Methodology
How the Global Radar works
Every number on the Global Radar comes from one open file and a few explicit rules. This page lists all of them, so any result can be checked by hand.
1. Data source
World Bank, World Development Indicators (WDI), extract dated 2026-10-01 (file health_equity_wdi_latest.json): for each country/economy and indicator, the latest available observation. The site reads a compact copy built from that file by scripts/build_global.py; the build checks that every value and year matches the original (… records, 0 mismatches). WDI compiles data from original producers (UN Population Division, WHO, UNICEF, UN IGME, ILO, UNESCO UIS, WHO/UNICEF JMP, the World Bank and others); each indicator's source is listed below. Restricted microdata (for example WHO STEPS) are not used.
2. Indicator selection
… indicators were selected for a health-equity lens and screened by licence (section 9). They are grouped into five product domains. "Disease burden" contains a single indicator (tuberculosis incidence); the prototype does not measure the overall burden of noncommunicable diseases.
| Indicator | Domain | Direction | Countries | Latest years | Source |
|---|
Direction is defined explicitly in the code (no automatic guessing): Higher = concern — under-5, infant and maternal mortality, tuberculosis incidence, out-of-pocket share of health spending, poverty headcount, unemployment. Lower = concern — life expectancy, UHC service coverage index, physicians, nurses and midwives, measles immunization, secondary school enrolment, safely managed drinking water and sanitation, access to electricity. Context (shown, never judged) — GDP per capita, health expenditure (% of GDP and per capita), crude birth and death rates (driven by age structure), fertility, population aged 65+, population, urban share.
3. Latest available year
Values are the latest available observation, not the current situation. Years differ by indicator and by country (for example, life expectancy is 2024 for all economies, while physician density ranges from 1980 to 2023). The interface shows the year next to every value with a freshness label: Recent 2024–2025 · Moderate 2022–2023 · Caution 2015–2021 · Old before 2015 — old values are displayed but excluded from comparisons, signals and the score. The data are not real-time.
4. Missing data
A missing value is shown as "No data available". It is never replaced by zero, imputed or estimated. A country without a value is left out of that indicator's comparison group; the score uses only indicators the country has (with a minimum, section 7).
5. Comparison method
For each indicator, a country's value is placed among countries in the same World Bank income group (as recorded in the dataset) that have a value from 2015 or later. If fewer than 10 such peers exist, all countries are used instead. A worldwide position is always shown as well.
The position is a mid-rank percentile: the share of peers with a more favourable value, counting ties as half. It is computed with the same function as the Korean Health Equity Radar (unfavorablePercentile in app/src/lib/equity/calculateGap.js). Peers' values may come from different years; the range of their years is shown with each explanation.
6. Priority Signal calculation
- Priority signal: less favourable than at least 80% of income-group peers with data.
- Watch: less favourable than 60–80%.
- Only the 16 directional indicators can signal; context indicators never do.
- Exclusion: poverty headcount at $3.00/day is shown but not used for signals or the score for high-income economies — values are near zero for most of them, so small differences are not meaningful at this line.
- Up to three signals are highlighted, preferring different domains. "Why this signal?" shows the value, year, peer count, peer median, peer years and the worldwide position.
- Possible action areas come from a rule file (
global/data/action_rules.json) and appear only for highlighted priority signals. They are areas to review, not prescriptions, and their suitability depends on national context.
7. Exploratory Priority Score
The score is the simple average of a country's relative positions (0–1) across the directional indicators it has, multiplied by 100. All indicators have equal weight; domain sub-scores are averages within each domain. No score is shown if fewer than 8 directional indicators are available. 0 means the most favourable positions among peers and 100 the least favourable.
This exploratory score is intended for screening and prioritization, not clinical or causal inference. It is not an official index, does not rank health systems, and is sensitive to which indicators are available for each country.
8. Limitations
- National averages hide differences within countries — the core of health equity. This prototype screens between countries; within-country analysis (as in the Korean implementation) needs sub-national data.
- Many values are modelled estimates (for example maternal mortality, under-5 mortality, unemployment) with their own uncertainty, which WDI does not carry in this extract.
- Comparing different observation years can mislead; years are always displayed and old values excluded, but moderate/caution years remain in comparisons.
