Per ora questa pagina è in inglese.
Method paper · scoring rules 0.2 · revision 1
Scoring rules 0.2: method paper
NationsRelations scores how two countries relate along four dimensions (UN votes, trade, security and shared clubs) and combines them into a relationship index from −100 (opposed) to +100 (very close). The rules are fixed, the same for every pair, and written out in full here. No score is produced by a language model. This paper specifies scoring rules 0.2, which score 18,721 pairs of 194 countries (the 193 UN member states and the observer state of Palestine) from data loaded in October 2026. It gives the data, the formulas and their parameters, a worked example, what is deliberately not scored, how sure we are of each number, the checks that were made, and the limits.
2026-10-23 · rules released 2026-11-01 · scores built 2026-11-01 · Download the PDF · What changed · Licence CC BY-NC 4.0
This is an earlier version. The rules now in force are 0.4. This page never changes, so numbers cited from it stay explainable.
1 What changed from the previous version
- Security is no longer computed for a pair in which one of the two is a UN non-member observer State (Palestine). UCDP records conflicts between states; the fighting over the territory of Palestine between the government of Israel and armed groups is a conflict between a government and non-state groups, so under 0.1 every pair with Palestine scored 0 and read as “no armed conflict recorded”. Under 0.2 the score is empty and the page says why. It affects the security score of the 193 pairs with Palestine and no index (those pairs have no shared UN votes, so they had no index under 0.1 either).
2 Purpose and principles
The site answers one question for any two countries: how do they really get along? It does so in three voices. Say is what governments say about each other in their General Debate speeches at the UN. Do is what they do: how they vote, how much they trade, whether they fought on opposite sides or the same side, which clubs they share. Live is how their people are connected (migration, refugees, football).
- One rule for every country. A score is never adjusted for one pair; a change of rule applies to all pairs and gets a new version.
- Rate claims, not countries. The index describes data. The labels describe the data, not the countries or their governments.
- Fingerprint first, one number second. The four scores are the main view; the index is always shown next to its weights, which readers can change.
- Words are shown, not scored, until stance labels have been checked by people. The Live layer is never scored.
- No score comes from AI. A language model may answer a reader's question from one page's data, and every number it uses is checked against that data.
3 Data
| Source | Used for | Period |
|---|---|---|
| Nations on Record: UN General Assembly votes (Voeten et al., UN Digital Library) | Yearly agreement between countries; resolution-level votes for topics | 1946–2025 |
| UN General Debate speeches (Baturo, Dasandi and Mikhaylov; Nations on Record) | Sentences in which one country names another (258,844) | 1946–2025 |
| IMF International Trade in Goods, via Nations on Record | Goods exports in each direction, by year | to 2025 |
| Uppsala Conflict Data Program, Dyadic and Battle-Related Deaths, version 26.1 | Who fought on which side, and when | 1946–2025 |
| Ideal points (Bailey, Strezhnev and Voeten) | A voting position per country and year, with a 90% range (11,459 values) | to 2025 |
| Group memberships, checked by hand (Nations on Record) | Shared clubs and NATO | current |
| UN DESA International Migrant Stock 2024; UNHCR; international football results | Live layer only; never scored | various |
Russia includes the Soviet Union and Germany includes West Germany, as in Nations on Record. Observer states, such as Palestine, do not vote in the General Assembly, so some pairs have no votes score and therefore no index. Licences and loading dates are listed on the Data page.
4 Pairs, orientation and time windows
A pair is stored once, with the two ISO 3166-1 alpha-3 codes in alphabetical order; a page shows the pair from the side named first in its address. Votes and trade use the last 3 years with data (2022–2024 for votes, 2023–2025 for trade in this release). Armed conflict is read up to the last year of the UCDP data (2025). Speech counts use all sessions since 1946 and, as “recent”, the last 10 General Debates.
5 The four scores
UN votes, −100 to +100
For each pair we take the mean yearly agreement over the window, where yearly agreement is the agreement score of Voeten et al. on recorded votes on whole texts (final passage): the same vote counts 1, an abstention against a yes or a no counts ½, a yes against a no counts 0 (Lijphart's index; identical to Signorino and Ritter's S-score). Before 8 November 2026 this paper called it the share of votes on which the two cast the same vote; the computation has not changed. The score does not use that share directly. Most of the world votes alike, so a global ranking would make France and Germany +62 and Canada and the United States +6. Instead each country ranks its own voting partners by that mean:
position of B for A = (average ascending rank of B among A's partners − 1) ÷ (number of partners − 1) position of the pair = average of the position of B for A and the position of A for B votes score = round(200 × position of the pair − 100)
A position of 1 means the closest partner, 0 the furthest. Countries that voted alike share a rank (average rank). A pair with no votes in the window has no votes score.
