Most US states do not record a voter's race, so vendors guess it from surname and neighbourhood. The guess is a probability. Then someone rounds it into one word.
Bars show each voter's modelled chance of being Black. The rest of each voter's chance is mostly white.
North Carolina records self-reported race, so the rounding can be checked. Black registered voters, as counted:
Taking the most likely category, the step Dong et al. (2025) attribute to a prominent commercial voter-file vendor, under-counts Black voters by 28.2 percentage points. The bar shows that published shortfall, not a headcount.
So when a vendor quotes a voter's race, ask how it was produced, and whether the prediction was kept continuous or collapsed to a single label.
Source. Dong, E. et al. (2025) 'Addressing discretization-induced bias in demographic prediction', PNAS Nexus, 4(2), pgaf027, doi.org/10.1093/pnasnexus/pgaf027. The ten voters are illustrative, not real records; only the North Carolina figure is published. Six US states, North Carolina among them, collect self-reported race at registration. Captured 10 September 2026.