Method & limitations

How the score is built, what the weights rest on, and where this model is knowably wrong. No econometrics assumed.

How the model is computed

Four steps. No step ever invents a value that is missing.

1. Log-transform the skewed indicators

Some indicators — aid received, CO₂ per person — have one island whose value dwarfs everyone else's. Stretched straight onto 0–100, that island sits at the extreme and the other eleven bunch into a band where nothing separates them. So log(1+x) is applied first, compressing the upper tail. The ordering never changes; only the spacing becomes readable.

13 of the 36 entries in the dictionary go through this transform.

This is arithmetic, not island data: five illustrative values and where each lands on the 0–100 scale.

Min–max on the raw values
Min–max after log(1+x)

Before the transform, four of the five values sit inside the bottom 14% of the axis. After it, all five spread across it. The ordering is unchanged.

▸Table view

Normalised position before and after the log transform
Illustrative value	Min–max on the raw values	Min–max after log(1+x)
0.2	0.0	0.0
1	1.6	13.6
3	5.6	32.1
7	13.7	50.6
50	100.0	100.0

2. Min–max normalise to 0–100 across the 12-island panel, flipping where higher is worse

Each indicator is stretched so the panel minimum is 0 and the panel maximum is 100. Indicators where a higher value is worse — external debt, youth unemployment, export concentration — are subtracted from 100 to flip them. After this step every indicator reads “higher is better”.

Warning: A 0 means “lowest in this panel”, not “none”. A 100 means “highest in this panel”, not “complete”. Change the panel and every score changes. The bounds are computed once over the 12 islands and then frozen — neither the contrast case nor the shock simulator can move them.

3. A metric is the mean of its available sub-indicators — never imputed

A metric is built from three base indicators. If only two are present, the metric is the mean of those two and the coverage reads “2/3”. A missing value is never filled in from the panel mean or a neighbouring country. If all three are missing the metric has no score — not a score of zero.

4. The composite is the weighted sum, renormalised by the weight actually available

The nine metrics are multiplied by their 12 / 11 / 9% weights and summed, then divided by the weight of the metrics that actually have a score. With all nine present the denominator is 100. If a metric is entirely missing the denominator shrinks with it, so a gap can never be quietly scored as a zero.

Worked end to end

Real numbers for Fiji, recomputed by the API when this page was rendered rather than copied into the source.

Data confidence100.0%Coverage 3/3

a. Normalise the three base indicators — M1 · Food & primary production security

a. Normalise the three base indicators — Food & primary production security
Key	Name	Raw value	Unit	Direction	Log	Normalised
aff	Agriculture/forestry/fishing share of GDP (MVI AFF)	20.69	index 0-100	Higher is worse	—	44.8
ain	Instability of agricultural production (MVI AIN)	36.93	index 0-100	Higher is worse	—	57.4
fss	Fish Stock Status (EPI)	10.66	score 0-100	Higher is better	—	37.4

b. The mean is the metric score

(44.8 + 57.4 + 37.4) / 3 = 46.5

M1 · Food & primary production security = 46.52 · Coverage 3/3

c. Weight the nine metrics, sum, divide by the available weight

Per-metric score, weight and score × weight
Name	Value	Coverage	Weight	Score × weight
M1 · Food	46.52	3/3	12%	558.2
M2 · Energy	67.05	3/3	12%	804.6
M3 · Diversification	57.17	3/3	12%	686.0
M4 · Fiscal	51.99	3/3	12%	623.9
M5 · Digital	51.00	3/3	11%	561.0
M6 · Climate	40.96	3/3	12%	491.5
M7 · Ocean	55.90	3/3	9%	503.1
M8 · Human capital	68.89	3/3	11%	757.8
M9 · Dependence	34.60	3/3	9%	311.4
Total			100%	5297.6

5297.6 / 100 = 52.98

Composite resilience · Rank 5 · Weight available 100%

Averaging the same nine scores with equal weights gives 52.68, moving the island from rank 5 to rank 6. The next section is about that difference.

