2 First steps: Working with confounders
This chapter covers
- The effects of confounding: Simpson’s paradox
- Removing confounding bias with the adjustment formula
- Determining when to apply the adjustment formula
Imagine that you have a causal question, such as wanting to assess the consequences of a past decision, or deciding which of two treatments is better, or determining which of two marketing strategies is more effective. You are comparing two options, called A and B. These may be past decisions, medical treatments, or marketing campaigns, and you want to know their effects on a specific result.
You may think that the best approach is to see what happens when A is used, see what happens with B, and compare the two outcomes to decide if A is better than B. However, as we will explore in this chapter, this intuitive method can often lead to wrong conclusions, especially when dealing with observational data. This is a critical point to consider! Think about how often you or others have made decisions this way.
In chapter 1, we discussed that when dealing with observational data (which isn’t collected from a controlled experiment), we often encounter confounding variables. These confounders are factors that influence both the treatment (or action) and the outcome. This means confounders can cause changes in both the treatment and the outcome, and we may not even know they exist or have any details about them.