I exploit Côte d’Ivoire’s 2011 post-presidential-election crisis as a natural experiment to document the effects of closing rural rail stations. Using nationally representative household microdata, I find sizable declines in per-capita spending and consumption in rural southern areas where stations were closed relative to adjacent areas that were never served by rail. These declines coincide with a reallocation of workers from higher-paying nonfarm sectors into lower-productivity family work. I use a Roy model with unobservable heterogeneous returns and sectoral mobility costs to rationalize these findings and recover marginal nonfarm returns. I show that the welfare gains from reopening rural rail stations exceed operating costs, though net gains diminish as coverage expands. Spatially targeted reopenings of a subset of the closed stations generate nearly the same welfare gains as full network restoration, implying a benefit–cost ratio almost twice as high.

We propose a new specification test to assess the validity of the judge leniency design. The test exploits all information in the data distribution to detect violations of the core assumptions and is asymptotically sharp and consistent. When validity is rejected, we propose a partial-monotonicity and exclusion approach that allows recovery of the Marginal Treatment Effect (MTE). Monte Carlo simulations show that our sharp test outperforms existing non-sharp tests. We apply it to U.S. criminal data (Stevenson, 2018), finding violations in three crime categories: robbery, drug selling, and drug possession.