In Defence of Liberty

Driven by data; ridden with liberty.

General Election 2017: Curtice-Firth Methodology of Exit Polls

At 10pm on election day, the polls close and the broadcasters reveal what the exit poll predicted who win the election, and how many seats each party will take.

general-election-2017-exit-poll-bbc.jpg

An exit poll (unlike pre-election polls) do not ask how people intend to vote, but ask how people who have actually gone to a polling station actually voted. In the UK, respondents are asked to fill in a duplicate ballot paper and place it inside a ballot box.

This survey is then used to compute seats. The most recent exit poll was jointly commissioned by the BBC, ITV News and Sky News, conducted by GfK and Ipsos MORI.

Prior to 2001, broadcasters (like the BBC and ITV) commissioned separate exit polls, with the accuracy of these exit polls varying widely: sometimes being good, sometimes being poor.

general-election-2017-exit-poll-errors.PNG

In 2001, the BBC commissioned an exit poll by experienced polling companies, with a team of academic researchers to analyse the data. Between 2001 and 2010, this team has included John Curtice, David Firth, Steve Fisher, Jouni Kuha, Clive Payne and Neil Shephard.

Data, Models and Luck

The Curtice-Firth methodology of exit polling takes data from several sources to reach its conclusions. This article is to serve as a shorter version of the University of Warwick’s explanatory page.

The previous election result

The constituency results at the previous general election are the first source. This is unsurprising, given voters at successive general elections tend to be the same people, and the error involved in constituency results is essentially zero (barring a few miscounted votes). Knowing how each constituency voted is an essential resource of voter behaviour.

The previous exit poll

Exit polling is made extraordinarily difficult in the United Kingdom, given there is no mandate for counting districts to produce results by each polling station.

The Curtice-Firth methodology crucially measures change between the previous exit poll and the current one, meaning that any particular biases resulting from the choice of polling stations, differential refusal and postal voting should cancel each other out.

The current exit poll

The current exit poll uses two experienced market research companies, GfK (formerly NOP) and Ipsos MORI, to conduct the duplicate ballot. There is also careful planning given to the exit poll panel — what polling stations are visited — as there may be good reasons as to why a particular polling station cannot be used.

Taken together, the exit poll receives high-quality data from electors, which underpins the entire exercise.

Electoral Change

Whilst the concept of the Butler swing from one party to another is widely accepted, this is insufficient for measuring change within a multi-party contest.

The exit poll is then used to calculate, for three parties (Conservatives, Labour, and Liberal Democrats): the change in the Liberal Democrat share of the three-party vote, and the change in the Labour share of the two-party vote (Conservatives and Labour combined). This method can be extended, where necessary, to more parties.

Statistical Modelling

With a survey collecting around 200 votes each at approximately 100 polling stations, this data is then used to arrive at a forecast in over 600 constituencies.

Getting from the data to a forecast necessitates a statistical model. The researchers use various constituency-level data (such as the previous election result, or census data) to ‘explain’ — using regression analysis — patterns of electoral change in different types of constituencies. An example would be electoral changes measured in urban and rural constituencies.

Coming up with a model, particularly under the requirements to produce a good forecast before 10pm, takes a great deal of psephological expertise and statistical imagination.

Probabilities of each seat

The exit poll analysis is then used to estimate what the vote share in each constituency is likely to be, and so, what the ‘win’ probability of each seat is for each party.

For example, we could have a seat where Labour are ahead of the Conservatives, but the two parties are close. The probabilistic model places a Labour win in the seat at 65% (0.65), with the Conservatives at 35% (0.35).

In the final forecast, this seat is used to count as 0.65 towards the Labour total, and as 0.35 towards the Conservative seat total.

Adding up the win probabilities across the 650 constituencies gives the final exit poll seat count, seen on our screens at 10pm.

Advertisements

2 comments on “General Election 2017: Curtice-Firth Methodology of Exit Polls

  1. David Firth
    June 16, 2017

    Thanks for this! But it should always be recognised that when I developed the statistical methods for modern exit polling at UK general elections, it was very much a collaborative effort. Without the knowledge and experience of John Curtice, and especially his contacts among the broadcasters and market-research companies, the new methods probably would never have been used. So while I appreciate very much your heading “Firth Methodology of Exit Polls”, it really is Curtice-Firth methodology (as in our 2008 published paper, with authors’ names listed alphabetically!).

    • Anthony Masters
      June 16, 2017

      Thanks so much for commenting, and congratulations on another successful exit poll.

      I will correct that misnomer now.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Information

This entry was posted on June 14, 2017 by in National Politics and tagged , , , .
%d bloggers like this: