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

We introduce a tree-based approach for assessing the performance impact of diverse self-selected interventions in management research. Our approach, which takes advantage of "Big Data", or observational data with large sample sizes and a large number of variables, offers important advantages over traditional propensity score matching. In particular, the tree-based approach to assessing the impact of interventions offers a data-driven methodology that applies to a wide range of intervention types (binary, polytomous, continuous), allows for examination of nascent interventions whose selection cannot be theoretically specified a priori, identifies pre-intervention variables that correlate with the self-selected intervention, and presents comparisons of ensuing performance in visuals that are easy to discern and understand. We illustrate the method and the insights that it yields in the context of two studies: analysis of the impact of an eGov service in India, and comparison of performance across different contractual pricing mechanisms and contract durations in the outsourcing of technology and technology-enabled business functions