One thing we’ve learned from the US election: think twice before ever trusting a poll
Hongkongers should take the findings of even the most seasoned local pollsters with a grain of salt. Just because they come with a PhD and work at a university doesn’t make them more credible
There is science, and then there is pseudo-science. Statistics is an exact science, but polling is not. There is physics, and there is economics. The global financial crisis that started in the United States almost a decade ago shook the very foundations of macroeconomics and its mathematical models. Mainstream, mathematically trained economists have been blamed for failing to predict the financial tsunami.
Now the shock victory of Donald Trump is doing something similar to American pollsters and their survey methods. Virtually no major poll in the lead-up to election day predicted that the Republican president-elect would win. The international science journal Nature has called it “the polling crisis”.
Americans practically invented public opinion polling, just as they developed the mathematical foundations of macroeconomics after the second world war. If so many bright and experienced pollsters with their financial and intellectual resources got it so spectacularly wrong, what hope is there for their counterparts in the rest of the world? This is especially worrying in societies facing deep divisions and conflicts. An immediate lesson for Hong Kong people: take the findings of even the most seasoned local pollsters with a grain of salt. Just because they come with a PhD and work at a university doesn’t make them more credible. Commonly cited polls in the US such as those by ABC/Washington Post, Ipsos, YouGov and Fox News showed the gap narrowing but still estimated a 3 to 4 percentage point lead for Hillary Clinton. Poll aggregators FiveThirtyEight and the New York Times forecast respectively a 65 and 85 per cent chance for a Clinton victory. The online Huffington Post projected a landslide win for Clinton.
Common polling methods include randomised telephone surveys and online questionnaires. The latter is problematic as people “self-select” to take part. Surveys usually need at least 1,000 participants to make them statistically significant. The trick is to make them representative of the population, factoring in income, education, gender, race and geographic location.
Interestingly, mobile phones are being partly blamed. Fewer people use land lines in the US as more of them own smartphones, which can screen out calls from strangers. But whatever the reasons for the failure, let it be a lesson for Hong Kong people to be sceptical about surveys by political parties or academics with an undisclosed agenda.