Trade, 0 to 100
Goods trade between the two is the sum of exports in both directions, averaged over the window. Each country’s total trade is its exports plus imports, averaged over the same window. The score follows the larger of the two shares (the more dependent side), on a logarithmic scale, because a percentile would make 0.1% of Iran’s trade look like a middling tie:
trade score = round(100 × clip((log10(larger share) − log10(0.1%)) ÷ (log10(25%) − log10(0.1%)), 0, 1))
A share of 0.1% or less scores 0 and 25% or more scores 100. The window ends at the last year in which at least 80% of countries report. If either country has no trade totals there is no trade score.
Security, −100 to +100
From the Uppsala Conflict Data Program. For each year and pair we record the most direct relation: primary (both are main parties of the same conflict, on opposite sides), direct (one is a main party and the other sends troops to a side), proxy (both only send troops to the sides of a third party’s conflict, as peacekeeping contributors do). Proxy relations count for little as fighting and never as alliance. A relation fades with the years since the last data year:
| Years since the relation was last recorded | Weight kept |
|---|---|
| 0 to 4 | 1 |
| 5 to 14 | 0.5 |
| 15 to 29 | 0.2 |
| more than 29 | 0 |
opposed = largest of: 100 × kept (primary), 60 × kept (direct), 30 × kept (proxy) if opposed > 0: security score = −opposed (recent fighting on opposite sides outweighs any alliance) otherwise: security score = min(80, 50 if both are in NATO + 30 × kept (fighting side by side, primary or direct))
No security score is computed when one of the two is a UN non-member observer State (Palestine). UCDP records wars between states; a conflict over the territory of a country it does not list as a state, fought between a government and armed groups, is not one, so a score of 0 would wrongly read as “no armed conflict”. The pair page shows such conflicts as context.
Shared clubs, 0 to 100
25 points for each tracked group both belong to (EU, AU, ARAB, ASEAN, BRICS, G7, G20, GCC, P5; P5 is the five permanent members of the Security Council) and 25 for the same UN regional group, capped at 100. Membership lists are checked by hand.
6 The index
The index is the weighted average of the scores that exist, with default weights votes 35, trade 20, security 30, inst 15 (sum 100); readers can change the weights on every pair page and see the effect at once.
index = round( sum of (weight × score) ÷ sum of the weights of the scores that exist )
The index is shown only if both a votes score and a trade score exist; otherwise the pair is labelled “Not enough data”. Labels describe the number:
| Index | Label |
|---|---|
| +55 and above | Very close |
| +25 to +54 | Close |
| 0 to +24 | Neutral |
| −30 to −1 | Distant |
| below −30 | Opposed |
7 A worked example
India and Pakistan, as built on 2026-10-20. Inputs: for India, Pakistan ranks 5 of 191 voting partners; for Pakistan, India ranks 82 of 191. Goods trade is 0.0654% of India’s trade and 0.769% of Pakistan’s. UCDP records them opposed as main parties in 2025, the last data year. They share no tracked group but belong to the same UN regional group.
| Score | Calculation | Result |
|---|---|---|
| UN votes | positions 98 and 57; average 78; 200 × 0.776 − 100 | +55 |
| Trade | larger share 0.769%; on the log scale | 37 |
| Security | primary, last year: 100 × 1 | −100 |
| Shared clubs | 0 groups + same region = 1 × 25 | 25 |
index = (35×55 + 20×37 + 30×(−100) + 15×25) ÷ 100 = +0.4, shown as 0: Neutral
8 What is not scored
- Words. The sentences in which one country names another, and the keywords beside the name, are shown and counted. A keyword in the same sentence says nothing about who is accused of what, so words do not enter a score until stance labels have been checked by people.
- People. Migration, refugees and football are shown for context. People moving between two countries says nothing about whether their governments get along.
- Moments. The dated events on featured pairs are editorial context with a reference; they never enter a score.
9 Votes by topic
Overall agreement hides what separates two governments, so agreement is also counted per issue: disarmament, Middle East, human rights, colonialism, nuclear weapons, economic development, using the topic labels of the UN voting dataset. Only roll-call votes on whole texts count (amendments and single paragraphs are left out), yes, abstain and no are counted vote by vote with the same rule as the headline agreement (the same vote 1, an abstention against a yes or a no ½, a yes against a no 0; until 8 November 2026 an abstention counted as a full difference here). The window starts in 2010. Disarmament, colonialism and economic development are coded together by the source from 2019, so they count votes to 2018. A share needs at least 5 shared votes. The thin line on each bar is the 95% Wilson range for the number of votes; votes are not independent draws, so it understates the uncertainty.
