Estimation in presence of local unobservables


The paper proposes a class of estimators based on spatial data transformation that allows practitioners to perform inference when models with local unobservable heterogeneity are implemented. In particular, we propose two estimators: the Spatial Differencing (SD) and the Spatial Within (SW) estimators as in the panel data setting. Both estimators require a spatial data transformation based on a given threshold distance. Gaining insight from the recent dyadic data regression literature, we study the asymptotic properties of the estimators and their finite sample performance using Monte Carlo simulations. We propose a Hausman-like test to detect the presence of local unobserved heterogeneity. The method is illustrated using data on the Kenyan project in Miguel and Kremer (2004). Consistently with Miguel and Kremer (2004) results, we find that the medical treatment spills over a radius of approximately 3 Km.

Work in progress (first draft available soon)