Fragility of the Marginal Treatment Effect

Many interesting and important economic questions relate to the effects of binary treatments such as starting a college degree or participating in a job training program. The causal effects of these treatments are likely to be heterogeneous and recent research has emphasized the estimation of heterogeneous treatment effects, with a particular focus on Marginal Treatment Effects (MTEs). In this note, I describe why common methods of estimating MTEs of binary treatments can be very sensitive to omitted higher powers of covariates and demonstrate this using simple Monte Carlo simulations. I conclude by discussing approaches that may be useful for researchers to address this problem in practice.