Scale and space in industrial policy implementation: Geospatial evidence from urban Ethiopia

Industrial policy in Africa has regained prominence, yet many registered manufacturing investments do not progress to production. This study examines the spatial patterning and correlates of implementation using a field-verified geospatial census of 2277 manufacturing projects in Shaggar City, Ethiopia. Projects were classified into five ordered implementation categories: Not Implemented, Abandoned, Under Construction, Construction Completed, and Operational. Ordered logistic regression was used to examine associations with network distance from Addis Ababa's central business district (CBD), firm size, industrial subsector, and distance to former town centers. In the primary specification, greater distance from the CBD was negatively associated in the primary ordinal specification with lower cumulative odds of being in a higher implementation category (OR = 0.921, 95% CI: 0.901–0.942, p < 0.001). The distance association was nonlinear in supplementary specifications, although its precise form varied across model formulations. Residual diagnostics, however, indicated substantial spatial autocorrelation across all four cumulative thresholds (Moran's I, p < 0.001), limiting the extent to which conventional model-based uncertainty can be interpreted as spatially independent inference. Distance to former town centers showed evidence of stage-specific association, concentrated at earlier implementation transitions. Firm size was also associated with implementation status in the primary ordinal model, with large projects having lower cumulative odds than small projects; this association weakened after adjustment for project timing and was directionally reversed in a binary multilevel sensitivity model. The latter model accounted for sub-city clustering but produced a non-positive-definite Hessian and is therefore treated as indicative rather than confirmatory. Missing firm-size information was spatially patterned, with missing records more common near the metropolitan core and among projects with higher operational rates. These findings show that manufacturing implementation is a spatially structured, multi-stage process and that observed associations depend on how implementation stage, project timing, missing information, and spatial dependence are handled. The study highlights the value of field-verified project-level data for distinguishing registered investment from actual implementation and for identifying where different stages of the implementation process may require closer monitoring and support.