There's an old saying that goes, "figures don't lie, but liars can figure." But sometimes even the figures can spin a confusing story. That's why I've always appreciated the power of understanding statistics. I even remember encouraging my sons to take a statistics class over a calculus course because, unlike calculus, I've used statistics almost every day, but I haven't used calculus since I happily departed differential equations class in college.
Many organizations fall in a common fundamental error when it comes to analyzing and attributing their own successes and failures. They confuse correlation with causation.
Correlation Versus Causation
Let's start by defining our terms. Correlation is when two items are linked in some way statistically. In other words, we can use stats to show that these two things are linked together with a high rate of probability.
Let's say that when I eat too much pizza, I feel pain in my stomach. The action of eating a lot of pizza is highly correlated with me feeling sick.
While these two actions are correlated, they don't necessarily mean that the first action caused the second. What if, for example, it turned out that I had Celiac disease--which is an allergy to wheat. In this case, that means that it's because the pizza crust was made with wheat which caused my illness--not that I ate too much.
The Correlation Trap
A more classic example of how it's easy to confuse correlation with causation is that we can statistically prove that 97% of people who got into a car accident drank at least one glass of water 24 hours before the accident. There is clear mathematical proof that drinking a glass of water is highly correlated with car accidents. But we all know there is no causation to this ridiculous relationship.
While that example is clear, the identical mistake is made thousands of times a day in businesses. Statistical analysis is performed between a factor and an outcome, and a high degree of correlation is found. This is a case of confusing correlation with causation. This comes out when the experiment is scaled based on those statistics and the outcome isn't generated. Something else was causing the outcome, not the non-causal, but correlated factor.
Thanks to the famous philosopher and writer John Stuart Mill, who studied the relationship between correlation and causation back in his , A System of Logic Ratiocinative and Inductive (1843), there are three questions you can ask to determine causality.
- Do X and Y events occur together? Are they always happening together at the same time?
- X precedes Y. Does Y always occur after X has occurred?
- There are no other reasonable explanations for the observed XY relationship.
If any of these three questions can't be answered in the affirmative, then there is not causation although there may be correlation.
Correlation in Action
A great example of how your organization might be confusing correlation with causation is in the world of digital marketing. You might have found that you saw a huge bump in sales recently after you launched a new Google ad or a new content marketing initiative. There is clearly a correlation between those two events. But is there causation?
When you peel back that onion, you might find that there were a host of environmental factors that came into play--like post-Covid consumer demand, supply chain issues, etc.--that were the true drivers for demand for your products. Unfortunately, none of these are controllable by the marketing team, who will be keen to increase budget based on the statistical correlation they found between their actions and the increased revenue.
I worked with a company that sold home security equipment and it was truly savvy with its advertising. And every time the marketing department put out a new ad, they said it drove sales. But over time, we saw that it was something else entirely--like a news story about a riot or a shooting--that led to the increase in sales. The ad was only a billboard that told people where to call or click to order a system.
Again, it can be easy to assign credit or blame to one action or event when something else was really the driver. That's why, in that security organization, our mantra became: correlation is not causation!
So, the next time someone presents a statistical case to you, ask yourself Mill's three questions to help drill down and see if there was truly causation at work.