Extended Data Fig. 6: Map of observed versus predicted regression results and residuals for DHS regions.

a, Observed levels of the PC1 development composite from available data. b, Model predictions based on linear regression of k with controls and country fixed effects, Extended Data Table 2. c, Difference between observed and predicted development composite (residuals). Positive residuals – where the model underpredicts development – such as in southern Angola/northern Namibia and northern Mauritania, are typically associated with regional effects not well captured by country fixed effects, such as near national borders. These regions are all rural and very sparsely populated, Extended Data Fig. 8c–e, with very high k, subject to possible infrastructure underestimation but, more likely, to limitations of the linear model at extreme k values, Extended Data Fig. 8a–b. Detailed residuals analysis using individual DHS measures show that these areas (Cunene, Angola; Oshikoto, Namibia; Ohangwena, Namibia; Omusati, Namibia; Assaba, Mauritania) have higher than expected levels of women who are literate, women with a secondary or higher education, women who are currently married using family planning, and with access to health facilities for labor and delivery. Regions showing lower development than predicted are primarily concentrated in the Western Sahel. Such differences have also recently been linked to armed conflict, which depresses human development despite infrastructure access. Panels a–c, subnational region boundaries made with DHS (using rdhs); national boundaries made with Natural Earth at naturalearthdata.com (using rnaturalearth).