When the denominator is not 100 — Jeju

Contrast caseData confidence20.7%·22

M1 70.4 (1/3)
M2 No data (0/3)
M3 23.8 (1/3)
M4 76.9 (1/3)
M5 No data (0/3)
M6 No data (0/3)
M7 No data (0/3)
M8 97.9 (1/3)
M9 No data (0/3)

22 of Jeju's 28 base indicators are undefined at sub-national level. Only four of the nine metrics get a score, and the weight of those four is the denominator. So this is not “66.6 out of 100” — it is “66.6, computed from the 47% of the weight that exists”. It is never ranked.

Jeju is a contrast case and is never ranked

The nine metrics

There are only three weight levels. A weight tuned to the decimal place would look precise and rest on nothing.

Tier 1 - non-substitutable constraint · 12%
Tier 2 - enabling capacity · 11%
Tier 3 - discounted for partial double-counting · 9%

One bar, segments ordered by tier. A bar rather than a pie because length compares more accurately than angle.

The nine metrics by tier: id, name, weight, base indicators and rationale
ID	Name	Weight	Indicators	Rationale
Tier 1 - non-substitutable constraintTier total 60%Things with no short-run substitute: calories, fuel, the structure of export earnings, fiscal room, physical climate exposure. If one of these fails, the others cannot cover for it.
M1	Food & primary production security	12%	aff Agriculture/forestry/fishing share of GDP (MVI AFF)ain Instability of agricultural production (MVI AIN)fss Fish Stock Status (EPI)	No short-run domestic substitute for calories.
M2	Energy security & transition	12%	modern_renew Modern renewable share (constructed proxy)co2_pc CO2 emissions per capitaclean_cook Access to clean cooking fuels and technologies	Fuel cutoffs cascade into water, transport and the cold chain.
M3	Economic diversification & trade	12%	xcon Export concentration (MVI XCON)xin Export instability (MVI XIN)tour_xp Tourism receipts, % of total exports	Monoculture turns one sectoral shock into a whole-economy shock.
M4	Fiscal & financial capacity	12%	ln_gdppc Log GDP per capitaextdebt External debt stocks, % of GNIcab Current account balance, % of GDP	Decides whether a shock is absorbed or becomes a debt crisis.
M6	Climate & disaster resilience	12%	ndgain_vul ND-GAIN climate vulnerabilitylecz Population in low-elevation coastal zone (MVI LECZ)vic Victims of natural disasters (MVI VIC)	Direct existential exposure for low-lying islands.
Tier 2 - enabling capacityTier total 22%These matter, but a shock does not transmit through them in weeks. They are discounted for transmission speed.
M5	Digital infrastructure & innovation	11%	egi E-Government Development Indextii Telecommunication Infrastructure Indexgci Global Cybersecurity Index	Sets medium-run adaptive capacity; shocks do not transmit through it in weeks.
M8	Human capital & youth capacity	11%	hci Human Capital Index (UN DESA)hdi Human Development Indexyouth_unemp Youth unemployment, ages 15-24	Determines whether young people stay; slower transmission.
Tier 3 - discounted for partial double-countingTier total 18%Already partly captured inside other metrics. Discounted so the same information is not counted twice.
M7	Ocean & ecosystem health	9%	ohi Ocean Health Indexmpa Marine protected areas, % of territorial waterstpa_land Terrestrial protected areas (% of total land area)	Partly already captured inside food security; discounted to avoid double counting.
M9	External dependence & connectivity	9%	rem_idx Remoteness (MVI REM)remit Personal remittances received, % of GDPoda Net ODA received, % of GNI	Partly already captured inside fiscal and diversification; discounted.

Sensitivity check: is the weighting driving the result?

No. The panel was rescored with all nine metrics equally weighted (11.1% each). 10 of the 12 islands do not move at all and 2 change place. The largest shift is 1. The ranking is therefore a product of the data, not of the 12 / 11 / 9 tiering.

