hide
Free keywords:
-
Abstract:
Climate‑smart agriculture (CSA) is promoted as a key strategy for enhancing food security and resilience in Kenya, yet adoption remains uneven across gender and agroecological strata. Conventional adoption studies struggle to diagnose the root causes of these disparities. We introduce a Bayesian‑Ising framework that unifies the analysis of simultaneous practice interactions and sequential adoption pathways under an intersectional lens, using survey data from 569 smallholders in Western Kenya. Three key findings emerge. First, complementarities are stratified, not universal. The pooled Ising network shows negative interactions between soil management (CSA1) and resilient crops (CSA3: Ĵ = −0.385, 95% CI −0.577 to −0.199) and between CSA1 and agroforestry (CSA4: Ĵ = −0.324, −0.486 to −0.141). Strong positive complementarities are concentrated in Male‑Headed Households in the Upper-Midland Zone (MHH‑UMZ: CSA4-CSA5 Ĵ = 0.686, 0.250–1.151; CSA5-CSA6 Ĵ = 0.524, 0.031–1.090). In Female‑Headed Households in the Lower-Midland Zones (FHH‑LMZ), the strongest edge is a negative interaction between water harvesting (CSA2) and energy efficiency (CSA5: Ĵ = −0.727, −1.386 to −0.203). Second, sequential analysis reveals distinct directed pathways. For FHH‑LMZ, the only directed edge is CSA2 → CSA5, yet these practices are antagonistic. For MHH‑LMZ, directed edges are CSA1 → CSA3 and CSA1 → CSA6, also with negative interactions. For MHH‑UMZ, a self‑reinforcing chain CSA3 → CSA4 → CSA5 → CSA6 aligns with positive complementarities. Cumulative probability of reaching CSA4 through the modeled sequence is 0.039 (0.024–0.059) for FHH‑LMZ, 0.035 (0.026–0.045) for MHH‑LMZ, and 0.056 (0.032–0.091) for MHH‑UMZ − a disparity of about 1.6‑fold between the most and least advantaged strata. The Intersectional Adoption Equity Index (IAEI) shows FHH‑LMZ achieves 0.747 (0.349–1.384) of the reference group’s cumulative adoption, while MHH‑LMZ achieves 0.666 (0.361–1.154). Third, diagnostic counterfactual simulations applying MHH‑UMZ transition probabilities yield modest and uncertain gains: FHH‑LMZ cumulative CSA4 increases by 0.023 (−0.010 to 0.062), MHH‑LMZ by 0.024 (−0.004 to 0.060); these gains are not statistically distinguishable from zero. We conclude that CSA scaling requires moving beyond one‑size‑fits‑all bundling. Policy should, ex ante, first diagnose context‑specific interaction patterns − positive synergies only in MHH‑UMZ, antagonisms in FHH‑LMZ and MHH‑LMZ − address the structural barriers underlying negative interactions, and support the directed pathways identified for each stratum. The Bayesian‑Ising framework offers a portable diagnostic toolkit for such context‑specific analysis, with emphasis on adaptive learning given the substantial uncertainty surrounding equity metrics.
