A Tree-Based Approach for Addressing Self-Selection in Impact Studies with Big Data