See our predictions for the next General Election.
Previously, we used uniform national swing (UNS) and applied it to every seat. For example, if Conservatives gained 3% and Labour lost 2% in the polls since the last general election, every seat across the country would, in effect, show Conservatives up by 3%, and Labour down by 2%. This gave a good ballpark figure for votes cast and seats, but it does not show where voters are going to or from.
Enter our new model: Response-weighted vote share
We take the data from the last three national polls, extracting the information about how voters voted in the last general election, and cross-referencing this with how they intend to vote in the next. Once we have this information, it is a simple case of identifying the swing from/to parties and adding or subtracting that from the share at the last election.
But not everyone is going to vote…
With that in mind, we also extract the data from the ‘likelihood of voting’ section in the responses, and add or subtract that for each 2017-2019 response cell, giving a final percentage of how the votes have changed and how likely these changing voter groups are to now vote.
You can see this in the table below. The rows are the 2019 votes from each 2017 party, and the columns are where these voters are moving to. So, LAB at the top, and LIB on the left denotes the total percentage of Labour voters who are very likely to vote Liberal Democrat in the next election.