🔎 What This Paper Questions

Spatial instruments—constructed from other units' realizations of an endogenous variable (for example, regional or global weighted averages)—are widely used in political science because they are easy to obtain, often statistically powerful, and can appear plausibly exogenous. However, the assumptions that would justify their use are poorly understood and deserve scrutiny.

🔍 Two Fundamental Problems Identified

• Spillovers and cross-unit interdependence: When the endogenous predictor exhibits cross-unit dependence, other cross-unit relationships (such as spillovers or interdependence) are likely present as well. Those relationships create channels through which the spatial instrument can be correlated with the outcome aside from the effect of the focal endogenous variable, risking violations of the exclusion restriction.

• Simultaneity in the first-stage: Spatial instruments typically include other units' outcomes as right-hand-side predictors while those outcomes are themselves simultaneously determined with the focal endogenous variable. Because the instrument and the endogenous regressor are jointly determined, the exclusion restriction is necessarily violated by construction.

📌 What Follows From These Concerns

• The two problems together—spatially induced spillovers and automatic simultaneity—undermine the core conditions required for a valid instrumental variable.
• These concerns challenge the credibility of spatial instruments across many common empirical settings.

🧾 Bottom Line

Given the risks of exclusion restriction violations from both cross-unit dependence and first-stage simultaneity, spatial instruments are rarely, if ever, valid for identifying causal effects.