What is the effect of salary on employee loyalty? This is the question that I, along with three teammates tackled as a part of a course on Causal Inference that I took during my Masters degree. Techniques we used included:
- Directed acyclic graphs (DAGs), structural causal models and the counterfactual outcomes framework to define the causal problem.
- Super Learning: Linear, Prototype, Tree-based, and Neural Network models to estimate the propensity to treatment and conditional expected outcome.
- Simple Substitution Estimator, Inverse Probability of Treatment Estimator and Targeted Maximum Liklihood Estimator for estimating the Average Treatment Effect (ATE).
My main contribution was in the Super Learning component, as well as the estimation of the ATE.