Modeling Agricultural Employment Across Africa: Panel Fixed-Effects and Robustness Analysis

Abstract

Agriculture remains a dominant employer across Africa, yet little is known about how within-country changes drive structural shifts in agricultural labor over time. Understanding the national-level factors affecting labour distribution, including youth employment, is essential for developing evidence-based policies that support resilient agrifood systems. In this work, we curated an annual panel dataset covering 54 African countries from 2010–2023 using data from the World Bank Development Indicators, FAO supplementary indicators, and ILOSTAT. Predictors analyzed included education spending, electricity access, urbanization, agricultural labour productivity, renewable energy, fertilizer use, and R&D expenditure. The target variable is agricultural employment as a percentage of total employment. Youth agricultural employment was also analyzed as a secondary outcome because of limited data availability. Country fixed-effects models with standard errors clustered at the country level were used to isolate within-country variation. A Hausman test favored the fixed-effects specification (p = 0.033). In the primary model (N = 477 country-years, across 46 countries; R² (within) = 0.28), greater electricity access (% of population) and higher agricultural value added per worker (constant USD) were associated with a lesser share of agricultural employment (p = 0.003; p = 0.012). Conversely, education expenditure (% GDP) and a limited subsample including R&D expenditure (% GDP) showed no consistent association. Fertilizer was not significant in the linear specification (p = 0.12). These results were robust to alternative specifications, including two-way fixed effects, one-year lagged regressors, a fertilizer extension, log-transformed specifications, and alternative input measures. Log-transformed specifications improved the fit (R2(within) ≈ 0.44-0.46) of the primary model without reversing key associations. A Wald test confirmed a significant heterogeneity between high- and low- development cohorts (p = 0.002). Models evaluating youth agricultural employment (N = 130) yielded weaker fit and less consistent estimates, reflecting data constraints. Overall, the findings suggest that infrastructure development and agricultural productivity are associated with structural labour transitions away from agriculture, although these relationships vary across levels of national development, highlighting the importance of country-specific development pathways for agrifood employment policy.