Rank and score under tier weighting and under equal weighting
Island	Region	Tier weighted	Equal weight	Move	Data confidence
Singapore	Asia (AIS)	1 · 72.56	1 · 72.15	0 No change	89.7%·3
Barbados	Caribbean	2 · 59.74	2 · 58.82	0 No change	89.7%·3
Mauritius	Indian Ocean (AIS)	3 · 58.88	3 · 58.30	0 No change	100.0%
Seychelles	Indian Ocean (AIS)	4 · 54.61	4 · 55.77	0 No change	93.1%·2
Fiji	Pacific	5 · 52.98	6 · 52.68	1 1 place down under equal weighting	100.0%
Jamaica	Caribbean	6 · 52.59	5 · 52.69	1 1 place up under equal weighting	89.7%·3
Cabo Verde	Atlantic (AIS)	7 · 46.50	7 · 45.90	0 No change	100.0%
Papua New Guinea	Pacific	8 · 44.54	8 · 44.45	0 No change	96.6%·1
Maldives	Indian Ocean (AIS)	9 · 41.78	9 · 42.80	0 No change	96.6%·1
Vanuatu	Pacific	10 · 35.61	10 · 34.72	0 No change	100.0%
Tuvalu	Pacific	11 · 33.16	11 · 33.16	0 No change	79.3%·5
Haiti	Caribbean	12 · 32.55	12 · 31.49	0 No change	100.0%

This is not evidence that the weights are right. It is evidence that the weights are not manufacturing the conclusion. The case for the tiering is the argument set out above, and it is an argument, not a statistic.

The five vulnerability channels (CVI)

The Cumulative Vulnerability Index measures how wide the channels are through which a shock arrives.

The five channels: weight, base indicators, note and orientation caveat
ID	Name	Weight	Indicators	Note
V1	Food & import supply	25%	aff Agriculture/forestry/fishing share of GDP (MVI AFF)fss Fish Stock Status (EPI)	Low domestic primary production + weak fishery = high import need.Orientation caveatInverted relative to the resilience layer. A large primary sector is a volatility risk (lowers M1) but also domestic calorie supply (lowers V1). Both readings are kept.
V2	Energy & fuel	20%	modern_renew Modern renewable share (constructed proxy)clean_cook Access to clean cooking fuels and technologies	Scored on MODERN renewable share, not raw renewable share - traditional biomass is not energy security.Orientation caveatEnters V2 as a capability, not as fuel independence. Clean cooking on a small island usually means imported LPG, so treating it as reduced energy dependence is the weakest link in this channel. Flagged, not hidden.
V3	External demand	20%	tour_xp Tourism receipts, % of total exportsxcon Export concentration (MVI XCON)xin Export instability (MVI XIN)	Concentration and volatility of the earnings base.
V4	Capital & external finance	20%	fdi FDI net inflows, % of GDPoda Net ODA received, % of GNIextdebt External debt stocks, % of GNIremit Personal remittances received, % of GDP	Reliance on externally sourced money.
V5	Trade logistics & remoteness	15%	rem_idx Remoteness (MVI REM)trade Trade (exports+imports), % of GDP	Amplifier of the other four; weighted lower to avoid double counting.Orientation caveatTrade openness is an amplifier here, not a shock: it measures how much of the economy has to physically cross an ocean.

The CVI is not the mirror of the resilience score

Deliberately so. Resilience measures the capacity to absorb a shock; the CVI measures the width of the channel the shock arrives through. The two axes share some base indicators but weight them differently, and a few indicators read in opposite directions on each. A large primary sector is a volatility risk (it lowers M1) and domestic calorie supply at the same time (it lowers V1). Define one axis as the inverse of the other and the resilience-versus-dependence quadrant becomes a tautology that says nothing. Both readings are kept for exactly that reason.

The simulator: the damage function

Two free parameters. Everything else is fixed by the data.

