Collective failure? Lessons from combining forecasts for the UK's referendum on EU membership

Research on election forecasting suggests there are benefits from combining different sources of information. This paper discusses the experience of a combined forecasting method that was developed for the UK’s EU referendum in 2016. The sources included opinion polls, vote expectation surveys, prediction and betting markets, expert and volunteer forecasts, and various forecasting models based on polling and other kinds of data. Averages of sources within each of these categories all, in our final forecast, suggested Remain was more likely to win but with varying degrees of certainty. Combining them produced a forecast that beat some but not others. Opinion polls and citizen forecasts came closest to the true outcome. Betting and prediction market participants and volunteer forecasters were the most overconfident that the UK would vote Remain. This may have been because they were distrustful of the polls following the 2015 general election miss and had too strong an expectation of a late swing towards the status quo similar to those in Scotland in 2014 and Quebec in 1995.