Abstract:
|
Recent developments on integrating causal modeling and efficient estimation strategies have demonstrated the potential of a flexible yet rigorous causal inference roadmap by better reflecting real world data knowledge, getting as close as possible to the actual causal question with clearly specified assumptions, and allowing more efficient adjustment. However, a common challenge for the corresponding statistical analysis in practice lies in the complexity of analytic calculation of efficient influence functions. In this manuscript, we focus on alternative numerical representations based on Highly Adaptive Lasso. We show that scalable and intuitive computerized higher order efficient estimation can be achieved via sequential projections of simple initial mappings such as inverse probability weighted functions or initial influence functions. A generalized algorithm without prerequisite projection structures of specific initial mappings is also investigated.
|