During a recent business trip to San Francisco, I struck up a conversation with my Uber driver: sports, weather and the standard “how’s business?” He said it fluctuates greatly, a problem he shares with his wife, who owns a company that produces and sells umbrellas. He then shares that he and his wife are typically both either busy or slow at the same times, in large part due to rainy weather causing people to stay dry in a car or under an umbrella. Hearing this, my inner analyst couldn’t resist a follow-up question: “So, can you predict your ride sales based on the revenues of your wife’s umbrella business, and vice versa?”
“No,” he replied, going on to explain that a very sunny day may drive business for his wife from shade-seeking tourists, but may slow down his as people prefer to walk instead of taking a taxi. Conversely, when there’s a major conference in town, he sees a boost in business for trips to and from the airport, but those same patrons don’t need her umbrellas when inside all day.
While it may be human nature to want to draw a connection (when it rains, more people take cabs and buy umbrellas), my recent experience highlights a very common problem among marketers: confusing correlation and causation. Correlation is a statistical measure that describes the size and direction of a relationship between at least two variables, in which the variables are not necessarily related or the change in one variable does not automatically cause the change in the other variables. Causation, on the other hand, indicates that one event is the result of the occurrence of one or more other events. To put it simply: correlation is often coincidence; causality is proven cause and effect.
Brands too often use correlation as a way to try to understand how a particular event -- such as the weather or a new ad campaign -- may lead to a conversion and then use that information to predict revenues. But, as this Khan Academy video shows, correlation does not imply causation, and can lead brands astray when determining what’s driving conversions and revenue, adding to ill-advised decision-making.
For example, a new client noticed that it receives a large number of conversions whenever it rains, and came to the conclusion that the company could look at weather patterns to anticipate conversion. In fact, the real reason for conversions was people actually needing the service it offer (causation). Now this brand is changing its marketing strategies based on causations instead of correlation-based insights.
In today’s advertising industry, nowhere is the confusion over correlation vs. causation more prevalent than with TV data. Several marketing measurement solutions take attributed digital data and overlay aggregated TV GRP data on top of it. In theory, this could provide valuable insight. In reality -- and with an understanding of the nuances of correlation vs. causation -- correlating aggregate TV data with digital touchpoint data dilutes the accuracy of the attributed data and only provides insight into a potential correlation. It would be like overlaying my Uber driver’s trip log on top of his wife’s umbrella sales data: There may be some interesting patterns, but if you make business decisions based on that correlation, you may make serious business mistakes.
To understand causation, it is essential to consider not only multiple data points, but all touchpoints across every channel. Only this type of “full picture” view can isolate causation from correlation, providing brands with the ability to understand the true cause of a conversion and optimize efforts with this understanding.
Robust attribution technologies and advanced marketing measurement solutions account for this subtle yet important distinction. Anything less is major risk to the optimization recommendations brands put into market.