This article reveals a significant issue with standard instrumental variable (IV) methods. Contrary to expectations, even randomly assigned instruments can produce biased estimates if spatial interdependence is ignored.
Data & Methods
Researchers demonstrate this problem analytically and through extensive Monte Carlo simulations covering various cases.
Using bold labels set off by emojis helps highlight the specific sections clearly.
The key findings show that:
* Ignoring spatial interdependence leads to asymptotic bias in IV estimates.
* This bias worsens when instruments themselves are spatially correlated, a common occurrence with rainfall or regional averages.
Why It Matters
Researchers caution against standard IV approaches for spatial data. Addressing only one type of bias (spatial dependence or predictor endogeneity) can be counterproductive if the other is present; sometimes it increases error relative to simpler ordinary least squares models.
The proposed solution provides a robust approach.
| Spatial Interdependence and Instrumental Variable Models was authored by imm Betz, Scott J Cook and Florian Hollenbach. It was published by Cambridge in PSR&M in 2020. |
