Spatially varying causal effect models
share
We establish causal effect models that allow for time- and spatially varying causal effects. Under the standard sequential randomization assumption, we show that the local causal parameter can be identified based on a class of estimating equations. To borrow information from nearby locations, we adopt the local estimating equation approach via local polynomials and geographical kernel weighting. Asymptotic theory is derived and a wild bootstrap inference procedure is given. The proposed estimator enjoys an appealing double robustness feature in the sense that, with a correct treatment effect model, the estimator is consistent if either the propensity score model or a nuisance outcome mean model is correctly specified. Moreover, our analytical framework is flexible enough to handle informative missingness by inverse probability weighting of estimating functions.
READ FULL TEXT
