Title: Modelling of Rice Production in Indonesia Using a Semi-Parametric Regression Approach on Panel Data
Authors: Suliyanto, Dita Amelia, Gabriella Agnes Budijono, Khansa Azizah, Raja Van Den Bosch Sihotang
Volume: 9
Issue: 6
Pages: 173-186
Publication Date: 2025/06/28
Abstract:
This study models rice production dynamics across Indonesia's 34 provinces (2022-2024) using a semi-parametric panel data approach with Generalized Additive Models (GAM). Rice (Oryza sativa L.) is Indonesia's primary staple crop, critical for national food security, yet production declined by 2.05% between 2022 (54.75 million tons GKG) and 2023 (53.63 million tons GKG). To address complex, potentially non-linear relationships, longitudinal data from the Central Bureau of Statistics (BPS) modeled provincial rice production (Y, tons) against key predictors: productivity (X?, quintal/ha), percentage of informal agricultural labor (X?, %), average temperature (X?, °C), average humidity (X?, %), and rainfall (X?, mm). Descriptive analysis revealed significant regional disparities: average production was 1,582,867 tons (SD = 2,609,910 tons), with Java provinces dominating nationally. A 2024 decline occurred in most provinces. Productivity and informal labor percentages decreased in 2024, while humidity peaked (87.76%) and rainfall showed high volatility (max 4950.5 mm in 2024). Correlation analysis indicated positive links between production and productivity (r = 0.55) and labor (r = 0.19), but negative associations with temperature (r = -0.21) and humidity (r = -0.32). The optimal GAM, selected via the lowest Quasi Information Criterion (QIC), utilized an Independence working correlation structure. Results identified significant non-linear effects of productivity, informal labor, and temperature on production. Humidity exerted a significant negative linear effect (coefficient: -196,600), while rainfall had a positive linear effect (coefficient: +397.4). This underscores the complex interplay of socio-economic and climatic factors, particularly the constraining role of high humidity and the beneficial, yet variable, role of rainfall. The findings support targeted, climate-resilient agricultural policies. Future research should incorporate variables like harvested area and irrigation and explore Generalized Additive Models (GAM) for extended timeframes.