Spatial heterogeneity of profitability among indigenous fruit and vegetable farmers in West Africa

Introduction The profitability of indigenous fruit and vegetable (IFV) farmers in West Africa is critical for food security, yet standard analyses often assume that the socioeconomic determinants of profitability operate uniformly across space. Methods This study tests that assumption using a Geographically Weighted Logistic Regression (GWLR) applied to survey data from 1,460 IFV farmers across nine West African countries. A global logistic regression identified gender, education, age, and extension visit as significant predictors of profitability, but significant residual spatial autocorrelation in its errors (Moran's I = 0.449, p < 0.001) indicated that this single pooled estimate masks substantial geographic variation. Results and discussion The GWLR confirmed profound spatial non-stationarity: no predictor was statistically significant at more than 27.3% of farmer locations, and several variables that were non-significant in the global model, including marital status, household size, farmer-based organization (FBO) membership, and IFV training, showed meaningful local effects in specific areas. Education displayed the most geographically extensive positive association with profitability of any predictor, while the direction and strength of the gender and extension-visit effects varied considerably across locations, including reversals of the global relationship in parts of the study area. The GWLR also substantially outperformed the global model in predictive validation (in-sample AUC = 0.914 vs. a 5-fold cross-validated AUC of 0.582 for the global model); because the GWLR figure is in-sample while the global figure is cross-validated, this comparison likely overstates the GWLR's true out-of-sample advantage, and should be read as suggestive rather than a like-for-like performance gain. These findings indicate that agricultural interventions for West Africa's IFV sector should be spatially targeted at a sub-national level rather than applied uniformly, and that decision-support tools should incorporate local socioeconomic and institutional context rather than relying on national or regional averages.