Estimation of Dose-Response Functions and Optimal Doses with a Continuous Treatment

This paper considers the continuous-treatment case and develops nonparametric estimators for the average dose-response function, the treatment level at which this function is maximized (location of the maximum), and the maximum value achieved by this function (size of the maximum). These parameters are identified by assuming that selection into different levels of the treatment is based on observed characteristics. The proposed nonparametric estimators of the location and size of the optimal dose are shown to be jointly asymptotically normal and uncorrelated. More generally, these estimators can be used to estimate the location and size of the maximum of a partial mean (Newey, 1994). To illustrate the utility of our approach, the techniques developed in the paper are used to estimate the turning point of the environmental Kuznets curve (EKC) for NOx, that is, the level of per capita income at which the emissions of NOx reach their peak and start decreasing. Finally, a Monte Carlo exercise is performed partly based on the data used in the empirical application. The results show that the nonparametric estimators of the location and size of the optimal dose developed in this paper work well in practice (especially when compared to a parametric model), in some cases even for relatively small sample sizes.