Modelling Factors Affecting the Middle Income Trap in Indonesia Using Generalized Additive Models (GAM)

Abstract

Indonesia is currently facing the risk of the Middle Income Trap (MIT), a condition in which economic growth stagnates after reaching middle-income status. This study aims to identify and model socio--economic factors affecting MIT at the provincial level in Indonesia during 2020--2023. The Generalized Additive Model (GAM) is employed to capture nonlinear and heterogeneous relationships between predictors and GRDP per capita with complex patterns that conventional linear or parametric models often fail to detect. The use of GAM in this context represents a methodological contribution, as studies applying GAM for MIT analysis in Indonesia remain very limited. This research therefore introduces a novel analytical approach by demonstrating how GAM can reveal flexible functional relationships and uncover nonlinear effects that are overlooked by traditional panel regression. GRDP per capita is modeled using six predictors: life expectancy, poverty rate, informal employment share, upper secondary education completion, food insecurity prevalence, and population density. The best model is obtained using the Gaussian family with an identity link, with five predictors showing nonlinear effects and food insecurity exhibiting a negative linear influence. The selected model demonstrates strong performance, indicated by an AIC value of 2743.279 and a R^2 of 98.6%, suggesting a very high explanatory power. In addition, the model achieves good predictive accuracy, with a MAPE of 8.04%. The findings support evidence-based policies aligned with Sustainable Development Goal (SDG) 8, promoting inclusive and sustainable economic growth.