Professor Stephen Fisher of Oxford University addressed the History of Statistics Section at Errol Street on 27 March 2019.
He began by describing the history of opinion polling, paying tribute to Nick Moon’s authoritative book Opinion Polls (1999); Nick was in the audience. The earliest known type of polling was a straw poll, used in Pennsylvania in 1824. A straw poll, an ad-hoc or unofficial vote, shows popular opinion on a certain matter and helps politicians decide what to say to gain votes.
Advances were made during the First World War for both political and commercial market surveys. In 1935, George Gallup founded what later became the Gallup Organization. It did better than other polls in the United States (particularly in 1936) until the 1948 election when all the major polls got it wrong. Polls predicted that the Republican Dewey would win but the incumbent Democrat, President Truman, won – one of the greatest upsets in American political history.
In the UK Gallup was alone in predicting a Labour win in 1945. In 1951 all three major polls predicted a Conservative lead; Labour got more votes but fewer seats. In the 1950s and 1960s polls usually got the winner right but overestimated the Labour vote.
After some poor results, there were inquiries – pollsters are not secretive about their methods. New methodology evolved, with face-to-face quota sampling, telephone interviewing, random-digit dialling. By 2010 most polls used the internet.
Both the BBC and ITV did exit polls in 1992, and successfully predicted the 2005 Labour majority. In 2015, there were changes to the way 'don’t knows' were allocated to estimated voting intentions. The polls suggested a big Conservative lead which was not borne out. Another inquiry recommended improving sampling procedures. The polls were accurate in 2017.
There is always a trend towards the status quo during an election or referendum campaign, and even the final polls overestimate the desire for change. Telephone polls show this more than internet polls and are less accurate. Professor Fisher quoted various examples of what polls say about opinion of political leaders. In the 1990s Black Wednesday was more important for Tony Blair than the move to the centre, and the banking crisis was good for Gordon Brown.
Governments start to recover from mid-term unpopularity about eight months before elections. In the 2016 EU referendum, polls were inaccurate, but the polling was considered better than other methods. The US presidential election in 2016 predicted a Clinton win, but state-level forecasts were more accurate. The currently preferred method is 'multilevel regression with post-stratification'. It was used to good effect by YouGov, but this forecast was officially dropped.
Betting markets predicted poorly in the EU referendum and in the US Presidential Election. There have been some improvements in polling over the course of nearly 200 years. Despite the unpredictable nature of the electorate, polls remain a useful guide to public opinion and continue to acknowledge how important it has become in our democracy.