Evaluation of Difference-in-Differences Estimators for Artificial Intelligence Agriculture Extension Targeting Under District-Level Partisan Gerrymandering: A Governance Causal-Identification Bound for Smallholder Access Equity

The problem is that the parallel-trends assumption, on which difference-in-differences identification rests, is not merely a statistical condition but a political one: district boundaries drawn to favour partisan incumbents can systematically sort smallholders into treatment and comparison units on unobserved determinants of extension access. A replication-based evaluation of the ordinary least squares estimator was conducted using 600 data sets of N = 500 drawn from a declared data-generating process (intercept 12.0, exposure 0.42, group -3.1, severity 1.35, error SD 4.2). The estimator recovered the exposure coefficient with negligible bias (-0.00034) and nominal 95% confidence interval coverage of 1.0, while the intercept showed bias of 0.0134 and root mean squared error of 0.87403. The bound, not the point estimate, is the appropriate object of governance inference when district boundaries are politically endogenous, and it specifies the conditions under which difference-in-differences estimates of extension targeting can be treated as causal.