Determinants of Smallholder Rubber Yield in Southern Nigeria: Empirical Evidence from OLS, Ridge, Lasso, PCR and PLS Models

Abstract

Natural rubber production remains important to smallholder agriculture in Southern Nigeria, but yields are constrained by suboptimal input use and correlated production factors. This study examined determinants of rubber yield using farm-level data from 200 farms across three Nigerian states for 2021–2025, obtained from the Rubber Research Institute of Nigeria. A log-log Cobb–Douglas framework was estimated using Ordinary Least Squares (OLS), Ridge, Lasso, Principal Component Regression (PCR), and Partial Least Squares (PLS) to assess the effects of farm size, tree density, fertiliser application, labour input, and farm age while addressing multicollinearity. Descriptive statistics, correlation coefficients, and Variance Inflation Factors supported the presence of multicollinearity among key production inputs. Across the five models, labour, farm age, tree density, and farm size had positive coefficient estimates. The fertiliser coefficient was negative under OLS, Ridge, and Lasso but positive under PCR and PLS, indicating sensitivity of its estimated effect to the modelling approach. Lasso produced the lowest mean squared error (MSE = 36,740.12) with R² = 0.4005, while Ridge, PCR, and PLS also reduced prediction error relative to OLS (MSE = 44,964.52). Overall, regularised and component-based methods provided better predictive performance than OLS in the presence of correlated inputs. The findings indicate that labour allocation, planting density, farm size, and plantation age are important considerations for smallholder rubber productivity in Southern Nigeria.