This chapter is from the book
This chapter is from the book
This chapter is from the book
11.5 Actionable Insights
In this chapter, here are the actionable insights:
Design is key for causal inference. Validation comes from design, not from results.
Regression discontinuity is a design that can be used to operationalize many different types of randomness that occur naturally in web products; it’s a creative design technique, rather just relying on a break or level change in the treatment variable.
Understanding time-series data is key to modeling some of the best RD/natural experiments where treatment is a function of time.
Seasonality decomposition is an extremely useful technique when working with time-series data.
Causal inference from observational data is vital to developing inference, since many difficult problems do not easily lend themselves to A/B testing. With a little thought, many designs in practice can be used to find causal factors.
Chapter 10 and 11 have demonstrated some of the most useful natural experimental and quasi-experimental designs for causal inference in observational data. The next chapter will cover statistical matching, another quasi-experimental design that’s more applicable to all cases, and Hill’s causality conditions, which we can use when all quasi-experimental designs fail.