Loss_i = 0.3512 × Σ_c [ S_c × E_ic × ( 1 − 0.20 × C_ic / 100 ) ]

Loss_i: First-round GDP loss for island i, in % of GDP. No feedback, no time index.
scale: Global shock-to-loss scale. One of the two free parameters, fitted on the 2020 event.
Σ_c: Sum over the six shock channels c. No interaction between channels is modelled.
S_c: Shock intensity on channel c — the user's input, in real-world units (a 40% fall in tourism receipts, not an index).
E_ic: Island i's exposure to channel c, 0–1. Fixed by the data and never fitted.
C_ic: The capacity that damps channel c, 0–100, built from the island's own metric scores through the capacity map below.
κ: How much damping capacity can buy — the second free parameter. At the fitted value of 0.20, a capacity of 100 removes at most 20% of the loss. A buffer, not an exemption.

What exposure is measured from, channel by channel

What exposure is measured from, channel by channel
Shock channel	How exposure E is defined
Tourism demand% fall in tourism receipts	Tourism receipts as a share of total exports, used directly rather than normalised: an 85% share means 85% of export earnings are at risk.
Fuel price% change in fuel import cost	Share of final energy that is NOT modern renewable, i.e. 100 - modern_renew. Traditional biomass does not count as energy independence.
Food price% change in food import cost	Low domestic primary production (MVI AFF, inverted) plus weak fishery (EPI Fish Stock Status, inverted).
External capital% tightening of external finance	Reliance on externally sourced money: external debt, remittances and FDI.
Freight & logistics% rise in freight cost / lead time	Remoteness (MVI REM) times trade openness — how much of the economy has to cross an ocean.
Physical climatecyclone / sea-level stress index	Direct climate exposure: low-elevation coastal population, disaster victims, and ND-GAIN vulnerability.

What it was fitted on

Correlation r 0.816
R² 0.666
Leave-one-out MAE 4.24 pp of GDP
Sample size 10 islands
Rank MAE 1.40

Fitted on the 2020 COVID event, which is a shock VECTOR - fuel moved favourably for importers while tourism collapsed. Only two parameters are free (global scale, damping exponent kappa); all exposures are fixed by the data. Leave-one-out mean absolute error is 4.24 percentage points of GDP. Jamaica and Tuvalu are excluded from the fit because tourism share of exports is unavailable for them.

Excluded from fit: Jamaica, Tuvalu — The islands excluded from the fit are the ones with no tourism share of exports, so no exposure could be built for them. The exclusion is stated rather than absorbed.

Two maps, and why they are not the same map

Which metrics DAMP a shock — fitted

Fitted machinery, carried over unchanged from the research code. The headline GDP-loss figure comes from this map.

Capacity map by shock channel
Shock channel	Metrics and share
Tourism demand	M3 Diversification 0.60M4 Fiscal 0.40
Fuel price	M2 Energy 0.60M4 Fiscal 0.40
Food price	M1 Food 0.70M4 Fiscal 0.30
External capital	M4 Fiscal 0.70M5 Digital 0.30
Freight & logistics	M9 Dependence 0.50M5 Digital 0.30M6 Climate 0.20
Physical climate	M6 Climate 0.70M4 Fiscal 0.30

Which metrics a shock DEGRADES — not calibrated

A presentation-layer derivation of the transmission map in the project spec. It exists to answer “which sector fails first” and has never been calibrated against anything.

Damage map by shock channel
Shock channel	Metrics and share
Tourism demand	M3 Diversification 0.55M4 Fiscal 0.25M8 Human capital 0.20
Fuel price	M2 Energy 0.60M3 Diversification 0.20M4 Fiscal 0.20
Food price	M1 Food 0.70M8 Human capital 0.15M4 Fiscal 0.15
External capital	M4 Fiscal 0.60M5 Digital 0.20M8 Human capital 0.20
Freight & logistics	M9 Dependence 0.50M3 Diversification 0.25M1 Food 0.25
Physical climate	M6 Climate 0.60M7 Ocean 0.20M1 Food 0.20

Warning: Conflating the two would have been easier and wrong: the metric that best absorbs a fuel shock is not necessarily the metric a fuel shock damages most. The GDP-loss figure is the calibrated output; the composite delta is explanation. They do not deserve equal trust.

