Causal Inference for Artificial Intelligence Fertilizer Recommendation Under Endogenous Soil Test Adoption: A Computer Policy Evaluation

Artificial intelligence (AI) fertilizer recommendation systems typically treat soil test adoption as an exogenous input, yet adoption is a farmer choice shaped by unobserved land quality, liquidity and extension access. The analysis specifies a structural recommendation model in which a latent profitability signal drives both soil testing and nitrogen application, so that naively estimated returns to AI-guided dosing conflate the effect of information with the effect of the unobserved signal. Identification is recovered through a control-function formulation that separates the recommendation rule from the adoption decision, and the resulting estimator is shown to be consistent under a conditional independence condition that is weaker than random adoption. The empirical component estimates the reduced-form relationship between electrification, urbanisation and GDP per capita (constant 2015 US$) using ordinary least squares with heteroskedasticity-robust (HC1) standard errors; the model achieves R-squared = 0.591 (adjusted R-squared = 0.5895, F = 184.764, p = 2.82e-62), with an electricity access coefficient of 1291.55405 (SE = 194.35159) and an urbanisation coefficient of 113.13952 (SE = 7.77493). These estimates are used to calibrate the scale of the confounding bias that an endogenous-adoption correction must remove.