Replication-script maintainers can consume the same data as JSON via /api/v1/peer-groupings/migration.
The summary mapping below shows the typical replacement for each retired bucket. There are deliberately rows where the mapping is one-to-many or many-to-one — that is the point of the change. The legacy structural_family column and API field remain for two quarterly vintages with Deprecation + Sunset headers pointing at 2027-03-31; the hard cut lands on that date.
| Old structural_family bucket | Typical V-Dem RoW | Typical CGV regime | Typical region+income |
|---|---|---|---|
| parliamentary_democracy | Liberal / electoral democracy | parliamentary_democracy | Distributed across many region × income cohorts |
| presidential_republic | Liberal / electoral democracy / electoral autocracy | presidential_democracy / civilian_dictatorship | Distributed |
| semi_presidential | Mixed | semi_presidential_democracy | Distributed |
| constitutional_monarchy | Liberal / electoral democracy | parliamentary_democracy | Distributed (high-income Europe + Asia + Pacific) |
| absolute_monarchy | Closed autocracy | royal_dictatorship | Mostly Middle East / Gulf |
| one_party_state | Closed / electoral autocracy | civilian_dictatorship | Distributed |
| military_rule | Closed / electoral autocracy | military_dictatorship | Distributed (typically lower-income) |
| theocracy | Closed autocracy | civilian_dictatorship / royal_dictatorship | Iran, Vatican (special-case) |
| directorial_republic | Liberal democracy | parliamentary_democracy | Switzerland (n=1; descriptive metadata only) |
Versioning + changelog
- v1.0 (2026-05-02). Initial publication.
References
- Our World in Data peer-grouping conventions (ourworldindata.org).
- World Bank lending categories overview (datahelpdesk.worldbank.org).
- International Monetary Fund, World Economic Outlook country grouping (imf.org).
- Our World in Data, HDI peer-grouping convention (ourworldindata.org).
- World Bank, Country and Lending Groups (annual) (datahelpdesk.worldbank.org).
- Cheibub, Gandhi, and Vreeland (2010). Democracy and dictatorship revisited. Public Choice 143(1–2): 67–101. (doi.org/10.1007/s11127-009-9491-2)
- Bjørnskov, C. and Rode, M. (2020). Regime types and regime change: a new dataset on democracy, coups, and political institutions. The Review of International Organizations 15: 531–551. (doi.org/10.1007/s11558-019-09345-1)
- Lührmann, A., Tannenberg, M. and Lindberg, S. I. (2018). Regimes of the World (RoW): opening new avenues for the comparative study of political regimes. Politics and Governance 6(1): 60–77. (doi.org/10.17645/pag.v6i1.1214)
- V-Dem Institute, Varieties of Democracy dataset documentation (v-dem.net).
- CGV original codebook (Cheibub, Gandhi, Vreeland 2010 replication archive) (sites.google.com/site/joseantoniocheibub).
- Bjørnskov-Rode regime data archive (sites.google.com/view/martinrode).
- Quality of Government Standard Dataset, distributing the BR/CGV data (gu.se/en/quality-government).
- CIA World Factbook (frozen January 2026) (cia.gov/the-world-factbook).
- Peer-grouping resolution v1, internal Civica planning archive (
peer-grouping-resolution-v1.md). - Peer-grouping deliberation transcript, internal Civica planning archive (
peer-grouping-deliberation-transcript.md).
Cite this page
Cite this page
Civica. (2026). Civica Atlas Methodology — Peer grouping: Peer grouping. Civica Atlas. Retrieved May 7, 2026, from https://civicaatlas.org/civica-index/methodology/peer-grouping