Independent Validation of a Modelled African Income Distribution: A Six-Country Study Against the World Bank and National Statistics Offices

Pan Africa Data (PAD) publishes a modelled income distribution across five income classes for 51 African countries at national level and 49 at functional-city and locality level, from 2000 to 2035. This paper documents PAD's national-level calibration approach and reports the results of an independent validation programme across six African countries: South Africa, Kenya, Nigeria, Ghana, Morocco and Egypt. The validation is structured as a three-way triangle: PAD's modelled marginalised class share, the World Bank Poverty and Inequality Platform (WB PIP) headcount at USD 3.65/day 2017 PPP, and each country's own national statistics office (NSO) headcount at its national poverty line. Because PAD calibrates to WB PIP by design - at USD 3.65 for lower-middle-income countries and USD 6.85 for upper-middle-income countries - direct PAD-versus-WB PIP comparisons at those thresholds are calibration fidelity checks, not empirical validation. The empirical content lives in the WB-PIP-versus-NSO edge (which PAD inherits transitively) and in South Africa, the sole upper-middle-income country in the panel, where PAD's marginalised share is genuinely out-of-sample at the USD 3.65 line. At the calibration threshold, PAD agrees with WB PIP to two decimal places for the five lower-middle-income countries. For South Africa at USD 3.65, PAD's 35.75% sits 5.36 percentage points above WB PIP's 30.39% - a real out-of-sample gap, disclosed openly, expected to close with a refined upper-middle-income country calibration in the next model version. The trajectory table (2019–2023) shows PAD tracking the World Bank benchmark through COVID, Nigeria's naira devaluation, Ghana's cedi crisis and Egypt's inflation shock; for the five lower-middle-income countries, every year in which the World Bank has subsequently published a USD 3.65 figure, PAD lands on it. The mechanism combining inputs to the class distribution is not disclosed and remains proprietary. Limitations of the v1 exercise and the planned v2 extensions are stated in full.