Opinion polls have been taken in the UK since 1943. Mark Pack has compiled almost every opinion poll taken in his publicly available database, and we have used this database to graph the results of these opinion polls.
Given that polls have a margin of error and that so many polls are taken, we took polling averages to make this graph clearer. We produced three polling averages, using different techniques to calculate them. In order to see which was most predictive of election results, we compared the averages from each election since 1950 with the actual result. All three produced similar results, but the most predictive was the linear average, which is illustrated below.
Whilst polling for the 2015 general election has been criticised as being unusually unpredictive, our linear average only gets the results wrong by about 1.6% per party. Whilst this is worse than 2005 and 2010, it is better than almost every other general election.
The issue with 2015’s polling is that its error was almost entirely between the Labour and Conservative Parties. It overstated Labour’s support by 3.6% and understated the Conservatives’ support by 3.0%. This was a similar effect to the polling error in the 1992 general election, though in 1992 the effect was much greater.
In general, general election results have given Labour a lower voteshare than our polling average would predict, and the Conservatives a higher voteshare. If we start from 1959, when this effect first became as pronounced as it has been in recent years, our linear average shows an average understatement of Conservative support of 0.9%, overstatement of Labour support of 3% and overstatement of Liberal/Liberal Democrat support of 0.5%.