Checking open geospatial inputs against official statistics: census-anchored population and facility counts for modelled walking access in Uasin Gishu County, Kenya

Abstract

Background. County-level accessibility analyses in Kenya usually rely on global gridded population and archived facility inventories, but they rarely check these inputs against the official statistics that county planners use. We compared open inputs with census, electoral and county health statistics for Uasin Gishu County, and tested whether anchoring population to official counts changes modelled walking access and ward priorities. Methods. Ward population totals from four gridded products (WorldPop 2016 at 100 m, GHS-POP 2015 and 2020, WorldPop 2026 constrained at 1 km) were compared with 2009 census ward populations, 2017 and 2022 registered voters by ward, and 2019 census sub-county totals. Because KNBS administrative sub-counties differ from constituency ward groupings, we built a ratio crosswalk from 2009 data. We then re-weighted WorldPop 2016 cells so that ward totals matched official shares in three steps. Using these weights, we recomputed the share of population beyond 60 minutes' walk from government dispensaries and health centres, and the six highest-scoring wards. Finally, we compared an archived facility inventory with County Department of Health counts for 2020–2024. Results. WorldPop 2016 matched 2009 census ward shares closely (Spearman ρ = 0.85; mean absolute error 0.21 percentage points), except for neighbouring Eldoret wards Langas (−2.6 points) and Kipkenyo. Agreement with 2022 registered voters was weaker for all products (ρ = 0.66–0.77). The 1-km WorldPop 2026 product agreed least at ward level (ρ = 0.69 against 2009 census). KNBS sub-county totals differed from constituency ward groupings by 18,619 people (Soy/Moiben) and 9,192 (Kapseret/Kesses) in 2009. Census anchoring changed the county share beyond 60 minutes by −0.04 to +0.44 percentage points (baseline 7.17%) and left the six highest-scoring wards unchanged in all twelve anchored analyses. The archive contained 110 located government dispensaries and health centres, compared with 133 (2020) and 143 (2024, provisional) in county statistics. In Kapseret it held 12 against 29. Conclusions. In this county, between-ward population totals were a minor source of uncertainty for modelled access and priorities compared with friction-surface values, provider eligibility and inventory completeness. Analysts should reconcile administrative units before anchoring to census sub-counties, and should check facility counts against county records before reporting access gaps. Preprint, not peer reviewed. Companion papers: measurement sensitivity, decision analysis, and external check of inputs (Uasin Gishu County, Kenya). Code and derived data: S1 zip.