Democratizing Propensity Score Matching Using Web Application

Abstract

Traditionally, data scientists use exploratory data analysis techniques such as correlation analysis, summary statistics, and regression analysis for identifying the most product enhancements and roadmap planning. However, these conventional approaches often yield biased conclusions and suboptimal solutions, leading to a waste of valuable time and missed opportunities for higher-value outcomes. In contrast, there are alternative techniques that involve the use of causal inference methods. However, these methods suffer from issues of limited accessibility, as they are not easily understandable or effectively utilized by inexperienced practitioners. Additionally, their implementation necessitates a substantial investment of time and effort. To this end, this paper tackles these challenges by democratizing one of the causal inference methods called Propensity Score Matching (PSM) and enhancing its accessibility for less technically inclined users through the automation of the entire workflow using a web application. Our approach not only fills this accessibility gap but also contributes to the existing literature by introducing a more rigorous model selection process and an enhanced sensitivity analysis. By overcoming the limitations of traditional exploratory data analysis methods, our web application has empowered data scientists at Booking.com to make better use of PSM, thereby improving the overall efficacy of their analyses.