Friday, June 09, 2017

A Victory For YouGov's New Model As Traditional Polls Fail Again

The average Conservative lead across all polls in the last week of the campaign was 7.5%. The actual vote share according to the BBC, with 649 seats declared, was a 2.4% Conservative lead.

The most accurate poll was Survation's prediction of a 1% Tory lead. But overall the opinion polls significantly overstated the Conservative lead. I previously wrote about how polling companies were potentially overstating the Conservative lead by adjusting their raw results such as reducing youth turnout among other factors to produce much larger leads in their headline figures. Below are some examples.

You can see how turnout adjustments, particularly lower turnout for younger people, increased ICM's 2% Conservative lead to 11%.

It does appear that around 66% of 18-24 year olds voted this time, compared to 43% in the last election. The potential impact of young people was key, and I previously wrote about how higher turnout could add 1.7m Labour votes. It looks as if Corbyn managed to mobilise young voters in a way that no previous party has done.

In terms of seats the current numbers with one seat to declare are as follows:

The most accurate model for predicting seats was YouGov's new data model, which predicted a hung parliament, albeit projecting just 304 Conservative seats. YouGov are to take a lot of credit for this new model which it was brave to publish. However, they did appear to lose their nerve on the eve of polling and said their final prediction based on their traditional approach was a 7% Conservative lead and an increased Conservative majority.

Their new data model predicted the following seats.

This is a reasonably accurate seat projection. More significantly YouGov predicted Canterbury would go Labour and that Kensington would be very close. Kensington is the one seat outstanding as it is too close to call.

I think this new modelling approach will receive a lot of attention in the coming weeks as it models individual seat characteristics and demographics rather than uniform or regional swing models. It is clear that there has been no uniform swing or even a clear regional swing. The characteristics of individual seats has strongly affected the results thus southern Remain constituencies with large numbers of young people went heavily to Labour in excess of any national swings.

Maybe it is time for the BBC to retire their swingometer.