Known limitations

Set out unsoftened. Most of these are faults where the fix is known and the data required for it does not exist.

High
The model is a structural snapshot of 2018-2021, not a 2026 nowcast
Indicator vintages are whatever the UNDP SIDS Data Platform publishes. Do not read any score as a current-year measurement.
High
Protected-area indicator is mislabelled at source
WDI ER.PTD.TOTL.ZS is terrestrial AND marine protection as a share of total territorial area, not terrestrial protection as a share of land. M7 therefore partly double-counts marine protection, and terrestrial protection is measured nowhere.
High
Four columns mix reference years
cab, tour_xp, oda and trade span 2010-2020 across islands. tour_xp is the worst (9-year spread) and is also the most load-bearing variable in the model - Jamaica's value is from 2011.
High
Jeju is a contrast case, not a ranked island
22 of 28 indicators are undefined at sub-national level and the rest inherit Korea's sovereign credit, national grid and national digital infrastructure. Jeju is shown separately and never ranked against the SIDS panel.
High
Food-dependence proxy inverts for subsistence atolls
V1 reads domestic food capacity from MVI AFF, the primary sector's share of GDP. In a tiny subsistence economy that share is large while absolute calorie output is negligible, so Tuvalu scores as the panel's LEAST food-import-dependent island when it in fact imports almost all staple food. Read V1 for Tuvalu, Vanuatu and PNG with this in mind. A real fix needs a calorie self-sufficiency or food-import-share indicator, which the SIDS platform does not publish.
Medium
The fiscal metric still uses external debt, not government debt
The single-vintage IMF replacement is missing for PNG, SGP, SYC, TUV and VUT. Switching over with five blanks would recreate the original high-income bias with different islands, so the switch is deliberately not made.
Medium
CO2 per capita rewards poverty
Low emissions from low income read as energy virtue. This holds Haiti's energy score at 50 rather than lower. A standard flaw in indices of this type.
Medium
Recovery time is a three-way band, not an estimate
Only 8 of 12 islands recovered, taking just three distinct values, and four are right-censored. Depth does not predict speed. A continuous estimate would need a wider panel and survival analysis.
Medium
The simulator is first-round only
Loss does not feed back into the composite, there is no time index, and it is calibrated on a single event (2020) with n=10.

Data & downloads

Every entry in the snapshot, including derived inputs and reference series that do not enter the score.

Indicator dictionary

Indicator names and source notes stay in the language the statistical sources publish them in. Translating them and hiding the original would make the model harder to audit, not easier.

