Causal Inference in Python

Chapter 5. Propensity Score

In Chapter 4, you learned how to adjust for confounders using linear regression. In addition to that, you were introduced to the concept of debiasing through orthogonalization, which is one of the most useful bias-adjusting techniques available. However, there is another technique that you need to learn—propensity weighting. This technique involves modeling the treatment assignment mechanism and using the model’s prediction to reweight the data, instead of building residuals like in orthogonalization. In this chapter, you will also learn how to combine the principles of Chapter 4 with propensity weighting to achieve what is known as double robustness.

The content of this chapter is better suited for when you have binary or discrete treatments. Still, I’ll show an extension that allows you to use propensity weighting for continuous treatment.

The Impact of Management Training

A common phenomenon in tech companies is for talented individual contributors (ICs) to branch out to a management track. But because management often requires a very different skill set than the ones that made them talented ICs, this transition is often far from easy. It comes at a high personal cost, not only for the new managers, but for those they manage.

Hoping to make this transition less painful, a big multinational company decided to invest in manager training for its new managers. Also, to measure the effectiveness of the training, the company tried to randomly select ...