2016 was a whopper of a year both for political upsets and debates about traditional polling and its relevance. In this talk I will discuss Pythonic survey analysis and will also highlight the pitfalls of polling and sampling generally. I will close with some thoughts on polling surprises from Brexit and the US Presidential election.
Between Brexit and the US Presidential election, it has been quite a year in the English-speaking world for pollsters. In this talk I will first introduce basic concepts of political polling design methodologies and traditional analytical techniques for dealing with the necessarily skewed data that results from traditional sampling. I will then give an overview of existing Python packages for tackling survey design and demonstrate sample code applying existing packages and also roll-your-own approaches. I will discuss current industry best practice for polling and explain how these traditional methods were deployed to monitor the year's two biggest political votes: the US Presidential election and the UK Brexit referendum.
I will explore how, and whether, the outcomes of these two votes was as much as a surprise to pollsters as the media indicated and what might have led to more accurate predictions. Finally, I will close with comments about how polling methodology is likely to change in the coming years and what, if anything, could have been done differently analytically to better predict the actual results of these important 2016 votes.