Causal Inference: The Case for Causality

Data scientists often focus on whether they can predict something, without pausing to consider whether they should. Modelers and analysts are often met with the cliché phrase: “correlation doesn’t equal causation.” But what even is causation in the first place?  How is it related to correlation?  How do we measure it? And how do we use it to enhance business decisions? In this blog, we’ll give you a high-level overview of causality and answer these often-unanswered questions.

What is causation? How is it related to correlation?

While correlation measures a linear relationship between two variables, it provides little, if any, information regarding if one variable actually causes the other. Causal inference, a field originating out of econometrics, focuses on isolating the sole impact of treatments or interventions on some outcome of interest.  Causal inference not only identifies whether a causal relationship exists but also quantifies the magnitude of that effect.

For instance, suppose you own an ice cream shop. You decide to increase the menu prices by 10% in the summer since ice cream demand is higher in the summer.  How would you determine the impact of the price change on demand?  Causal inference methods would help you determine how much of the shop’s change in demand (or revenue) can be directly attributed to the price change. These methods also help quantify the impact of an intervention while accounting for external factors like seasonality and macroeconomic trends that can simultaneously affect a company’s revenue.

Now that I understand causation, how can I measure it effectively?

While a fully controlled experiment is the gold standard for determining casual effects (e.g. clinical trials), it is often not possible due to constraints such as high cost, impractical logistics, or ethical concerns. For example, say we run multiple ice cream shops, how might we estimate the effect of a price change on our customers’ demand for ice cream? We could offer a different price to each customer, an ideal randomized control trial, but we would have a lot of disgruntled customers who thought our pricing was not fair. Instead, we can borrow a concept from economics called a quasi-experiment. In a quasi-experiment, we might assign one shop to maintain the original price and another to implement the price increase. The challenge here is estimating the effect from this price intervention when each ice cream shop is impacted by their own unique set of demand factors: competition, seasonality, and population density.

In these situations, where controlled experiments are not feasible, we require specialized causal modeling techniques to estimate the true effect of price on demand, given the other factors impacting demand. While there are many different causal inference techniques—which we will explore in more detail in the causal inference blog series—the core idea is that a model is created based on historical data (such as historical demand at the ice cream shop before the intervention) to predict what the demand would have been after the intervention (i.e. price change) if the intervention had never taken place. The difference between the predicted and the actual observed demand gives us an estimate of the causal effect size.

I’d like to know more. How can I use causal analysis to improve business decisions?

While it's important to understand how causal inference works, it is just as critical to know when to apply it. Causal inference is most valuable when addressing cause-and-effect questions and moving beyond simple associations or correlations. If you want to determine the impact of a price change, an intervention, a marketing campaign, or any other decision, causal inference provides you with the tools to determine and measure that effect. ‍

Final Takeaway

Under the right conditions, causal inference transforms observational data into actionable insights. It enables businesses to learn from past behaviors, forecast future outcomes, and make informed decisions.

Want to learn more? If causal inference sounds like something that would benefit you, stay tuned for more blogs in our Causal Inference Series where we will dive deeper into specific modeling techniques and case studies.