A Study of Counterfactual Inference Based on Instrumental Variables and Machine Learning

Abstract

Causal inference is based on the inference of cause to effect and is part of causal analysis. It is an important method in data analysis and data science, and its application is very widespread in many fields. Counterfactual inference is a very important part of causal inference, which is the activity of thinking in which facts that have occurred in the past are negated and re-represented in order to construct a hypothesis of possibility and this paper specifically investigates two ideas for solving the counterfactual inference problem, the first approach is to use instrumental variables and the second approach is to use machine learning. In addition, based on the previous work which introduced Balancing Neural Network (BNN), this paper illustrates two ideas of modifying the architecture of BNN using shortcut connection. The assessment of performance of these two ideas will be done in future works.