ArpitGarg's Weblog

Polling (just like risk calculation in businesses) is as much a mathematical calculation as it is an art. You need to have the sense that few have to get it right. In last 10 years or so, it has become more of mathematical model than anytime before. Just like the 2007-09 financial crisis which brought to fore risk calculation model errors, similarly the current election result has brought to notice the polling model errors.

Trump campaign used to attack polls as fake and political hit jobs regularly. For my analysis I want to assume the intentions behind polls were pure. So why did they get it so wrong, barring a couple of polls (which can be owed to chance). I believe the polls had severe model errors, which didn’t took into account a lot of factors.

Just like 2007-09 when polls failed to take into account correlation risk across geographies, the polls failed to take into account the anger among the electorate. Also Trump was a new phenomenon, who was not someone one would like to say one likes, but was bang on message and policies. That created shy voters who did not admit to pollsters they were voting for Trump.

Polls also failed to take into account the contempt the main stream media had for Trump. So much so that anyone saying he/she supported Trump was bashed not just on social media but by mainstream media pundits too for being a bigot. I remember one of the Kardashians had to roll back on her comment of voting for Trump after the backlash. Would she have voted for Clinton? I very much doubt so. When bashing of a candidate becomes an acceptable thing and anyone supporting him is touted as a bad person by responsible people in media, it becomes difficult for one to admit the support anymore.

The anger, the shyness gave rise to the errors in legacy polling methods.

What polls also failed to take into account was the contempt for the media. Gallup poll which has tracked trust in media for last 40 years or so, found the trust to be the lowest in 2016. Two-thirds don’t trust media. And that same distrusting media was out to get Trump in full force. Media distrust fueled Trump support.

All these factors have to be modeled into polling methodology in future. The change election with a flawed candidate with an appealing message, the anger in the populace, the creation of shy voters and the bias in the media were few of the factors that broke the polls this year.

Hope the pollsters learn from these errors and evolve for better.