Incentives for REDD − i.e., reductions in emissions from deforestation and degradation − motivate application of static economic modeling of land use to assess heterogeneity over space in the business-as-usual baselines for land use required for forest policy evaluations. That some forested locations face higher threats is now recognized as an important factor in the evaluation and targeting of policy. Given this point − now often included in impact evaluation via matching − further theory is required to explain variations in policy impact. We show this need by analyzing impacts of Mexican protected areas (PAs) on land cover. Applying static land-use economics improves the baselines for our impact estimation and we find, on average, a 2.5% lower rate of 2000-05 natural land cover loss within the PAs. Stricter PAs appear closer to cities and have greater impact (4.4%) than less strict (2.3%), yet static baselines do not explain why. Nor do they explain why impact gradients by type differ across countries, or why PA spillovers vary across states − as we show for Mexico. We suggest an initial political economy model of impacts by type of PA and also provide examples of the economic and political dynamics required to understand PAs' spillovers.