Beyond farm size: spatial determinants of cocoa productivity in Ashanti Region, Ghana

Cocoa yields in Ghana have declined 23% since 2020 despite favorable prices, yet the spatial dimensions of this productivity crisis remain under-researched. This study analyzed 2,612 georeferenced cocoa farms across 10 districts in Ghana’s Ashanti Region (5°30’N–7°45’N, 0°15’W–2°25’W) to identify spatial patterns and determinants of yield variation. We integrated GIS-based spatial analysis with multivariate spatial regression modeling, applying Moran’s I statistics, and Ripley’s K-function. Results revealed strong spatial clustering of productivity (Moran’s I = 0.594, p < 0.001), indicating that unobserved spatially structured factors significantly shape yield outcomes, with seven distinct clusters identified through point pattern analysis. A two-hurdle spatial regression approach was employed to analyze cocoa yield determinants, while addressing the substantial zero inflation in production data. More than 50% of sampled farms (n = 2,612) reported zero yields during the 2022/2023 season, reflecting distinct economic and agronomic processes governing participation versus productivity decisions. In Hurdle 1, a spatial autoregressive probit model was applied to estimate the binary participation decision, while Hurdle 2 employed spatial error regression among producing farms to identify conditional yield determinants. The results reveal strong spatial dependence at both decision stages, with spatial autoregressive coefficient ρ = 0.698 in the participation model and spatial error coefficient λ = 0.690 in the productivity model. Tree density exhibits super-elastic effects (elasticity = 1.29), while farm size shows an inverse productivity relationship (elasticity = -0.0287). The model explains 92.07% of yield variation on the original scale using Duan’s smearing estimator for backtransformation. Policy implications emphasize geographically targeted interventions, smallholder intensification through replanting programs, and sustained extension engagement.