Extension Contact and Rice Productivity of Non-Swamp Farmer Groups in Mamasa Regency, West Sulawesi

Abstract

Rice production in West Sulawesi fell by 16.83% in 2023, driven mainly by a 14.71% contraction in harvested area, while implied provincial productivity declined only slightly (from about 5.10 to 4.97 t ha⁻¹). Extension services are expected to raise productivity, yet their association with yields in heterogeneous highland rice systems is poorly documented. This study examined the cross-sectional association between extension contact and rice productivity among non-swamp farmer groups in Mamasa Regency, defined as lowland rice (sawah) groups outside tidal and inland swamps, cultivating rainfed, semi-technically irrigated, or technically irrigated fields. Field-survey data from 179 farmer groups in all 17 sub-districts were analyzed. Extension exposure was measured by three categorical items (meeting regularity, frequency of contact with extension workers, and mentoring activeness) and productivity as the chairperson-reported mean yield across cultivated seasons. Because a reflective latent specification showed inadequate reliability (ρ_c=0.629; AVE=0.411) and a redundant outcome indicator, the primary analysis used regression on the observed indicators with robust and bootstrap confidence intervals, with PLS-SEM as a secondary check. Routine contact with extension workers (88.3% of groups) was associated with 1.59 t ha⁻¹ lower productivity than rare or occasional contact (95% CI −2.65 to −0.54; p=0.003), whereas meeting regularity and mentoring activeness showed no association. The association was stable across alternative outcome definitions, influential-case exclusions, sub-district fixed effects, and PLS-SEM specifications but attenuated and became non-significant after exploratory adjustment for agroecological, technology-access, and group covariates (−0.88; 95% CI −2.05 to 0.29). Out-of-sample predictive accuracy was low (Q²_predict=0.070 against a naive mean benchmark). The negative association should not be read as a harmful effect of extension; need-based targeting, poorly matched extension content, missing complementary inputs, and unmeasured agroecological differences remain competing hypotheses. Documenting targeting criteria, collecting baseline productivity and agroecological data, and evaluating context-specific extension packages over time are recommended.