Like a lot of people in the insights and research industry, I tuned into the U.S. election last month with interest, not only because it would decide the fate of the free world, but also because it is probably the most important example (along with Covid-19 this year) of communicating data to a mass audience.
It’s easy to forget not everyone was following the ins and outs, polling, forecasting and electoral college as closely as I was. For the vast majority of the U.S. this was going to be the only interaction they had with those concepts for four years.
There are significant difficulties communicating data in any situation, let alone during the most high-profile event of the year. Here are two things that I learned from the U.S. Election about communicating data.
Communicating uncertainty
I called it a night only a couple of hours after the polls had closed, just after Miami-Dade had reported and gave a more favourable swing to Trump than expected. I woke up with the President having already claimed victory, and betting prediction markets now putting Biden at odds of 7/4 or worse to win. I should have taken that bet.
People saw Biden’s c.90% pre-election probability of taking the White House and took it to mean he would definitely win. When Trump started to perform better than expected, suddenly they shifted too far in the other direction, memories of 2016 no doubt coming to mind.
The fact is, the pre-election predictions ended up being fairly accurate. Trump outperformed expectations, but he still lost, badly in the popular vote, and significantly enough that they could only qualify for a recount in Georgia and Wisconsin.
Philip Tetlock’s Superforecasting is a great explanation of how we should actually read prediction percentages, what those percentages actually mean, and how to distinguish between a sure thing and a strong favourite. In the end, the strong favouring of Biden was probably about right, but pollsters and predictors needed to better communicate the details of their predictions, taking into account the need for certainty people feel.
For the wider research industry, it’s a lesson around communicating the true meaning of numbers and not forgetting the context they come from.
Looking for patterns in numbers
Many people breathed a sigh of relief when the U.S. networks began calling the result in Biden’s favour. Trump’s own former director of communications posted this tweet.
Anthony Scaramucci on Twitter: “For my friends in the UK, this is what it’s been like since Tuesday night.”
In the video, we can see the election being likened to Watford scoring in the play-off semi-finals a few years ago.
Arizona, passing to Wisconsin, Wisconsin crossing to Michigan, who heads the ball back for Nevada to score. The crowd goes wild.
And yet, that’s not how it happened at all. That’s how the votes were counted, not how they were cast, and if you look at how they were cast the narrative switches away from Biden coming from behind to stage a dramatic comeback.
He was never behind. He started off comfortably ahead, and it was only towards the end, when Trump’s supporters came out en masse on election day, that the gap narrowed.
Republicans understand the importance of narrative, which was why they refused to let mail-in ballots be counted before election day in key swing states. It is also the only reason why Trump’s claims of election fraud have even a thin veneer of plausibility to some people.
When communicating data, its important to remember that no single piece is seen in isolation, rather that they are connected and linked to everything that comes before and after. Even if they aren’t related, the human brain has a tendency to create a story from the illusion of a pattern. If the human brain is going to create patterns, why not construct those patterns yourself to make the process easier?
A new horizon
As well as showing us what the next four years would look like, the U.S. election was a great learning opportunity around some of the pitfalls of communicating data.
While these examples came in a political context, they contain important lessons for all of us. The human brain didn’t evolve to deal with numbers, probabilities or unrelated events. It evolved to look for patterns where there are none, and to look for the certainty in ambiguity.
When communicating data in any setting, we need to keep in mind these cognitive biases and make use of them when constructing a story.
