abstract: 

We consider evaluating the causal effects of dynamic treatments,i.e. of multiple treatment sequences in various periods, based on double machinelearning to control for observed, time-varying covariates in a data-driven wayunder a selection-on-observables assumption. To this end, we make use ofso-called Neyman-orthogonal score functions, which imply the robustness oftreatment effect estimation to moderate misspecifications of the dynamicoutcome and treatment models. This robustness property permits approximatingoutcome and treatment models by double machine learning even under highdimensional covariates and is combined with data splitting to preventoverfitting. In addition to effect estimation for the total population, weconsider weighted estimation that permits assessing dynamic treatment effectsin specific subgroups, e.g. among those treated in the first treatment period.We demonstrate that the estimators are asymptotically normal and root-nconsistent under specific regularity conditions and investigate their finitesample properties in a simulation study. Finally, we apply the methods to theJob Corps study in order to assess different sequences of training programsunder a large set of covariates.

to participate to the event clic here: JoinMicrosoft Teams Meeting