Right term data; or what Trump can teach you about using data correctly
In the months, weeks and days leading up to last November’s US elections, one thing was clear: Hilary Clinton would win. According to The New York Times on the day of the election, the probability of a Clinton victory was 85%. Nate Silver’s 538 website, renowned for its statistical prowess, calculated the probability of Clinton winning the popular vote but losing the electoral college at just 10.5%.
The consensus was plain: all the analysis pointed to a resounding Clinton victory.
Well, most of the analysis. There were a few lone voices, calling in the wilderness, pointing in the other direction. Pointing towards an emotionally unstable reality TV host with an orange face and large hair. They too were basing their predictions on data, but coming to the opposite conclusion. For example, Allan J. Lichtman, a political historian at American University in Washington said just two days before the election that Trump would win.
Now as we know, the majority was wrong. The expert consensus proved to be founded on faulty assumptions and faulty analysis. Even with what looked like the most accurate, most up to date and most comprehensive data available to them, the actual outcome came as a nasty surprise.
So what went on? And what can we learn from it?
The answers can be found in the type of data that was being analyzed. What we saw in the elections showed what happens when we look at the wrong type of data; specifically, the wrong-term data.
There were in fact at least two distinct terms of data being used to predict voter behaviour: short term (e.g. polls) and long term (e.g. historical election trends). And in the marketing world, the same categories of short and long term data also exist. Short-term data is typically good at seeing what’s going on right now: what’s trending, what’s being purchased, how people currently feel. In the world of marketing and media, a lot of the digitally-sourced data is short-term, providing a seemingly-clear window into what’s being seen, heard or clicked on. Long-term data on the other hand typically looks wider and further: how people have behaved over time, how their feelings towards brands evolve, who they’ve actually voted for.
But all data have strengths and weaknesses. Long-term data is not great at picking up short-term activity. Short-term data is not good at showing longer-term effects. In the elections, the ones like Lichtman that got the result right looked at long-term data; all the others, those who got it wrong, focused solely on the short-term. In other words, the elections were a grand experiment in the effects of leveraging your future on wrong-term data. Too many people were looking at short-term data (polls) to predict long-term behavior (voting preferences). And the result was a nasty surprise.
All too often, we fall into the same wrong-term data trap when looking at our brands and businesses. In seeking to embrace all the exciting-looking short-term data available to us, we fail to ask the critical question of whether it’s the right kind of data. And to answer that, there is one criterion that is more important than anything else: is this data fit for purpose? Is it best suited for doing the job I need it to do? If I need to assess long-term behavioural trends (e.g. voting preferences, or brand sentiment), then I should be looking at long-term data, however tantalizing the short-term data may seem. Anything else is wrong-term data.
When it comes to marketing, all brands share the same common purpose: sustained increase of sales. This is inherently a multi-term purpose: short term (sales) and also long term (sustained increase). The analysis and data that feeds it should therefore also be multi-term. Any assessment of ROI needs to embrace a multi-term view: the return cannot only be seen in the short-term.
What smart marketers are doing is ensuring that the data they are looking at is aligned to the purpose they have in mind. Short-term data serves short-term purposes well. But for all those who are concerned with anything past the short-term, long-term data is essential. Because applying short-term data to long-term purposes is using wrong-term data, and that can lead to nasty surprises.
And I think we can all agree we’ve had quite enough of those recently.