Filter by metricSearchShowing 36 of 36

Indicator dictionary: key, name, unit, direction, log transform, derived flag, metric, channels and source
Key	Name	Unit	Direction	Log-transformed	Derived	Metric	Channels
aff	Agriculture/forestry/fishing share of GDP (MVI AFF)	index 0-100	Higher is worse	—	—	M1 · Food	V1 · Food & import supply
Source: UNDP SIDS Platform / MVI (LDC EVI component). High share = economy leans on a volatile, weather-exposed primary sector.
ain	Instability of agricultural production (MVI AIN)	index 0-100	Higher is worse	—	—	M1 · Food	—
Source: UNDP SIDS Platform / MVI. Year-to-year volatility of domestic food & crop output.
fss	Fish Stock Status (EPI)	score 0-100	Higher is better	—	—	M1 · Food	V1 · Food & import supply
Source: UNDP SIDS Platform / Blue block, Environmental Performance Index. Health of the fishery the island eats from.
renew	Renewable energy, % of final energy consumption	%	Higher is better	—	—	—	—
Source: UNDP SIDS Platform / Climate block (WDI EG.FEC.RNEW.ZS, 2015).
co2_kt	CO2 emissions	kilotonnes	Higher is worse	—	—	—	—
Source: UNDP SIDS Platform / Climate block (WDI EN.ATM.CO2E.KT, 2018). Converted to tonnes per capita in the model.
xcon	Export concentration (MVI XCON)	index 0-100	Higher is worse	Yes	—	M3 · Diversification	V3 · External demand
Source: UNDP SIDS Platform / MVI. How few products/services the island's export earnings rest on.
xin	Export instability (MVI XIN)	index 0-100	Higher is worse	Yes	—	M3 · Diversification	V3 · External demand
Source: UNDP SIDS Platform / MVI. Volatility of export earnings.
tour_xp	Tourism receipts, % of total exports	%	Higher is worse	Yes	—	M3 · Diversification	V3 · External demand
Source: UNDP SIDS Platform / MVI block (WDI ST.INT.RCPT.XP.ZS). Treated as monoculture risk, not as income.
gdppc	GDP per capita	current US$	Higher is better	—	—	—	—
Source: UNDP SIDS Platform / Finance block (WDI NY.GDP.PCAP.CD, 2020). Log-transformed in the model.
extdebt	External debt stocks, % of GNI	%	Higher is worse	Yes	—	M4 · Fiscal	V4 · Capital & external finance
Source: UNDP SIDS Platform / Finance block (WDI DT.DOD.DECT.GN.ZS, 2019).
cab	Current account balance, % of GDP	%	Higher is better	—	—	M4 · Fiscal	—
Source: UNDP SIDS Platform / Finance block (WDI BN.CAB.XOKA.GD.ZS).
egi	E-Government Development Index	index 0-1	Higher is better	—	—	M5 · Digital	—
Source: UNDP SIDS Platform / Digital block (UN DESA EGDI, 2020).
tii	Telecommunication Infrastructure Index	index 0-1	Higher is better	—	—	M5 · Digital	—
Source: UNDP SIDS Platform / Digital block (UN DESA TII, 2020).
gci	Global Cybersecurity Index	score 0-100	Higher is better	—	—	M5 · Digital	—
Source: UNDP SIDS Platform / Digital block (ITU GCI, 2020).
ndgain_vul	ND-GAIN climate vulnerability	index 0-1	Higher is worse	—	—	M6 · Climate	—
Source: UNDP SIDS Platform / Climate block (Notre Dame GAIN, 2019). Higher = more exposed/sensitive, less adaptive.
lecz	Population in low-elevation coastal zone (MVI LECZ)	index 0-100	Higher is worse	—	—	M6 · Climate	—
Source: UNDP SIDS Platform / MVI. Share of people living <5 m above sea level.
vic	Victims of natural disasters (MVI VIC)	index 0-100	Higher is worse	Yes	—	M6 · Climate	—
Source: UNDP SIDS Platform / MVI. Population affected by disasters, indexed.
ohi	Ocean Health Index	score 0-100	Higher is better	—	—	M7 · Ocean	—
Source: UNDP SIDS Platform / Blue block (OHI, 2020).
mpa	Marine protected areas, % of territorial waters	%	Higher is better	Yes	—	M7 · Ocean	—
Source: UNDP SIDS Platform / Blue block (WDI ER.MRN.PTMR.ZS, 2018).
tpa	Terrestrial AND marine protected areas (% of total territorial area)	%	Higher is better	Yes	—	—	—
Source: WDI ER.PTD.TOTL.ZS, 2018. RETAINED FOR REFERENCE ONLY - not used in M7. Denominator is total territorial area including waters, so it overlaps `mpa` and is not a land-protection measure despite its original label.
