Methodology

1. Data sources

Pan Africa Data aggregates national data from reputable international institutions, cleans and standardises it across all 54 African countries, and builds proprietary sub-national extensions on top.

2. Value type definitions

Every record returned by the API and included in downloads carries a value_type field. This tells you exactly how the figure was produced.

3. Forecast methodology

Standard global forecasting approaches — simple trend lines, single global inflation floors, unemployment trends anchored on crisis years — produce systematically wrong results for African economies. Pan Africa Data's forecast model addresses five structural issues specific to African data.

Inflation

Hyperinflation spike years are detected using a country-specific threshold and excluded from trend calculations. Forecasts then converge toward central bank targets where one exists:

For countries without a central bank target, a global floor of 0.5% is used rather than 2.0% — this prevents CFA franc zone countries with structurally low inflation from being artificially lifted.

Unemployment

Forecasts revert toward country-specific structural rates rather than extrapolating from COVID-era spikes. South Africa's structural rate is 28.0%; Nigeria 5.0%; Kenya 5.5%; Ghana 4.5%; Egypt 7.5%; Morocco 10.0%; Tunisia 15.0%. Eleven countries with available structural rate estimates use this mean-reversion approach; all other countries use trend extrapolation.

Gini coefficient

Country-specific Gini floors prevent forecasts from converging toward implausibly low inequality levels. High-inequality Southern African economies (South Africa, Namibia, Botswana, Eswatini, Lesotho, Zimbabwe) retain elevated floors reflecting structural inequality, with gradual improvement modelled over the forecast horizon.

4. GDP in three currencies

GDP per capita is provided in three formats:

GDP total

GDP total (current US$) for 2025–2030 is taken directly from IMF WEO. The 2031–2035 extension applies a growth rate that converges linearly from the 2030 IMF growth rate toward a country-specific long-run potential growth rate, reflecting differing structural growth potential across the continent — resource-driven and frontier economies typically converge toward higher long-run rates than mature, diversified economies.

5. Sub-national income distribution

Pan Africa Data has constructed proprietary sub-national income distribution data across two geographic layers below the national level — city (metropolitan) and locality/suburb. The city layer resolves to 338 metropolitan areas; the locality/suburb layer resolves to 2,044 individual localities, across 49 African countries. Coverage includes every locality with a population of 50,000 or more, based on the United Nations World Urbanization Prospects 2025 revision. (Eritrea, Libya and Somalia are excluded for lack of underlying data, and Seychelles and Eswatini have no locality meeting the inclusion threshold — see limitations below.) To our knowledge, this is the only standardised Africa-wide dataset providing income distribution at both the functional-city and locality/suburb level across a 5-class taxonomy.

For each geography, five income classes are provided: marginalised, low income, middle income, upper-middle income, and high income. Each class includes population share, population count, and income bounds in current USD, constant USD, and PPP, plus the Gini coefficient. Cities and localities currently carry their country's national Gini coefficient.

How sub-national figures are derived

The model starts from national income distribution anchored to World Bank WDI measured data. Sub-national income relativities — derived from GDP per capita estimates, urbanisation rates, population density and economic activity — are then applied to produce income bounds and class shares at both city and locality/suburb level.

Sub-national geographies currently carry the national Gini coefficient of their country. Differences between cities and localities in the same country come from their income relativities, which shift each geography's class shares, not from separate city-level inequality estimates.

Forecasts to 2035 use the same national forecast methodology described in section 3, with sub-national relativities held stable over the forecast horizon.

PPP framework — a note on vintage

Pan Africa Data currently uses the World Bank's 2017 PPP framework, with poverty-line anchors at $2.15 (extreme), $3.65 (lower-middle-income) and $6.85 (upper-middle-income) per day. In June 2025 the World Bank released updated 2021 PPP thresholds of $3.00 / $4.20 / $8.30. We have deliberately retained the 2017 PPP framework for two reasons.

First, our primary calibration source — the World Bank Macro Poverty Outlook (MPO), used to anchor country-level poverty forecasts — continues to publish country projections at the $2.15 / $3.65 / $6.85 lines as of the latest available release (April 2026). Migrating our thresholds to 2021 PPPs before MPO does would produce internally inconsistent calibration, where the forecast anchor and the class boundaries reference different PPP vintages.

Second, the majority of African country-level poverty statistics reprocessed under 2021 PPPs remain preliminary in the World Bank's Poverty and Inequality Platform (PIP), and the 2017 PPP framework preserves comparability with the body of African poverty research published between 2022 and 2025.

6. Household vs per capita income

Both per capita and household income bounds are provided for every city and income class.

Household income bounds are derived by multiplying per capita income bounds by a regional household size coefficient, calibrated using available household survey data by region — household sizes vary meaningfully across Africa, and East African urban households differ structurally from West African urban households, for example.