- Income-group peers are a pragmatic comparison set, not a judgement that countries should have similar results.
- Poverty at $3.00/day is near zero in most high-income economies; it is therefore excluded from their signals and score (section 6).
- Relative position is not a cause, a prediction or an evaluation of any policy or government.
- Historical trends use each country's own observed years, which differ between countries and indicators; a short or irregular series can look flat or steep by chance. Trends are descriptions, not explanations (section 10).
9. License and attribution
Indicators included in this prototype were screened using the licence metadata supplied in the World Development Indicators dataset: only series whose WDI License Type field reads CC BY-4.0 were kept (all …). This is a metadata-level screening, not a legal opinion, and it does not mean that every World Bank dataset may be used without conditions; third-party terms of original producers may apply. Source and attribution information are retained for each indicator (table above and in the country view).
Attribution: World Bank, World Development Indicators (CC BY 4.0), with original sources as listed per indicator. Map boundaries: Natural Earth (public domain); boundaries do not imply any judgement on the legal status of any territory. This prototype is not affiliated with or endorsed by the World Bank, WHO or any data provider.
10. Historical trends (v0.2)
Data. A second file from the same WDI extract (health_equity_wdi_timeseries_2000_latest.csv, … observed values, … economies, … indicators, …). The full file is not published; the build (scripts/build_global_trends.mjs) writes one small file per economy with its observed points exactly as recorded. Before writing, it checks that the last observation of every country–indicator series equals the latest value used above (… of … series matched, … mismatches; the remaining … latest values were observed before 2000 and have no series). SHA-256 of the file: ….
- Observed years only. A year without an observation stays empty. Nothing is interpolated, carried forward, smoothed or estimated, and a missing value is never 0. Charts draw a line only between consecutive observed years; a gap in the line means no observation.
- Recent trend. A straight line (ordinary least squares) through the observed values in the 10 years up to and including the country's last observation, only when at least 4 years were observed in that window (Korean engine function
olsSlope). Because last observation years differ, windows differ between countries; each window is displayed. - Classification (signal indicators only, last observation 2015 or later): Improving or Worsening when the 95% range of the slope lies entirely on the more or less favourable side of zero; otherwise No clear change. This rule does not depend on the indicator's scale. The high-income poverty exclusion of section 6 also applies here. …
- Compared with peers. The slope is also placed among the slopes of the same income group (all economies if fewer than 10), using the Korean engine's
trendClasstertiles: most favourable, middle or least favourable third. - Long-term change. The difference between the first observation since 2000 and the last observation, with both years shown.
- Peer line. For each year, the median of the economies in the income group that were observed in that year; years with fewer than 10 observing economies are omitted, and the number of economies is shown on hover.
- Kept apart from signals. Priority signals, Watch and the exploratory score still use the latest available value only (sections 5–7), so they are identical to v0.1. The trend is shown next to the current position, never merged into it. The "Trend watch" list shows signal indicators classified as Worsening and is not part of the signals or the score.
- Not causal. Trend describes past observed values; it does not explain why they changed or predict future values. Many series are modelled estimates that are revised as a whole, so a past value may change in later WDI releases.
11. Evidence links (v0.2)
Each possible action area (section 6) can list public sources in global/data/evidence_registry.json: identifier, title, publisher, address, type (fact sheet, guideline, report, technical document, topic page), language, the date it was checked and the relevance-review status. scripts/verify_global_evidence.py opens every address and records the HTTP status, the page title it saw and whether the listed key terms appear on the page; a source appears on the site only if all three checks pass and the address did not redirect elsewhere. Addresses were added only after they were opened and checked; none are guessed. Currently … sources are shown, linked to … of … action areas; areas without a checked source show "No verified evidence linked yet".
A checked link confirms that the document exists at that address with that title — not that it supports a particular action in a particular country. Relevance to each area has not yet been reviewed by a domain expert, and listing a source does not mean its publisher endorses this prototype.
About Health Equity Radar
Health Equity Radar transforms fragmented public-health data into interpretable signals for priority setting. The Republic of Korea is the initial implementation (health-profile.kr, sub-national official statistics). The Global prototype tests whether the same decision framework can operate across countries using standardized open data.