hci	Human Capital Index (UN DESA)	index 0-1	Higher is better	—	—	M8 · Human capital	—
Source: UNDP SIDS Platform / Digital block (UN DESA HCI, 2020). Schooling + literacy composite.
hdi	Human Development Index	index 0-1	Higher is better	—	—	M8 · Human capital	—
Source: UNDP SIDS Platform / Profile block (UNDP HDR, 2019).
youth_unemp	Youth unemployment, ages 15-24	%	Higher is worse	—	—	M8 · Human capital	—
Source: World Bank WDI SL.UEM.1524.ZS (ILO modelled estimate, 2024). SUPPLEMENTARY - not on the SIDS platform.
rem_idx	Remoteness (MVI REM)	index 0-100	Higher is worse	—	—	M9 · Dependence	V5 · Trade logistics & remoteness
Source: UNDP SIDS Platform / MVI. Distance-weighted trade remoteness from world markets.
remit	Personal remittances received, % of GDP	%	Higher is worse	Yes	—	M9 · Dependence	V4 · Capital & external finance
Source: UNDP SIDS Platform / Finance block (WDI BX.TRF.PWKR.DT.GD.ZS, 2020). Treated as external-income dependence.
oda	Net ODA received, % of GNI	%	Higher is worse	Yes	—	M9 · Dependence	V4 · Capital & external finance
Source: UNDP SIDS Platform / Finance block (WDI DT.ODA.ODAT.GN.ZS).
fdi	FDI net inflows, % of GDP	%	Higher is worse	Yes	—	—	V4 · Capital & external finance
Source: UNDP SIDS Platform / MVI block (WDI BX.KLT.DINV.WD.GD.ZS, 2019). Used in the dependence layer only.
trade	Trade (exports+imports), % of GDP	%	Higher is worse	Yes	—	—	V5 · Trade logistics & remoteness
Source: UNDP SIDS Platform / Finance block (WDI NE.TRD.GNFS.ZS). Openness = exposure in the dependence layer.
tpa_land	Terrestrial protected areas (% of total land area)	%	Higher is better	—	—	M7 · Ocean	—
Source: WDI ER.LND.PTLD.ZS, 2018. Added 8 Aug 2026. True land-protection measure; replaces `tpa` inside M7 Ocean & ecosystem health.
co2_pc	CO2 emissions per capita	tonnes/person	Higher is worse	Yes	Yes	M2 · Energy	—
Source: Derived: co2_kt x 1000 / population.
ln_gdppc	Log GDP per capita	ln(current US$)	Higher is better	—	Yes	M4 · Fiscal	—
Source: Derived: natural log of GDP per capita, to stop Singapore compressing the panel.
gdppc_preshock	Pre-shock (2019) GDP per capita	current US$	Higher is better	—	Yes	—	—
Source: Derived: 2020 GDP per capita backed out by the 2020 real growth rate.
pop_density	Population density	people/km2	Higher is worse	—	Yes	—	—
Source: Derived: population / surface area. NB the denominator is surface area, not land area.
clean_cook	Access to clean cooking fuels and technologies	% of population	Higher is better	—	Yes	M2 · Energy	V2 · Energy & fuel
Source: WDI EG.CFT.ACCS.ZS, 2022 (Our World in Data / Tracking SDG7). Added to correct the traditional-biomass problem.
modern_renew	Modern renewable share (constructed proxy)	%	Higher is better	Yes	Yes	M2 · Energy	V2 · Energy & fuel
Source: CONSTRUCTED PROXY, not an official statistic: renewable share x clean cooking access / 100. Strips fuelwood and charcoal out of the headline renewable share.
govdebt_2024	General government gross debt	% of GDP	Higher is worse	—	Yes	—	—
Source: IMF WEO 2024. INCOMPLETE - missing for PNG, SGP, SYC, TUV, VUT, so the fiscal metric still uses external debt.

CSV downloads

Provenance

Every number on this site is recomputed at request time from the snapshot. Nothing is read from a stored table of pre-computed results. The snapshot itself is a file generated by the research pipeline in artifacts/build, and the service invents no value outside it. The service therefore cannot drift from the research artefacts: drifting would require the snapshot to change, and then both move together.

Snapshot version: 1.0.0
Data vintage: 2018-2021 (structural snapshot, not a nowcast)
Primary source: UNDP SIDS Data Platform - https://sids.data.undp.org/
Generated from: artifacts/build (raw_data.py, panel_build.py, score.py, sim_v3.py)
Model: Island Resilience & Sustainability Model (IRSM)

This model is a structural snapshot of 2018–2021. It is not a present-day estimate.