7. Known data limitations

GDP per capita historical values for pre-2010 periods may diverge from household survey-based estimates due to methodological differences between national accounts data and consumption survey approaches. Pan Africa Data uses national accounts (WDI) as the primary source for consistency and comparability across all 54 countries.

8. Update schedule

National data is refreshed twice a year, in April and October, aligned to the IMF WEO and World Bank MPO releases. Sources are World Bank WDI, IMF WEO, World Bank MPO, UN Population Division, OECD DAC and World Bank RPW. Sub-national data (city and locality/suburb) is refreshed alongside each national update.

Questions about methodology, data sources, or a specific figure? [email protected]

9. Validation

Pan Africa Data's model is validated on three dimensions — calibration accuracy against the World Bank's own poverty benchmarks, city-level population fidelity to authoritative UN sources, and internal integrity of the distributional outputs. Validation is ongoing rather than a one-time launch check.

Calibration against WB MPO poverty benchmarks

The model is calibrated against World Bank Macro Poverty Outlook country-level poverty rates at the $3.65/day threshold. Of the 51 African countries with WB MPO calibration coverage, 44 land at exact zero-percentage-point gap between the modelled marginalised share and the WB MPO benchmark. Three upper-middle-income economies (South Africa +8.2pp, Botswana +4.7pp, Equatorial Guinea +3.4pp) sit above the $3.65 line by design, because their income structure is calibrated to the $6.85/day upper-middle-income poverty line rather than the $3.65/day lower-middle-income line. Four low-population economies (Gabon, Mauritius, Cabo Verde, Seychelles) sit within a ≤2pp watch-level tolerance. A zero gap at the calibration line is a check that the calibration was applied faithfully, not independent validation; independent validation is reported on the Validation page. Two countries (Algeria, Senegal) have no WB MPO POV365 coverage — a genuine World Bank data gap disclosed in the country limitations table.

City-level populations sourced directly from UN WUP2025

The 2,044-locality Layer-3 population base is sourced from the United Nations World Urbanization Prospects 2025 revision (DEGURBA methodology, cities ≥50,000 population). A pipeline integrity check confirms faithful preservation of the UN source across the full 2000–2035 horizon — across every city and year sampled, database populations match the UN WUP2025 series to within a single person on values ranging from 279,000 (Maseru) to 26.2 million (Greater Cairo). Representative rows:

Layer-3 populations reflect the UN's DEGURBA cluster boundaries — Johannesburg 6.6M represents the DEGURBA cluster, not the wider functional-city definition. Layer-2 functional cities (Johannesburg ~11.5M, for example) aggregate multiple DEGURBA clusters into the metropolitan economic unit — see section 5 for the layer definitions.

Functional-city size reconciliation against Statistics South Africa Census 2022

For Layer-2 functional cities, Johannesburg reconciles at 1.03–1.06× the Census population (depending on functional boundary definition), Tshwane at 0.92× (Census, Soshanguve-inclusive), and Durban at 1.05×.

Class-share integrity

For every country, every geography, and every forecast year, the five income class shares must sum to 100%. A production integrity check across 51 countries × 11 forecast years × 3 administrative levels — 561 country-year-level combinations in total — records zero violations.

Country-level income distribution — Angola worked example

For Angola, the World Bank's 2018 official survey records a marginalised share of 52.9% at a Gini of 51.3. Pan Africa Data's 2019 modelled result gives 58.5% at a Gini of 50.83 — a ~5.6pp overstatement of marginalised population against an almost-identical Gini. The direction of the miss is conservative (over-poverty rather than under-poverty), and part is attributable to timing (Angola's economy deteriorated between 2018 and 2019).

Model responsiveness to macro shocks — Egypt worked example

Between 2021 and 2024, Egypt experienced substantial EGP devaluation. Pan Africa Data's marginalised share for Egypt rises sharply across the same window (7.1% in 2021 to 12.6% in 2024) and then plateaus — capturing the real welfare impact of the currency shock. Static extrapolations from pre-crisis baselines would miss the shift entirely.

Cross-source reference check

Beyond the World Bank and UN benchmarks named above, Pan Africa Data's national income distributions have been sense-checked against an independent third-party commercial reference dataset across 11 African countries. That comparison was used purely as a post-calibration reference check, not as a calibration input. The World Bank remains the sole calibration anchor because it is a live, publicly-verifiable, actively-maintained release. Where Pan Africa Data and the third-party reference diverged, differences were traced to classification-methodology differences (e.g. different class-boundary definitions) rather than treated as calibration errors to correct against.

Ongoing validation

The model is re-calibrated against every new World Bank MPO release and every material Stats SA or national statistics office update. Where a client identifies a country figure that materially disagrees with in-country ground truth, we investigate and update the calibration in the next release rather than defending the model against the country expert.