A pinboard by
George Ng

Ph.D in Biotechnology and an expert in machine learning


Would a larger sample size population help machine learning models match election results in polls?

Angry Millenials Laura Yaffe was the picture of the typical Brit, prim and proper. When small talk at a dinner turned to Brexit. Her resentment at the chief architects of the populist Leave camp movement was open and vile. UK Millenials were horrified when the Brexit results failed to match their expectations. According to Laura, ‘Anyone and everyone who can leave has left the UK for jobs in other countries‘.

While ubiquitous smartphones have enabled greater participation in online opinion polls. Bigger poll samples do not equal a true representation of the sample population. Machine learning models will fail at predicting elections when voters hide their preferences. On June 8th, UK will go to the polls. Recent opinion polls forecast Tories with a 43% majority. Will May lead the country to an election win? Or will millennials be looking for payback at the snap election?

Shy Tory Factor Polls that fail to match election results are not new. The Bradley Effect, explain how white voters deceived pollsters to avoid appearing racist. And refers to an erroneous forecast of Tom Bradley in the 1982 California Governor’s Election. “The Shy Tory factor”, tells of voters refusing to reveal their loyalties in the 1992 UK election. It skewed polls that predicted a Labour landslide win. In the recent 2016 USA Election. This effect reared its ugly head when polls at all levels forecast a Clinton landslide win. Trump won marginally.

Big Data And Polls It is obvious by now that polling has limitations and isn't gospel truth. Deviations arising from inherent randomness results in sampling bias. While biased sampling techniques leads to statistical errors. Both sampling bias and statistical errors plague polls. Compounding the error, voters faced with unpopular candidates lie about their preferences. With no intention to vote. With all these problems polls can turn to social media as a precursor of election results.