Comparative analysis on the three popular causality modeling methodologies

Abstract

The idea of causality has lasted for over thousands of years. Unlike the idea of statistical correlation and regression, performing causal modeling and prediction is an even more challenging job. Under the intervention framework of causality, causal modeling is gaining popularity given the advances of big data and computational ability in recent years. In different scientific research areas, there exist three powerful causal modeling methodologies, namely, the potential outcomes method in statistics, the instrumental variables method in economics and Judea Pearl’s causal diagram method (do-calculus) in computer science and artificial intelligence. In this paper, by linear causal modeling assumption, we prove that the above three causal methodologies are equivalent. That is, given a causal problem, all of the three modeling methods will generate the same causal relationship conclusion, despite that they own different causal inference processes. During the past one-and-half years, the global economy suffers severe impacts from the COVID-19 pandemic. To fight the deadly pandemic, various social distancing measures and actions, taken by the countries, are effective in curbing the impact of the pandemic over the population. However, such social distancing policy has an adverse effect over the global economy growth; if more stringent measures were taken, then there would be suffering in the forms of much slower economic growth and higher unemployment. In this paper, we study the causal relationships between social distancing, fatality rate and economy growth. This work provides a useful tool for the governments to keep balance between controlling the pandemic and maintaining economic growth.