Causal Effect

A causal effect is the difference between an outcome after an action and the counterfactual—what would have happened without that action. For instance, the causal effect of taking aspirin is the pain relief you feel compared to the headache you would still have. Because we can't observe the counterfactual, we must find clever ways to estimate it. Try this: When you see a change, ask what would have plausibly happened if no action was taken.

The core of understanding a causal effect lies in the concept of the counterfactual. Since we can't observe the same situation with and without an intervention at the same time, we must find ways to approximate this 'what if' scenario.

This is why correlation is not causation. Observing that traffic decreased after bike lanes were installed doesn't prove the lanes were the cause. Perhaps the city also launched a new bus line, or a major employer switched to a remote work policy. These are confounding variables that offer alternative explanations for the outcome.

To isolate a causal effect, researchers use methods to create a credible comparison group. The gold standard is a Randomized Controlled Trial (RCT). For example, a company uses an A/B test to measure the causal effect of a new website button. By randomly showing half its users the new button (treatment group) and half the old button (control group), it can isolate the button's impact on clicks, since all other factors are, on average, equal between the groups.

In daily life, we can apply this thinking by looking for natural experiments or comparison groups. Before concluding that your new marketing campaign caused a sales spike, check if your main competitor simultaneously ran out of stock. The competitor's issue provides a strong alternative explanation for your success, making the causal effect of your campaign uncertain.

Modern causal inference is built on the Potential Outcomes Framework, which defines a causal effect as the difference between an outcome with a treatment and its potential outcome without it (the counterfactual). The core challenge is that the counterfactual is unobservable, making it a missing data problem. Major contributions to this field include:

• Rubin (1974): Formalized the potential outcomes model, defining the individual causal effect and framing the central challenge as one of missing data. [1]
• Pearl (2009): Developed a complementary graphical framework using Directed Acyclic Graphs (DAGs) to make causal assumptions explicit and identify confounders. [2]
• Angrist & Pischke (2009): Advanced the 'credibility revolution' by promoting robust empirical methods like natural experiments to credibly estimate causal effects from real-world data. [3]

The primary limitation is the fundamental problem of causal inference: for any individual, we can never observe both their outcome with treatment and their outcome without treatment at the same time. All statistical methods, including randomized trials, are attempts to estimate the average causal effect across a population, not the precise effect on a single individual. Furthermore, causal claims are only as strong as the assumptions they are built on; if you fail to account for a key confounding variable, your estimate of the effect will be wrong.