Three reasons you are surprised that Trump won

Media outlets and your social network are reverberating with shock at Trump’s victory. How did we get to this point, and why is everybody so surprised? Here are a few insights from cognitive science, all of which we’ve known about since the 1970s:

Many faces across the world this morning resembled this one (albeit less orange)

1) People don’t understand probabilities

As America headed to the polls, high-quality forecasters gave Trump a 35% chance of victory. Somehow, people misinterpreted this as ‘Trump is unlikely to win the election’. Something that has a 35% chance of happening is not unlikely.

Kahneman and Tversky showed in 1972 that people are absolutely dreadful at understanding probabilities presented in this manner. It’s much easier to understand probabilities in terms of absolute numbers. Let’s say we have 100 elections. We would expect a candidate who polling suggests has a 35% chance of winning to win in 35 of 100 of them. In 35 out of 100 parallel realities, you woke up this morning to a Trump victory. In 65, it was Clinton. If I were to offer you an operation in which the patient died 35 times out of 100, would you think ‘well, it’s more LIKELY that I survive’, or would you think ‘Christ, I don’t fancy that’?

Something that has a 35% chance of happening is not unlikely.

This kind of misapprehension probably undermined the urgency with which people campaigned on behalf of, and turned out to vote for, Hillary. It may even have predisposed people to cast protest votes for Trump. People thought Trump was very unlikely to win. That’s not what the numbers said.*

2) Everything’s going to be alright.

Even those who really understood what the probability estimates were saying might not have really taken them on board. Humans tend to display an optimism bias — we overestimate the probability good things will happen, and underestimate the probability bad things will. Not only are people biased when guessing the probabilities of good and bad things, but they change their beliefs differently in the two cases. If I tell you something good is more likely than you thought, that sticks. If I tell you something bad is more likely, you’re much less likely to change your mind.

So those of us who wanted to believe that a Trump victory was unlikely believed just that.

3) Somebody else will do it.

In 1964, a girl named Kitty Genovese was stabbed on a residential street in NYC. 38 people witnessed the incident from their homes. As lights came on and curtains opened, the attacker fled the scene. Curiously, no doors opened. In fact, nobody even called the police. The attacker, emboldened by the apparent indifference of the neighbourhood, returned 30 minutes later to finish the job.

Kitty’s case provides the most famous example of the ‘bystander effect’. People find it very easy to do nothing in situations where something bad is happening. This seems to be a mixture of diffusion of responsibility — ‘if it’s important, somebody else will probably do it’, social modelling — ‘it can’t be that important, nobody else seems bothered’, and audience inhibition — ‘it’d be embarrassing if I got all upset about this and it turned out to be nothing’. All of these might have played a role in the lacklustre turnout for Hillary, and the lukewarm response that her campaign inspired in the people who she needed to enthuse. If you rolled your eyes at somebody’s Facebook post about supporting Hillary or campaigning on her behalf, you’ve probably fallen victim to the bystander effect.

Thanks for reading. If you have any thoughts on or extensions to the above, pop them in the comments below.

*Additional confusion was probably caused by the distinction between estimated vote share and probabilities, both of which were frequently presented in a similar manner (percentages between 40 and 60). In terms of vote share, a rock-solid 20% difference is extremely persuasive. If high quality polling had established that Hillary was 60% and Trump 40%, we might have had cause to rest on our laurels. But they didn’t — they placed them at 49 and 45%. Which leads to the probability estimates of 65/35, which is far from reassuring.

Product manager & data scientist. Writing about AI, building things, and climate change.

Product manager & data scientist. Writing about AI, building things, and climate change.