10 Voting positions
Ideal points (Bailey, Strezhnev and Voeten) summarise all of a country’s recorded votes in one number per year, with a 90% range; higher means voting more like the United States and Western democracies. It is not a moral score. Two countries are called clearly apart on this dimension only if their 90% ranges do not overlap; if they overlap, their votes alone cannot tell the positions apart.
11 How sure we are
Each score carries a level, High, Medium, Low or None, that says how much data stands behind it. The level never changes a score.
| Score | High | Medium | Low | None |
|---|---|---|---|---|
| UN votes | 150 or more shared whole-text votes in the window | 50 to 149 | fewer than 50 | no shared votes |
| Trade | goods trade reported in all 3 years | in 2 years | in one year or none | no trade series |
| Security | always: UCDP covers every UN member state | – | – | pairs with a non-member observer State: not computed |
| Shared clubs | always: the list is fixed | – | – | – |
| Index | as sure as the weaker of votes and trade | no index |
150 shared votes give a sampling range of about ±8 points for a share near 50%, 50 about ±14.
12 Checks that were made
- Recomputation. The trade, security, shared-clubs and index scores of all 18,721 pairs were recomputed from the stored facts with the formulas above and equal the stored values; the votes score was reproduced to within 2.6 points by the simple rank formula (ties between countries explain the difference).
- Topics. For a test pair the numbers of shared votes and shares per topic equal those of the topic view of Nations on Record, which applies the same rules.
- Automated tests cover the scoring rules, the explanations, the confidence levels, the data downloads and the API.
- Editorial review. The thirty featured pairs were reviewed by the owner; hooks that did not match the data were corrected. Mistakes reported by readers are decided in public on the Corrections page.
13 Limits
- UN votes and speeches are what governments say in one forum. Bilateral diplomacy, treaties, embassies and sanctions are not in the data of this version.
- Trade covers goods, not services, investment or energy pipelines, in current dollars; customs data cannot tell zero trade from trade that was not reported.
- UCDP counts conflicts between state or organised parties with at least 25 battle-related deaths a year. Violence by groups without a government side does not count against a government.
- One index compresses a two-sided relationship; the trade score follows the more dependent side, and a pair can be close for one of them and distant for the other.
- The windows are short by design (three years), so a score reflects now rather than history; the yearly series on each pair page show the history.
- Source data are updated once a year when the General Assembly sessions are published; scores can move when the data move, and the version and build date say which data.
14 Reproducibility and versions
Every score on a page, in a file or in a PDF carries the rules version and the build date. The rules version changes only when a rule or parameter changes; a change applies to all pairs, is announced on the What changed page, and gets its own frozen page like this one. Page addresses never change. The code that builds the scores is deterministic, the source files and the database are backed up with a manifest of the rules version and loading dates, and the scores of all pairs can be downloaded.
15 How to cite
NationsRelations. (2026). Scoring rules 0.2: method paper (revision 1). https://nationsrelations.org/method/0.2. Rules released 2026-11-01; scores built 2026-11-01. Licence: CC BY-NC 4.0. A DOI is not yet assigned; one is planned with the first yearly release.
16 References
- Bailey, M. A., Strezhnev, A., and Voeten, E. (2017). Estimating dynamic state preferences from United Nations voting data. Journal of Conflict Resolution, 61(2), 430–456.
- Bailey, M., Strezhnev, A., and Voeten, E. United Nations General Assembly Ideal Points. Harvard Dataverse. doi:10.7910/DVN/LEJUQZ.
- Baturo, A., Dasandi, N., and Mikhaylov, S. J. (2017). Understanding state preferences with text as data: Introducing the UN General Debate corpus. Research and Politics, 4(2).
- Uppsala Conflict Data Program (UCDP). Dyadic Dataset and Battle-Related Deaths Dataset, version 26.1. Uppsala University. https://ucdp.uu.se
- United Nations, Department of Economic and Social Affairs, Population Division (2024). International Migrant Stock 2024.
- UNHCR Refugee Population Statistics. https://api.unhcr.org
- Jürisoo, M. International football results from 1872 to present. CC0.
- Wilson, E. B. (1927). Probable inference, the law of succession, and statistical inference. Journal of the American Statistical Association, 22(158), 209–212.
- Nations on Record. https://nationsonrecord.org