capture

Trump wins, pollsters lose. Biases rampant.

What is really surprising about these elections is not that Trump won, but that so many people were surprised about his win. The pollster gave Clint a 75% to 98% chance of winning just yesterday, yet Trump won. And it wasn’t a marginal win. It was a landslide. He hit the jackpot. Presidency, House, Senate, maybe even the popular vote. And it wasn’t just a lonely swing state. It was pretty much across the board.

Here a few examples – Trump last polls and election results:

Michigan: 44.2 –> 47.7 / Arizona: 47.6 –> 49.7 / Ohio: 47.7 –> 52.1 / Georgia: 49.5 –> 51.4 / Iowa: 48.0 –> 51.7 / Pennsylvania: 45.2 –> 48.8 / Utah: 37.3 –> 44.7 / Wisconsin: 44.3 –> 47.9 / Maine: 42.0 –> 48.0 / Montana: 52.1 –> 58.4 / Tennessee: 53.1 –> 61.1

Admittedly Clinton did win by a larger margin in the states she was expected to win, but considerably less than Trump.
Now, when polls are so wrong so early before the event, I find it hard to call it a surprise, and I’d rather call it a mistake. Pollsters have been embarrassingly wrong, and in a similar way to Brexit. Pollster didn’t recognise a number of factors, or did not properly account for a well-known factor at least, which is shy voters and the social desirability bias. I think few people can argue that Clinton has been presented by the media as a more socially desirable candidate. This shies away lean republican in the poll, which causes a bias toward Clinton. Then, when you are safe in your ballot station, you cast your real vote. The non-social desirable one. And Trump wins. Surprise? Not really.

And this is why this morning, entering the office, most of my colleagues admitted that my doubts about Clinton chances to win were spot on. I thought it was a coin toss. Not a magician, these conclusions were readily available to anyone with a bit of knowledge of statistics, and not emotionally charged.

The social desirability bias is more common than you imagine, and understandably so. You study it in Statistical Sampling 101. When you have a polarising event, biases do exist and need to be accounted for. But if you are part of the polarised set, you probably fail to recognise it and account for it. You fall prey of confirmation bias. And we can hardly argue that Clinton has been presented as the favourite candidate by the Media. Whether it was a pro-Clinton or an against Clinton running the polls, they were victim of the same social desirability bias. They saw the data, got confirmation of what they *knew* was right (Clinton looked more socially desirable – to both Democrats and Repubicans! -, so no surprise she is ahead), and failed to adjust for social desirability. They were victims of the same bias they were expected to remove! Plus one, of course, the confirmation bias. This is what happens when you have a wide and growing gap between society segments. And this is the concerning aspect of these elections: there appears to be a gap, between large parts of society, which is not recognised or, even more worryingly, is neglected. And then you have it: Brexit, Trump. What is next? Austria? Italy? France?

One thought on “Trump wins, pollsters lose. Biases rampant.

Leave a Reply