“Data-Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Today’s column is written by Carl Spaulding, executive vice president of strategy at NCSolutions.
Advertisers need a refresher course on the difference between attribution and causation – and why proving the latter is so much harder but also more valuable.
“Correlation does not imply causation,” the classic saying goes, meaning just because two variables might appear to have a cause-and-effect relationship, that’s not necessarily the case. There can be other factors affecting the variables, and good science demands exploring what those factors may be.
Any market research professional worth his or her salt would understand this distinction and take it into account when conducting a study. But lately, I’ve been dismayed to find so many people in our field making a similar mistake when discussing “attribution” and incorrectly stating what attribution does and doesn’t mean with respect to the effectiveness of an ad campaign.
Contrary to what most people think and say, attribution does not imply causation, and confusing the two prevents us from developing a more nuanced and helpful understanding of our industry.
The word “attribution” entered the corporate lexicon about 10 years ago, when last-click attribution vendors began selling products that track which links customers clicked on before buying certain items. But last-click didn’t take into account the customer’s entire path to purchase, so then came multitouch or multichannel attribution, which took into account all the websites a customer visited and the advertising they were exposed to on their conversion journey.
Neither of those models has anything to do with measuring the effectiveness of advertising. They might be able to determine which websites a consumer visited before buying a product, but they can’t say if that consumer bought that product as a direct or indirect result of advertising.
In other words, attribution measures correlation: Consumer X visited these websites and then bought product Y.
I, however, am more interested in causation, which is much harder to tease out, but, in the end, much more valuable. It says, Consumer X bought product Y because he saw advertisement Z. Causation provides so much more intelligence about not only what happened, but also how it happened and what can be done differently to improve results in the future.
This is nothing against attribution providers.
But many people mistakenly use attribution as a synonym for causality. Multitouch attribution doesn’t measure the complex purchase consumption dynamics that factor into a person’s shopping decisions.
Causation does take those things into account. The way to determine the causal factors that matter is through good science. And by that I mean identifying control and test groups, noting how their behaviors differ and controlling for mitigating factors such as geography, demographics and propensity to buy. Only then can the divergence in purchase behaviors between the two groups be recorded.
This kind of research has been done for 10 years now. There’s no sexy, easily digestible buzzword for this line of work.
I prefer to just call it what it is: good science.